v0.50.210: gpt-5.5, cron titles, agent cache, bfcache fix, onboarding fix, mermaid CSP, PWA auth (#1056)

* feat(models): add gpt-5.5 to openai, openai-codex, copilot catalogs

Adds GPT-5.5 and GPT-5.5 Mini entries to the static _PROVIDER_MODELS
catalog so they appear in the model picker for the openai, openai-codex,
and copilot providers.

Signed-off-by: Pix (PiClaw, claude-opus-4-7) via Hermes Agent

* fix(models): add gpt-5.5-mini to copilot provider catalog

* fix(renderer): suppress Mermaid Google Fonts CSP violation via fontFamily inherit (#1044)

Mermaid's built-in 'dark' and 'default' themes inject an @import for
fonts.googleapis.com/Manrope into every generated SVG. The CSP style-src
only allows cdn.jsdelivr.net, so this request is blocked on every diagram
render, filling the console with CSP errors.

Fix: pass fontFamily:'inherit' (and fontSize:'14px') in the themeVariables
block of mermaid.initialize() in renderMermaidBlocks(). This suppresses
Mermaid's external font import and uses the page's existing font stack.

Avoids adding fonts.googleapis.com to the CSP — no new external dependency,
no font FOUT, consistent with the rest of the UI typography.

3 regression tests added in tests/test_1044_mermaid_csp_font.py.
2215/2215 tests passing.

* fix(onboarding): non-standard provider/path cluster (#1029)

* fix(bfcache): restore full layout on tab/session restore — rail, topbar, panels (#1045)

The pageshow handler added for #822 only cleared the session search filter
and re-rendered the session list. This left the rest of the layout chrome
(topbar, rail icons, workspace panel, resize handles, gateway SSE) in the
stale bfcache DOM state, causing a broken layout (oversized search icon,
uninitialized rail) that required a hard refresh to fix.

Fix: extend the pageshow handler to re-run the full set of layout sync calls
that the boot IIFE runs on a fresh page load:

  syncTopbar()              — restores model chip, title, topbar state
  syncWorkspacePanelState() — restores workspace panel open/closed
  _initResizePanels()       — reattaches panel resize drag listeners
  startGatewaySSE()         — reconnects the gateway SSE watcher
                              (bfcache-persisted connections are dead)

All four calls are typeof-guarded for safe degradation if a helper is not
yet defined. The existing #822 fixes (sessionSearch clear +
renderSessionListFromCache) are preserved unchanged.

loadSession() is intentionally NOT re-called — it would cause message
flicker; the sync calls above are sufficient to restore visual state.

7 regression tests added in tests/test_1045_bfcache_layout_restore.py.
2219/2219 tests passing.

* fix(bfcache): also close open dropdowns on bfcache restore (#1045)

Additional symptom noted in issue #1045: bfcache freezes the DOM including
any open dropdown/popover state. The thinking-level selector (and other
composer dropdowns) left open when navigating away would appear open without
user interaction on tab restore.

Extend the pageshow handler to call all four named close functions before
the layout sync:
  closeModelDropdown()     — composer model selector
  closeReasoningDropdown() — thinking/reasoning effort selector
  closeWsDropdown()        — workspace chip dropdown
  closeProfileDropdown()   — profile switcher dropdown

All calls are typeof-guarded, matching the style of the layout sync calls
already in the handler.

2 new tests (9 total in test_1045_bfcache_layout_restore.py):
- pageshow closes all four named dropdowns
- dropdown closes appear before layout sync calls (clean state first)

2221/2221 tests passing.

* fix(bfcache): remove _initResizePanels() — bfcache preserves listeners

* fix(bfcache): remove _initResizePanels from pageshow — bfcache preserves listeners; update test

* fix(sessions): use cron job name as session title when available (#1032)

* fix(test): add id column to messages table in cron title test fixture

* fix(merge): inject cron title lookup into read_importable loop, remove stale sqlite3 block

* fix(pwa): redirect to /login client-side on 401 — fixes iOS PWA auth expiry trap (#1038)

When an auth session expires, the server returns a 302→/login for page
requests. In a normal browser this works fine, but in an iOS PWA running
in standalone mode the redirect navigates out of the PWA shell into Safari,
leaving the app permanently stuck on 'Authentication required' with no
recovery path.

Fix: intercept 401 responses client-side before surfacing any error.

- workspace.js api(): check res.status===401 first; call
  window.location.href='/login' and return immediately (no throw)
- ui.js: add _redirectIfUnauth() helper; wire into all direct fetch()
  calls that bypass api() — api/models, api/models/live, api/upload

All fetch paths that could receive a 401 now redirect cleanly within
the PWA frame rather than opening Safari.

6 regression tests added in tests/test_1038_pwa_auth_redirect.py.
2175/2175 tests passing.

* fix(pwa): preserve current URL in ?next= param on 401 redirect

* fix(test): update 401-redirect assertion to accept ?next= URL format

* feat(pwa): add _safeNextPath() to login.js so ?next= param is honored after re-login

Addresses reviewer suggestion: the ?next= URL set on 401 redirect was ignored by
the login success handler (always redirected to ./). _safeNextPath() validates and
returns the ?next= param with open-redirect guards: rejects non-path-absolute inputs,
// protocol-relative URLs, backslash variants, and control characters.
4 new regression tests added.

* Implement session agent cache for AIAgent reuse

Added session agent cache to reuse AIAgent across messages.

* Implement agent caching for session management

* Implement session agent eviction on session deletion

Added session agent eviction to prevent turn count leakage in recycled sessions.

* docs: v0.50.210 release notes — 7 PRs, 2239 tests (+27)

* docs(changelog): drop stale [Unreleased] entries duplicated by v0.50.210

Three entries in the [Unreleased] section are duplicates of items now
listed under v0.50.210:

  - Mermaid CSP font fix (#1044)        → v0.50.210 / Mermaid Google Fonts CSP
  - bfcache layout restore (#1045)      → v0.50.210 / bfcache layout and dropdown restore
  - iOS PWA auth redirect (#1038)       → v0.50.210 / Login redirects back to original URL

The original drafts landed in [Unreleased] when individual PRs (#1047,
#1048, #1043) were approved; the v0.50.210 release-notes commit then
added the same items under the version section without removing the
[Unreleased] copies. Drop the duplicates so users reading the CHANGELOG
don't see the same fix listed twice.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Signed-off-by: Pix (PiClaw, claude-opus-4-7) via Hermes Agent
Co-authored-by: Pix (Hermes) <aliceisjustplaying@users.noreply.github.com>
Co-authored-by: nesquena-hermes <nesquena-hermes@users.noreply.github.com>
Co-authored-by: qxxaa <mrhanoi@outlook.com>
Co-authored-by: Nathan Esquenazi <nesquena@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.env.example

@@ -26,3 +26,6 @@
 
 # Path to your Hermes config.yaml (for toolsets and model config)
 # HERMES_CONFIG_PATH=~/.hermes/config.yaml
+
+# Display name for the assistant in the UI (default: Hermes)
+# HERMES_WEBUI_BOT_NAME=Hermes

.github/workflows/release.yml

@@ -52,5 +52,6 @@ jobs:
           push: true
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
+          build-args: HERMES_VERSION=${{ github.ref_name }}
           cache-from: type=gha
           cache-to: type=gha,mode=max

.github/workflows/tests.yml

@@ -0,0 +1,30 @@
+name: Tests
+
+on:
+  pull_request:
+    branches: [master]
+  push:
+    branches: [master]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ['3.11', '3.12', '3.13']
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install pyyaml>=6.0 pytest pytest-timeout
+
+      - name: Run tests
+        run: pytest tests/ -v --timeout=60

.gitignore

@@ -16,12 +16,26 @@ archive/
 .env
 .env.*
 !.env.example
-.claude/*
+.claude/
+CLAUDE.md
+AGENTS.md
+.cursorrules
+.windsurfrules
+.aider*
+copilot-instructions.md
 
 # Generated screenshots and transient artifacts
 screenshot-*.png
 full-UI.png
 
+# Version file written by Docker/CI build — generated, never committed
+api/_version.py
+
 # OS files
 .DS_Store
 Thumbs.db
+
+# Local reference clones — never committed (except tracked design/UI-UX reference pages)
+docs/*
+!docs/ui-ux/
+!docs/ui-ux/**

AGENTS.md

@@ -1,53 +0,0 @@
-# Web UI MVP Instructions
-
-Canonical source: <repo>/
-Symlink (for imports): <agent-dir>/webui-mvp -> <repo>
-Runtime state: ~/.hermes/webui-mvp/sessions/
-
-Purpose:
-- Claude-style web UI for Hermes. Chat, workspace file browser, cron/skills/memory viewers.
-
-Start server:
-  cd <agent-dir>
-  nohup venv/bin/python <repo>/server.py > /tmp/webui-mvp.log 2>&1 &
-  # OR: <repo>/start.sh
-
-Run tests:
-  cd <agent-dir>
-  venv/bin/python -m pytest <repo>/tests/ -v
-
-Health check: curl http://127.0.0.1:8787/health
-Logs: tail -f /tmp/webui-mvp.log
-SSH tunnel from Mac: ssh -N -L 8787:127.0.0.1:8787 <user>@<your-server>
-
-Living documents (always update after a sprint):
-  <repo>/ROADMAP.md
-  <repo>/ARCHITECTURE.md
-  <repo>/TESTING.md
-
-Sprint process skill: webui-sprint-loop
-
-# Workspace Convention (Web UI Sessions)
-
-When running as an agent invoked from the web UI, each user message is prefixed with:
-
-  [Workspace: /absolute/path/to/workspace]
-
-This tag is the single authoritative source of the active workspace. It reflects
-whichever workspace the user has selected in the UI at the moment they sent that message.
-It updates on every message, so if the user switches workspaces mid-session, the very
-next message will carry the new path. Always use the value from the most recent tag.
-
-This tag overrides any prior workspace mentioned in the system prompt, memory, or
-conversation history. Never infer or fall back to a hardcoded path like
-~/workspace when this tag is present.
-
-Apply it as the default working directory for ALL file operations:
-
-  - write_file: resolve relative paths against this workspace
-  - read_file / search_files: resolve paths relative to this workspace
-  - terminal workdir: set to this path unless the user explicitly says otherwise
-  - patch: resolve file paths relative to this workspace
-
-If no [Workspace: ...] tag is present (e.g., CLI sessions), fall back to
-~/workspace as the default.

ARCHITECTURE.md

@@ -7,20 +7,42 @@
 >
 > Keep this document updated as architecture changes are made.
 
+> Current shipped build: `v0.50.36-local.1` (April 16, 2026).
+> Baseline: upstream `nesquena/hermes-webui` `v0.50.36`.
+> Intentional local delta: first-time password enablement from Settings immediately issues a `hermes_session` cookie so the current browser remains signed in. The previous `Assistant Reply Language` customization has been removed, legacy `assistant_language` settings are filtered out on load/save, the workspace panel closed/open state is preloaded via a `documentElement` dataset marker before `style.css` paints to avoid a first-load desktop flash, transcript disclosure cards now animate caret rotation and body expansion with transitionable `max-height`/`opacity` states instead of `display:none/block`, and thinking cards now share the same rounded bordered card chrome as tool cards while keeping their gold palette.
+> Automated coverage: 1353 tests collected (`pytest tests/ --collect-only -q`).
+
 ---
 
 ## 1. Overview and Purpose
 
 The Hermes Web UI is a lightweight web application that gives you a browser-based
 interface to the Hermes agent that is functionally equivalent to the CLI. It is modeled on
-the Claude-style interface: a three-panel layout with a sidebar for session management,
-a central chat area, and a right panel for workspace file browsing.
+the Claude-style interface: a sidebar for session management, a central chat area,
+and a demand-driven right panel used for workspace browsing and preview surfaces.
+The right panel is closed by default on desktop and opens only when it is actively
+being used for browsing or previewing content.
+
+To prevent a visible first-paint mismatch on refresh, `static/index.html` preloads the
+saved workspace panel state into `document.documentElement.dataset.workspacePanel`
+before the main stylesheet loads. Desktop CSS honors that preload marker immediately,
+and `static/boot.js` keeps the dataset synchronized with the runtime panel state machine.
 
 The design philosophy is deliberately minimal. There is no build step, no bundler, no
 frontend framework. The Python server is split into a routing shell (server.py) and
 business logic modules (api/). The frontend is seven vanilla JS modules loaded from static/.
 This makes the code easy to modify from a terminal or by an agent.
 
+For the current local build, the codebase is intentionally as close to upstream as possible:
+the app now tracks upstream `v0.50.36`, keeps the password-session continuity patch in the
+settings/onboarding flow, and does not carry forward the prior reply-language preference
+feature.
+
+Hermes-level chrome is intentionally consolidated: the sidebar has no dedicated brand header.
+Instead, the footer exposes a single "Hermes WebUI" launch button that opens one tabbed
+control-center modal for global preferences, conversation import/export, and clear-conversation
+actions. The topbar remains focused on conversation context and the workspace/files toggle.
+
 ---
 
 ## 2. File Inventory
@@ -28,7 +50,8 @@ This makes the code easy to modify from a terminal or by an agent.
     <repo>/
     server.py              Thin routing shell + HTTP Handler + auth middleware. ~81 lines.
                            Delegates all route handling to api/routes.py.
-    start.sh               Discovery script: finds agent dir, Python, starts server.
+    bootstrap.py           One-shot launcher: optional agent install, deps, health wait, browser open.
+    start.sh               Thin wrapper around bootstrap.py for shell-based startup.
     Dockerfile             python:3.12-slim container image (~23 lines)
     docker-compose.yml     Compose config with named volume and optional auth (~22 lines)
     .dockerignore          Excludes .git, tests/, .env* from Docker builds
@@ -39,7 +62,9 @@ This makes the code easy to modify from a terminal or by an agent.
       helpers.py           HTTP helpers: j(), bad(), require(), safe_resolve(), security headers (~71 lines)
       models.py            Session model + CRUD, per-session profile tracking (~137 lines)
       profiles.py          Profile state management, hermes_cli wrapper (~246 lines)
+      onboarding.py        First-run onboarding status, real provider config writes, and readiness detection.
       routes.py            All GET + POST route handlers (~1180 lines)
+      startup.py           Startup helpers: auto_install_agent_deps() (~50 lines)
       streaming.py         SSE engine, run_agent, cancel, HERMES_HOME save/restore (~236 lines)
       upload.py            Multipart parser, file upload handler (~78 lines)
       workspace.py         File ops: list_dir, read_file_content, workspace helpers (~77 lines)
@@ -48,11 +73,12 @@ This makes the code easy to modify from a terminal or by an agent.
       style.css            All CSS incl. mobile responsive (~670 lines)
       ui.js                DOM helpers, renderMd, tool cards, model dropdown, file tree (~977 lines)
       workspace.js         File preview, file ops, loadDir, clearPreview (~185 lines)
-      sessions.js          Session CRUD, list rendering, search, SVG icons, overlay actions (~533 lines)
+      sessions.js          Session CRUD, list rendering, search, SVG icons, dropdown actions (~533 lines)
       messages.js          send(), SSE event handlers, approval, transcript (~297 lines)
       panels.js            Cron, skills, memory, workspace, profiles, todo, settings (~974 lines)
       commands.js          Slash command registry, parser, autocomplete dropdown (~156 lines)
-      boot.js              Event wiring, mobile nav, voice input, boot IIFE (~338 lines)
+      onboarding.js        First-run wizard overlay, provider setup flow, and settings/workspace orchestration.
+      boot.js              Event wiring, mobile sidebar/workspace nav, voice input, boot IIFE (~338 lines)
     tests/
       conftest.py          Isolated test server (port 8788, separate HERMES_HOME) (~240 lines)
       test_sprint{1-20b}.py Feature tests per sprint (21 files, 415 test functions)
@@ -347,7 +373,7 @@ highlighting) and Mermaid.js (diagrams) from CDN, both loaded async/deferred wit
 Six JS modules loaded in order at end of <body>:
   1. ui.js       (~846 lines) DOM helpers, renderMd, tool card rendering, global state
   2. workspace.js (~169 lines) File tree, preview, file operations
-  3. sessions.js  (~532 lines) Session CRUD, list rendering, search, SVG icons, overlay actions, project picker
+  3. sessions.js  (~532 lines) Session CRUD, list rendering, search, SVG icons, dropdown actions, project picker
   4. messages.js  (~293 lines) send(), SSE event handlers, approval, transcript
   5. panels.js    (~771 lines) Cron, skills, memory, workspace, todo, switchPanel
   6. boot.js      (~175 lines) Event wiring + boot IIFE
@@ -358,10 +384,19 @@ inherit `currentColor` for consistent theming.
 
 Three-panel layout (in static/index.html):
 
-    <aside class="sidebar">    Left panel: session list, nav tabs, model selector
+    <aside class="sidebar">    Left panel: session list, nav tabs, sidebar-footer Hermes WebUI trigger
     <main class="main">        Center: topbar, messages area, approval card, composer
     <aside class="rightpanel"> Right panel: workspace file tree and file preview
 
+Composer footer layout (current):
+
+    left cluster   attach button, mic button, per-conversation model selector
+    right cluster  compact circular context-usage badge, send button
+
+The model selector is still the authoritative control for new-session creation
+and session updates; it was moved out of the sidebar so model choice feels scoped
+to the active conversation rather than a global app setting.
+
 ### 5.2 Global State
 
     const S = {
@@ -406,11 +441,19 @@ Approval:
     stopApprovalPolling   clearInterval
 
 UI helpers:
-    setStatus(t)          Updates #statusText in composer footer
+    setStatus(t)          Fallback helper: shows a toast for non-chat status/error messages
+    setComposerStatus(t)  Updates the inline composer status label for turn-scoped states
     setBusy(v)            Sets S.busy, disables/enables Send button, clears status on false
     showToast(msg, ms)    Bottom-center fade toast (default 2800ms)
+    showConfirmDialog(o)  Shared in-app confirmation modal, resolves true/false
+    showPromptDialog(o)   Shared in-app input modal, resolves string/null
     autoResize()          Auto-resize #msg textarea up to 200px
 
+Dialog policy:
+    Native browser confirm()/prompt() are not used in the Web UI.
+    Destructive actions use showConfirmDialog(...), then a toast on success.
+    Lightweight naming flows (new file/folder/project) use showPromptDialog(...).
+
 Files:
     loadDir(path)         GET /api/list, rebuild #fileTree
     openFile(path)        GET /api/file, show in #previewArea
@@ -463,7 +506,7 @@ Known gaps:
 - Nested lists: single regex pass, multi-level indentation not handled
 - Mixed bold+link in same line: may produce garbled output
 
-### 5.5 Model Chip Label (Fixed in Sprint 1)
+### 5.5 Model Label Resolution (Fixed in Sprint 1, reused by composer selector)
 
 B3 was resolved in Sprint 1. Current code uses a MODEL_LABELS dict:
 
@@ -474,10 +517,10 @@ B3 was resolved in Sprint 1. Current code uses a MODEL_LABELS dict:
       'anthropic/claude-haiku-3-5': 'Haiku 3.5', 'google/gemini-2.5-pro': 'Gemini 2.5 Pro',
       'deepseek/deepseek-chat-v3-0324': 'DeepSeek V3', 'meta-llama/llama-4-scout': 'Llama 4 Scout',
     };
-    $('modelChip').textContent = MODEL_LABELS[m] || (m.split('/').pop() || 'Unknown');
+    getModelLabel(m) => MODEL_LABELS[m] || (m.split('/').pop() || 'Unknown');
 
 Fallback: any unlisted model shows its short ID (after the last /) rather than a wrong label.
-To add a new model: add an entry to MODEL_LABELS and add an <option> to the <select>.
+To add a new model: add an entry to MODEL_LABELS and add an <option> to the composer footer <select>.
 
 ### 5.6 Session Delete Rules (from skill)
 
@@ -1095,7 +1138,7 @@ The model chip label bug is now fixed. The MODEL_LABELS object in syncTopbar():
       'deepseek/deepseek-chat-v3-0324':  'DeepSeek V3',
       'meta-llama/llama-4-scout':        'Llama 4 Scout',
     };
-    $('modelChip').textContent = MODEL_LABELS[m] || (m.split('/').pop() || 'Unknown');
+    getModelLabel(m) => MODEL_LABELS[m] || (m.split('/').pop() || 'Unknown');
 
 Fallback: splits on '/' and uses the last segment, so any unlisted model shows its
 short identifier rather than a wrong hardcoded label.
@@ -1586,3 +1629,19 @@ and #rightpanelResize. On mousemove: computes delta and clamps to min/max. On mo
 saves width to localStorage. Widths restored at boot via localStorage.getItem().
 CSS: .resize-handle with position:absolute, width:5px, cursor:col-resize.
 body.resizing added during drag to suppress text selection.
+
+
+## Workspace path trust levels
+
+`api/workspace.py` has two distinct trust functions — do not collapse them:
+
+**`validate_workspace_to_add(path)`** — used by `/api/workspaces/add` (explicit user registration).
+Permissive: blocks only non-existent, non-directory, and system root paths. The user is
+consciously registering an external path (e.g. `/mnt/d/Projects` in WSL), so we trust intent.
+
+**`resolve_trusted_workspace(path)`** — used for actual file read/write operations inside
+an existing workspace. Strict: path must be under home, in the saved workspace list, or under
+`BOOT_DEFAULT_WORKSPACE`. Prevents path traversal and unauthorized file access.
+
+The distinction matters because add uses permissive validation to avoid the circular
+dependency: you cannot get a path into the saved list if you need the saved list to add it.

BUGS.md

@@ -10,6 +10,18 @@ This file tracks UI bugs and polish items. Fixed items are kept for reference.
 
 ---
 
+## Known Limitations
+
+- **Two-container Docker setup: tools run in WebUI container** — In the two-container setup (hermes-agent + hermes-webui as separate containers), WebUI-initiated agent sessions run tools in the WebUI container, not the agent container. This is a known architectural constraint. Workaround: use the combined single-image approach, or initiate sessions via the CLI in the agent container. (#681)
+
+- **Image-in-chat vs. saved-to-workspace mismatch** — When the agent displays an inline image (from a URL) and the user asks it to save that image, the agent issues a fresh download which may return a different file if the source URL is CDN-rotated or parameterized. The WebUI correctly renders whatever URL the agent provides. Fix requires agent-side URL caching. (#641)
+
+- **MCP tools not available in WebUI sessions** — MCP servers must be configured in the active profile's config.yaml under mcp_servers:. If MCP tools are not appearing, check that the profile is correct and the MCP server process is reachable from inside the WebUI container. (#628)
+
+- **os.environ race condition in concurrent sessions** — Concurrent agent sessions share process-level os.environ for TERMINAL_CWD, HERMES_SESSION_KEY, and HERMES_HOME. _ENV_LOCK serializes mutations but does not fully isolate env vars during agent execution. Upstream fix pending in hermes-agent. (#195)
+
+---
+
 ## Fixed
 
 ### ~~Session title truncation / hover actions~~ -- Fixed (Sprint 16)

CONTRIBUTING.md

@@ -0,0 +1,171 @@
+# Contributing to Hermes WebUI
+
+Thanks for contributing.
+
+Hermes WebUI is intentionally simple to work on: Python on the server, vanilla JS in the browser, no build step, no bundler, no frontend framework. The best pull requests preserve that simplicity while solving a real problem cleanly.
+
+## Two Paths to a Strong Pull Request
+
+### Path 1: Small, Focused Changes
+
+This is the fastest path to review and merge.
+
+- Fix one clear bug or add one tightly scoped improvement
+- Touch the fewest files you can
+- Avoid drive-by refactors mixed into functional changes
+- Run the relevant tests locally before opening the PR
+- Keep the PR description concise and specific
+
+These are the changes that are easiest to review and safest to merge quickly.
+
+### Path 2: Bigger Changes
+
+If you want to change architecture, reshape a workflow, add a substantial UI feature, or alter core behavior, align on direction first.
+
+- Open an issue, start a discussion, or open a draft PR early
+- Explain the problem you are solving, not just the implementation you want
+- Call out tradeoffs, migration risk, and any alternatives you considered
+- Keep the final PR easy to review by separating unrelated work
+
+Large changes are welcome, but surprise rewrites are hard to review well.
+
+## What We Expect in Every PR
+
+### 1. One Logical Change Per PR
+
+Keep each PR focused. A small related group of fixes is fine. A bug fix plus a CSS cleanup plus a refactor plus a docs rewrite is not.
+
+### 2. Local Verification
+
+Run the test suite locally:
+
+```bash
+pytest tests/ -v --timeout=60
+```
+
+CI also runs this suite on Python `3.11`, `3.12`, and `3.13`.
+
+If your change affects browser behavior, also run the relevant manual checks from [TESTING.md](TESTING.md).
+
+### 3. Clear PR Description
+
+There is currently no PR template in this repo, so include the important sections yourself:
+
+- Thinking Path
+- What Changed
+- Why It Matters
+- Verification
+- Risks / Follow-ups
+- Model Used
+
+If the change is user-visible, include screenshots or a short video.
+
+For UI or UX changes, before/after images are required. PRs that change the interface or interaction flow without before/after images will likely be ignored, or closed in a regular maintainer sweep without review.
+
+### 4. AI Usage Disclosure
+
+If AI helped produce the change, say so in the PR description.
+
+Include:
+
+- Provider
+- Exact model name or ID
+- Any notable mode or tool use that mattered
+
+If no AI was used, write: `None — human-authored`.
+
+### 5. Keep the Docs Honest
+
+If your change alters behavior, architecture, testing, setup, or user-facing workflows, update the relevant docs in the same PR.
+
+Common files:
+
+- [README.md](README.md) for setup, usage, and contributor-facing commands
+- [ROADMAP.md](ROADMAP.md) for shipped features and sprint history
+- [ARCHITECTURE.md](ARCHITECTURE.md) for implementation details and design constraints
+- [TESTING.md](TESTING.md) for manual and automated verification guidance
+- [CHANGELOG.md](CHANGELOG.md) when maintainers want release-note-ready entries
+
+## Project-Specific Guidelines
+
+### Preserve the Design Constraints
+
+Hermes WebUI is deliberately:
+
+- No build step
+- No bundler
+- No frontend framework
+- Easy to modify from a terminal
+
+Do not introduce new infrastructure or dependencies unless the gain is clear and the tradeoff is justified.
+
+### Match the Existing Shape of the Codebase
+
+- Server logic belongs in `api/` with `server.py` staying thin
+- Frontend behavior belongs in the existing `static/*.js` modules
+- Prefer extending current patterns over introducing parallel abstractions
+- Keep changes legible to future contributors working directly from the repo in a terminal
+
+### Be Careful With User-Facing Changes
+
+This project is heavily UI-driven. If you change interaction flows, session behavior, workspace browsing, onboarding, or mobile layouts:
+
+- test the happy path
+- test reload behavior where relevant
+- test narrow/mobile layouts where relevant
+- include before/after images in the PR
+
+### Security and Safety Matter
+
+This app can expose workspace contents, run agent actions, and optionally sit behind a reverse proxy or Docker deployment. Treat auth, path handling, uploads, streaming, and environment handling as high-risk areas.
+
+If your PR touches security-sensitive behavior, say so explicitly in the PR description and explain how you verified it.
+
+## Writing a Good PR Message
+
+Start with a short Thinking Path that explains the chain from project goal to the specific fix.
+
+Example:
+
+> - Hermes WebUI aims for near 1:1 parity with the Hermes CLI in a browser
+> - Long-running chat turns rely on SSE streaming and session recovery
+> - Reloading during an in-flight turn can leave the UI in an inconsistent state
+> - The bug was that recovered sessions restored messages but not the live stream state
+> - This PR fixes the recovery path so in-flight turns reconnect cleanly after reload
+> - The benefit is that users can refresh or reconnect without losing visibility into active work
+
+Another example:
+
+> - Hermes WebUI is intentionally a simple Python + vanilla JS application
+> - The right panel is used for workspace browsing and previews
+> - On mobile, panel state changes need to be obvious and touch-friendly
+> - The existing close affordance was inconsistent with the bottom-nav flow
+> - This PR fixes the mobile panel close behavior and aligns it with the current navigation model
+> - The result is fewer dead-end UI states on phones
+
+After that, cover:
+
+- what you changed
+- why you changed it
+- how you verified it
+- what risks remain
+
+## Review Tips
+
+Want the smoothest review?
+
+- Keep diffs tight
+- Name things clearly
+- Avoid unnecessary rewrites
+- Add short comments only where the code would otherwise be hard to follow
+- Respond directly to review feedback and update the PR description if the scope changes
+
+## Development References
+
+- [README.md](README.md)
+- [ARCHITECTURE.md](ARCHITECTURE.md)
+- [TESTING.md](TESTING.md)
+- [ROADMAP.md](ROADMAP.md)
+- [SPRINTS.md](SPRINTS.md)
+
+Questions are best raised early, before a large change is finished.

Dockerfile

@@ -3,21 +3,94 @@ FROM python:3.12-slim
 LABEL maintainer="nesquena"
 LABEL description="Hermes Web UI — browser interface for Hermes Agent"
 
-WORKDIR /app
+# Install system packages
+ENV DEBIAN_FRONTEND=noninteractive
 
-# Copy source
-COPY . /app
+# Make use of apt-cacher-ng if available
+RUN if [ "A${BUILD_APT_PROXY:-}" != "A" ]; then \
+        echo "Using APT proxy: ${BUILD_APT_PROXY}"; \
+        printf 'Acquire::http::Proxy "%s";\n' "$BUILD_APT_PROXY" > /etc/apt/apt.conf.d/01proxy; \
+    fi \
+    && apt-get update \
+    && apt-get install -y --no-install-recommends ca-certificates wget gnupg \
+    && rm -rf /var/lib/apt/lists/* \
+    && apt-get clean
 
-# Install Python dependencies
-RUN pip install --no-cache-dir -r requirements.txt
+RUN apt-get update -y --fix-missing --no-install-recommends \
+    && apt-get install -y --no-install-recommends \
+    apt-utils \
+    locales \
+    ca-certificates \
+    sudo \
+    curl \
+    rsync \
+    openssh-client \
+    && apt-get upgrade -y \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
+
+# UTF-8
+RUN localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8
+ENV LANG=en_US.utf8
+ENV LC_ALL=C
+
+# Set environment variables
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PYTHONIOENCODING=utf-8
+
+WORKDIR /apptoo
+
+# Every sudo group user does not need a password
+RUN echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers
+
+# Create a new group for the hermeswebui and hermeswebuitoo users
+RUN groupadd -g 1024 hermeswebui \ 
+    && groupadd -g 1025 hermeswebuitoo
+
+# The hermeswebui (resp. hermeswebuitoo) user will have UID 1024 (resp. 1025), 
+# be part of the hermeswebui (resp. hermeswebuitoo) and users groups and be sudo capable (passwordless) 
+RUN useradd -u 1024 -d /home/hermeswebui -g hermeswebui -s /bin/bash -m hermeswebui \
+    && usermod -G users hermeswebui \
+    && adduser hermeswebui sudo
+RUN useradd -u 1025 -d /home/hermeswebuitoo -g hermeswebuitoo -s /bin/bash -m hermeswebuitoo \
+    && usermod -G users hermeswebuitoo \
+    && adduser hermeswebuitoo sudo
+RUN chown -R hermeswebuitoo:hermeswebuitoo /apptoo
+
+USER root
+
+COPY --chmod=555 docker_init.bash /hermeswebui_init.bash
+
+RUN touch /.within_container
+
+# Remove APT proxy configuration and clean up APT downloaded files
+RUN rm -rf /var/lib/apt/lists/* /etc/apt/apt.conf.d/01proxy \
+    && apt-get clean
+
+USER root
+
+# Pre-install uv system-wide so the container doesn't need internet access at runtime.
+# Installing as root places uv in /usr/local/bin, available to all users.
+# The init script will skip the download when uv is already on PATH.
+RUN curl -LsSf https://astral.sh/uv/install.sh | env UV_INSTALL_DIR=/usr/local/bin sh
+
+USER hermeswebuitoo
+
+COPY --chown=hermeswebuitoo:hermeswebuitoo . /apptoo
+
+# Bake the git version tag into the image so the settings badge works even
+# when .git is not present (it is excluded by .dockerignore).
+# CI passes: --build-arg HERMES_VERSION=$(git describe --tags --always)
+# Local builds that omit the arg get "unknown" as the fallback.
+ARG HERMES_VERSION=unknown
+RUN echo "__version__ = '${HERMES_VERSION}'" > /apptoo/api/_version.py
 
 # Default to binding all interfaces (required for container networking)
 ENV HERMES_WEBUI_HOST=0.0.0.0
 ENV HERMES_WEBUI_PORT=8787
 
-# State directory (mount as volume for persistence)
-ENV HERMES_WEBUI_STATE_DIR=/data
-
 EXPOSE 8787
 
-CMD ["python", "server.py"]
+CMD ["/hermeswebui_init.bash"]
+

HERMES.md

@@ -1,165 +1,176 @@
 # Why Hermes
 
-Hermes is a persistent, autonomous AI agent that lives on your server. It remembers everything,
-schedules work while you sleep, and gets more capable the longer it runs. This document explains
-the mental model, why that matters, and how Hermes compares to every major AI tool available today.
+Hermes is a persistent, autonomous AI agent that runs on your server. It has layered memory that
+accumulates across sessions, a cron scheduler that fires jobs while you're offline, and a
+self-improving skills system that saves reusable procedures automatically. You reach it from a
+terminal, a browser, or a messaging app — and it's the same agent with the same history every time.
+
+This document explains the mental model, how Hermes compares to other tools honestly, and where
+it is and is not the right choice.
 
 ---
 
-## The Core Idea: Assistants Forget. Agents Don't.
+## The real problem: most tools are excellent in the moment and weak over time
 
-Every time you open Claude Code, Codex, or a chat window, the tool starts from zero. It does not
-know who you are, what you worked on yesterday, how your repo is structured, or what bugs you
-already fixed. You re-explain yourself every single session. The tool is powerful in the moment
-and useless the next day.
+Memory is no longer a differentiator on its own. ChatGPT, Claude, Cursor, and GitHub Copilot all
+have some form of memory now. Anthropic, OpenAI, and Microsoft are all shipping scheduling and
+agent features. The category boundaries that existed twelve months ago are blurring fast.
 
-Hermes fills that gap. It runs on your server, retains context across every session, and acts
-on your behalf whether or not you are at a keyboard.
+Hermes is not the only tool with memory or automation. It is the tool that makes those
+capabilities durable, self-hosted, cross-surface, and cumulative on your own server. The
+distinction that matters is not "has memory" vs. "has no memory" — it's whether context persists
+across sessions automatically, whether execution happens on hardware you control, whether you can
+reach the same agent identity from any device, and whether the system gets meaningfully better at
+your specific workflow over time without manual configuration.
 
 ```
-Assistant model:  You -> [Tool] -> Answer -> Done
-                  (tool forgets everything when the window closes)
+Session-scoped:   You -> [Tool] -> Answer -> Done
+                  (some tools now carry memory, but the execution is stateless)
 
-Agent model:      You <-> [Hermes] <-> (memory, skills, schedule, tools)
-                  (persistent, learns your stack, acts on your behalf, runs while you're offline)
+Persistent agent: You <-> [Hermes] <-> (memory, skills, schedule, tools, surfaces)
+                  (runs on your server, accumulates context, acts on your behalf offline)
 ```
 
 ---
 
-## The Three Pillars
+## A note on convergence
 
-### 1. Memory That Compounds
+The market is converging. Chat assistants are adding task scheduling and file connectors. IDE
+tools are launching cloud agent modes. CLI tools are adding skills systems and mobile surfaces.
+The lines between "assistant," "editor," and "agent" are dissolving.
 
-Hermes has layered memory that survives every session, every reboot, every model swap:
+This makes comparisons harder but also makes the question sharper: what actually matters when
+every tool is claiming some version of every feature? For Hermes, the answer is synthesis. Any
+single feature — memory, scheduling, messaging — is available somewhere else. The value is
+having all of them in one self-hosted system, running continuously, with a persistent identity
+that accumulates real knowledge of your stack over time.
 
-- **User profile** -- who you are, your preferences, your communication style, things you've
-  corrected Hermes on
-- **Agent memory** -- facts about your environment, your toolchain, your project conventions
-- **Skills** -- reusable procedures Hermes discovers and saves; it never has to relearn how to
-  deploy your app, run your tests, or review a PR
-- **Session history** -- every past conversation is searchable; Hermes can recall what you
-  worked on last Tuesday
+---
+
+## The three pillars
+
+### 1. Memory that compounds
+
+Hermes has layered memory that survives every session, every reboot, and every model swap:
+
+- User profile — who you are, your preferences, your communication style, things you've corrected Hermes on
+- Agent memory — facts about your environment, your toolchain, your project conventions
+- Skills — reusable procedures Hermes discovers and saves automatically; it never has to relearn how to deploy your app, run your tests, or review a PR
+- Session history — every past conversation is searchable; Hermes can recall what you worked on last Tuesday
 
 When you correct Hermes, it remembers. When it solves a tricky problem, it saves the approach.
-When it learns your stack, that knowledge carries into every future session.
+When it learns your stack, that knowledge carries into every future session. You never configure
+this manually — it happens in the background as a side effect of normal use.
 
-### 2. Autonomous Scheduling
+### 2. Autonomous scheduling
 
-Hermes can run jobs without you present -- every hour, every morning, on any cron schedule.
-It fires up a fresh session, runs the task, and delivers the result to wherever you want it:
-Telegram, Discord, Slack, Signal, WhatsApp, SMS, email, and more.
+Hermes can run jobs without you present — every hour, every morning, on any cron schedule. It
+fires up a fresh session with full access to your memory and skills, runs the task, and delivers
+the result wherever you want it: Telegram, Discord, Slack, Signal, WhatsApp, SMS, email, and more.
 
 Things Hermes can do while you sleep:
 
 - Review new pull requests on your GitHub repo and post a full verdict comment
-- Send you a morning briefing of news, markets, or anything else you care about
+- Send a morning briefing of news, markets, or anything else you track
 - Run your test suite and alert you if something breaks
 - Watch a competitor's blog for new posts and summarize them
 - Monitor a datasource and notify you when a threshold is crossed
 
-### 3. Reach It From Anywhere
+The difference from cloud-scheduled alternatives is that the job runs on your server, with your
+memory and skills, and your data never leaves your hardware.
+
+### 3. Reach it from anywhere
 
 Hermes runs on your server and is reachable from every surface: terminal over SSH, the web UI
 (this project), and messaging apps including Telegram, Discord, Slack, WhatsApp, Signal, and
 Matrix. Start a task from your phone, check it from the browser on your laptop, continue it in
-a terminal on a remote server. The same agent, memory, and history follow you everywhere.
+a terminal on a remote server. The same agent, memory, and history follow you across all of them.
 
 ---
 
-## A Framework for AI Tools
+## How AI tools are layered today
 
-There are four distinct categories of AI tool. Understanding the category tells you what a tool
-can and cannot do.
+The old four-category model — chat, editor, CLI, agent — is too clean. These layers are actively
+collapsing into each other. Here is a more honest picture:
 
-### Category 1: Chat Assistants
-*Claude.ai, ChatGPT, Gemini*
+Chat assistants (Claude.ai, ChatGPT) now have persistent memory, task scheduling, 50+ service
+connectors, and in some cases full agent modes with computer use. They are no longer "just chat."
 
-You open a window, ask something, get an answer. No persistent memory beyond the conversation,
-no ability to run code or touch files, no way to act on your behalf. Excellent for Q&A,
-drafting, and brainstorming. You re-explain your context every session.
+IDE tools (Cursor, Windsurf, Copilot) have shipped or are shipping cross-session memory,
+cloud-based background agents, and in Cursor's case a full Automations platform with Slack
+integration. Cursor v3.0 (April 2026) is explicitly agent-first.
 
-### Category 2: IDE Integrations
-*GitHub Copilot, Cursor, Windsurf, Zed AI*
+CLI tools (Claude Code, Codex, OpenCode) have added hooks, skills, desktop app automations,
+and multi-surface reach. Claude Code now spans terminal, IDE, desktop, and browser. Codex has
+become a product family: CLI, IDE extension, desktop app, and Codex Cloud.
 
-Deep inside your editor. Autocomplete, inline diffs, refactors -- all excellent. Windsurf was
-earliest with workspace-scoped memory (Cascade Memories); Copilot has been shipping repo-level
-memory since late 2025 and is catching up. Cursor has no native memory as of early 2026. None
-have scheduling or messaging access. Tied to one machine and one editor.
+Persistent self-hosted agents (Hermes, OpenClaw) sit at the intersection: they combine the
+tool-use power of CLI agents, the memory of chat assistants, the scheduling of automation
+platforms, and the cross-surface reach of messaging integrations — running continuously on
+hardware you own.
 
-### Category 3: Agentic CLI Tools
-*Claude Code, Codex CLI, OpenCode, Aider*
-
-The current frontier for most developers. Can use real tools -- run shell commands, read and
-write files, search the web, call APIs. Great for deep, multi-step tasks in a single terminal
-session. All are adding memory and scheduling features to varying degrees (see comparisons below),
-but the core model is still session-scoped: you invoke it, it works, it stops.
-
-### Category 4: Persistent Autonomous Agents
-*Hermes, OpenClaw (as of early 2026)*
-
-All the tool use of Category 3, plus memory that accumulates across sessions, plus always-on
-scheduling, plus multi-modal access from any device or messaging app. Gets more useful over time
-rather than resetting to zero. Hermes and OpenClaw are the two primary open-source, self-hosted
-tools in this category. OpenClaw is a gateway-centric automation platform; Hermes is a
-self-improving agent that writes and reuses its own procedures from experience.
+The question is not which category a tool belongs to. The question is which combination of
+capabilities you actually need, where that execution lives, and whether the system gets better
+at your specific context over time.
 
 ---
 
-## How Hermes Compares
+## How Hermes compares
 
 ### vs. OpenClaw
 
-OpenClaw is the most direct comparison to Hermes and the question most people ask first.
-Both are open-source, self-hosted, always-on agents with persistent memory, cron scheduling,
-and messaging app integration. If you're evaluating Hermes, you should evaluate OpenClaw too.
+OpenClaw is the most direct comparison and the question most people ask first. Both are
+open-source, self-hosted, always-on agents with persistent memory, cron scheduling, and messaging
+app integration. If you're evaluating Hermes, evaluate OpenClaw too.
 
-OpenClaw (MIT, ~347k GitHub stars) is built around a **Gateway** control plane written in
-Node.js/TypeScript. It excels at broad personal automation: native Chrome/Chromium control for
-browser automation, the widest messaging platform support in the space (WhatsApp, Telegram,
-Signal, iMessage, LINE, WeChat, Slack, Discord, Teams, Matrix, and more), voice wake words,
-and a ClawHub skill marketplace where users share pre-built automations. The community is large
-and the ecosystem is growing fast.
+OpenClaw (MIT) is built around a Gateway control plane written in Node.js/TypeScript. It has the
+widest messaging coverage in the space — 24+ channels including WhatsApp, Telegram, Signal,
+iMessage, LINE, WeChat, Slack, Discord, Teams, Matrix, Google Chat, Feishu, Mattermost, IRC,
+Nextcloud Talk, and more. It has native Chrome/Chromium control via CDP, voice wake words on
+macOS and iOS, and a ClawHub marketplace with 10,700+ skills. The community is large (350k+
+GitHub stars, 16,900+ commits) and growing.
 
-Hermes takes a different approach. It is built in Python and centers on a **self-improving
-agent loop** rather than a gateway control plane. The core difference is in how skills work:
-OpenClaw skills are primarily human-authored plugins installed from a marketplace; Hermes
-**writes and saves its own skills automatically** as part of every session. When Hermes solves
-a problem a new way, it saves the procedure and reuses it going forward without any user effort.
+Hermes is built in Python and centers on a self-improving agent loop rather than a gateway
+control plane. The core architectural difference is in skills: OpenClaw skills are primarily
+human-authored plugins installed from a marketplace. Hermes writes and saves its own skills
+automatically as part of every session. When Hermes solves a problem a new way, it saves the
+procedure and reuses it without any user effort. That's not a subtle distinction — it's the
+reason Hermes gets meaningfully better at your workflow without you maintaining a plugin library.
 
-Beyond the skills architecture, there are two other practical differences worth knowing:
+Two practical differences worth knowing directly:
 
-**Stability.** OpenClaw's community forums and GitHub issues document a recurring pattern of
-update-breaking regressions -- for example, Telegram integration was broken across multiple
-releases in early 2026. The unofficial WhatsApp Web protocol OpenClaw uses is known to
-disconnect and requires periodic re-pairing (this is documented in OpenClaw's own FAQ).
-Hermes has had no equivalent release breakages.
+Stability. OpenClaw's GitHub issues and community forums document recurring update-breaking
+regressions. Telegram integration was broken across multiple releases from early 2026 through
+at least April 2026. The unofficial WhatsApp Web protocol OpenClaw relies on disconnects and
+requires periodic re-pairing — this is in OpenClaw's own FAQ.
 
-**Security.** ClawHub's open publishing model has been exploited repeatedly. A community audit
-identified over a thousand malicious skills in the marketplace including prompt injections and
-tool-poisoning payloads; the community-maintained awesome-openclaw-skills list tracks confirmed
-removals and flags known bad actors. Hermes has no third-party marketplace and a correspondingly
-smaller attack surface.
+Security. ClawHub's open publishing model has been exploited at scale. Three separate audits in
+early 2026 found serious problems: Koi Security (January 2026) linked 335 skills to a campaign
+called "ClawHavoc" that delivered Atomic Stealer malware on macOS; Bitdefender found roughly
+900 malicious packages representing about 20% of the ecosystem at the time; Snyk's "ToxicSkills"
+report (February 2026) found malicious skills across roughly 4,000 scanned packages. China's
+CNCERT issued a national warning about ClawHub. Hermes has no third-party marketplace and a
+correspondingly smaller attack surface.
 
-**OpenClaw's genuine strengths** are worth stating plainly: it has broader messaging coverage
-(iMessage, LINE, WeChat, Teams -- platforms Hermes does not support), native browser and
-computer control via Chrome CDP, voice wake words on macOS and iOS, a larger community, and
-more third-party integrations than Hermes. If those capabilities matter most to you, OpenClaw
-is worth a serious look.
+OpenClaw's genuine strengths are worth stating plainly: broader messaging coverage (iMessage,
+LINE, WeChat, Teams, Google Chat — platforms Hermes does not support), native browser and
+computer control via Chrome CDP, voice wake words, a larger community, and more third-party
+integrations than Hermes. If those capabilities matter most, OpenClaw is worth a serious look.
 
-Where Hermes is the better fit: you want an agent that self-improves from experience without
-manual plugin authoring, you work in Python and want access to the ML/data science ecosystem,
-you want a stable deployment that does not break between updates, or you want a full web chat
-UI rather than a monitoring dashboard.
+Where Hermes fits better: you want an agent that self-improves from experience without managing
+a plugin library, you work in Python and want the ML/data science ecosystem, you want a stable
+deployment that doesn't break between updates, or you want a full web chat UI rather than a
+control dashboard.
 
 | | OpenClaw | Hermes |
 |---|---|---|
 | Persistent memory | Yes | Yes |
 | Scheduled jobs (cron) | Yes | Yes |
-| Messaging app access | Yes (15+ platforms, incl. iMessage/WeChat) | Yes (10+ platforms) |
-| Web UI | Gateway dashboard (monitoring only) | Full three-panel chat UI |
+| Messaging app access | Yes (24+ platforms, incl. iMessage/WeChat/LINE) | Yes (many platforms) |
+| Web UI | Chat UI + control dashboard | Full three-panel chat UI |
 | Self-hosted | Yes | Yes |
 | Open source | Yes (MIT) | Yes |
-| Self-improving skills | Partial (AI can generate skills; not the default loop) | Yes (automatic, first-class) |
+| Self-improving skills | Partial (AI can generate; not the default loop) | Yes (automatic, first-class) |
 | Browser / computer control | Yes (native Chrome CDP) | Via shell / tools |
 | Voice wake words | Yes (macOS/iOS) | No |
 | Python / ML ecosystem | No (Node.js) | Yes |
@@ -167,209 +178,312 @@ UI rather than a monitoring dashboard.
 | Multi-profile support | Via binding-rule routing | Yes (first-class named profiles) |
 | Provider-agnostic | Yes | Yes |
 | Update reliability | Moderate (documented regressions) | High |
+| Memory inspectability | Limited | Yes (markdown files, editable) |
+| Self-hosted autonomous execution | Yes | Yes |
 
 ### vs. Claude Code (Anthropic)
 
-Claude Code is Anthropic's official agentic CLI and one of the best tools in Category 3.
-In a single focused session it is capable -- deep code understanding, shell access, file
-editing, multi-step reasoning.
+Claude Code is Anthropic's official agentic tool and one of the strongest options for focused
+coding sessions. It has deep code understanding, shell access, file editing, and multi-step
+reasoning. It has been expanding rapidly — it now spans terminal, IDE plugin, desktop app, and
+browser surfaces — and the gap is closing in several areas.
 
-Claude Code has been adding features rapidly and the gap is narrowing:
+What Claude Code has that's worth knowing:
 
-- **Hooks system** -- 13 event types (SessionStart, PreToolUse, PostToolUse, Stop, etc.) with
-  4 handler types (shell command, HTTP endpoint, LLM prompt, sub-agent); deterministic
+- Hooks system — 26 event types (SessionStart, PreToolUse, PostToolUse, Stop, and more) with
+  4 handler types (shell command, HTTP endpoint, LLM prompt, sub-agent); gives deterministic
   non-LLM control over the agent lifecycle
-- **Plugins / Skills** -- installable via `/plugin install`, hot-reloaded from `~/.claude/skills`,
-  with a marketplace; skills and slash commands unified as of v2.1.0
-- **Scheduling** -- `/loop` (session-scoped), cloud-managed cron via `claude.ai/code/scheduled`
-  (Anthropic infrastructure, minimum interval applies), and desktop app automations
-- **Messaging channels** -- Telegram, Discord, iMessage, and webhooks via the Channels feature
-  (research preview, v2.1.80+); deep Slack integration that triggers cloud sessions and creates PRs
-- **Claude Cowork** -- a separate product for knowledge workers; connects to 38+
-  services via MCP including Slack, Gmail, Microsoft Teams, Notion, Jira, Salesforce, and more
-- **Memory** -- CLAUDE.md and MEMORY.md for project-level context; auto-memory rolling out
+- Plugins / Skills — installable via `/plugin install`, hot-reloaded from `~/.claude/skills`,
+  with a marketplace; includes the official ralph-wiggum plugin (`/ralph-loop`) for
+  autonomous iteration toward a completion goal (distinct from `/loop`)
+- `/loop` — a native bundled skill, available in every session without any plugin, that runs
+  a prompt on a repeating schedule within an active CLI session (polling/monitoring use case);
+  session-scoped, dies when the terminal closes
+- Scheduling — cloud-managed cron (Anthropic infrastructure, minimum 1-hour interval) and
+  desktop app scheduled tasks (run locally while the app is open, minimum 1-minute interval,
+  full local file access); no self-hosted cron
+- Messaging channels — Telegram, Discord, and iMessage via the Channels feature (research
+  preview, requires Bun runtime); Slack is the most-requested addition and has not yet shipped
+- Memory — CLAUDE.md and MEMORY.md for project-level context; auto-memory since v2.1.59+
+- Claude Cowork — a separate knowledge-worker product connecting 38+ services via MCP
+  including Gmail, Microsoft Teams, Notion, Jira, Salesforce, and more
 
-These are real features. The key differences that remain:
+Claude Code's source was briefly and accidentally made public in March 2026 before being taken
+down. The CLI ships as minified/bundled TypeScript compiled with Bun — it is not open source.
 
-- Claude Code's scheduling runs on **Anthropic's cloud** (or requires the desktop app open),
-  not a self-hosted server; cloud jobs have a minimum interval and your data leaves your hardware
-- Memory is **project-file-based** (CLAUDE.md / MEMORY.md), not a knowledge graph that
-  accumulates automatically across all your work; auto-memory is still rolling out
-- **Not provider-agnostic** -- routes through Bedrock or Vertex but always hits a Claude model;
-  you cannot switch to GPT, Gemini, or a local model
-- **Not open source** -- proprietary; the CLI ships obfuscated JavaScript
-- Messaging channels are a **research preview** requiring Bun runtime; not yet production-grade
+Key differences that remain:
+
+- Scheduling requires cloud (Anthropic infrastructure, data off your hardware, 1-hour minimum)
+  or the desktop app (runs locally, but the app must stay open — not a headless server process);
+  neither runs as a server daemon the way Hermes cron does
+- Memory is project-file-based (CLAUDE.md / MEMORY.md plus rolling auto-memory); it doesn't
+  automatically accumulate a cross-project knowledge graph the way Hermes does
+- Not provider-agnostic — routes through Anthropic, Bedrock, Vertex, or Foundry, but always
+  a Claude model; you can't switch to GPT, Gemini, or a local model
+- Messaging channels are still a research preview, not production
 
 Hermes can use Claude Code as a sub-agent. For large implementation tasks, Hermes can spawn
 Claude Code to handle the heavy lifting and fold the result back into its own memory and history.
 
 | | Claude Code | Hermes |
 |---|---|---|
-| Persistent memory (automatic) | Partial (CLAUDE.md / MEMORY.md, rolling out) | Yes |
-| Skills / hooks system | Yes (Hooks + Plugin/Skills marketplace) | Yes (auto-generated from experience) |
+| Persistent memory (automatic) | Partial (CLAUDE.md / MEMORY.md + auto-memory v2.1.59+) | Yes |
+| Skills / hooks system | Yes (26-event Hooks + Plugin/Skills marketplace) | Yes (auto-generated from experience) |
 | Scheduled jobs (self-hosted) | No (cloud or desktop-app only) | Yes |
-| Messaging access | Partial (Telegram/Discord/iMessage via research preview; Slack native) | Yes (10+ platforms, production) |
+| Messaging access | Partial (Telegram/Discord/iMessage research preview; Slack not yet) | Yes (many platforms, production) |
 | Cowork connectors (Slack, Gmail, etc.) | Yes (via Claude Cowork, separate product) | Via agent tool use |
 | Web UI | Yes (claude.ai/code, Anthropic-hosted) | Yes (self-hosted) |
-| Provider-agnostic | No (Claude models only, via Bedrock/Vertex) | Yes (any provider) |
+| Provider-agnostic | No (Claude models only) | Yes (any provider) |
 | Self-hosted scheduling | No | Yes |
 | Open source | No | Yes |
+| Background/cloud agent mode | Yes (cloud-scheduled) | Yes (self-hosted cron) |
 | Runs as sub-agent of Hermes | Yes | N/A |
+| Memory inspectability | Partial (CLAUDE.md readable; auto-memory less so) | Yes (markdown files) |
 
 ### vs. Codex CLI (OpenAI)
 
-Codex CLI is OpenAI's open-source agentic terminal tool (Apache 2.0, ~73k GitHub stars). It
-supports 10+ providers including Anthropic, Google, Mistral, Groq, and local models via Ollama.
-It added persistent session memory in v0.100.0 with `codex resume`. The desktop app has an
-Automations feature for scheduled local tasks.
+Codex CLI (Apache 2.0, ~60k GitHub stars) started as a straightforward terminal tool and has
+expanded into a product family. It was rewritten from TypeScript to Rust. It now includes an IDE
+extension, a desktop app with an Automations feature, and Codex Cloud for remote execution. A
+Skills system is shared across surfaces. It supports 12+ built-in providers: OpenAI, Anthropic,
+Google/Gemini, Mistral, Groq, Ollama, OpenRouter, LM Studio, Together AI, DeepSeek, xAI,
+Azure OpenAI, and custom endpoints.
 
-The CLI itself has no native scheduling (open feature request as of early 2026). Memory is
-session-history-based rather than a living knowledge graph. No messaging app access. A strong
-tool for single-session coding; Hermes adds the always-on layer on top.
+The CLI itself has no native scheduling (open feature request). Session continuity is available
+via `codex resume`. Memory is session-history-based plus AGENTS.md project context — not a
+living knowledge graph that accumulates across all your projects. No first-party messaging
+integration. The Automations feature in the desktop app covers scheduled local tasks but doesn't
+reach the cross-session, cross-surface continuity Hermes has.
 
 | | Codex CLI | Hermes |
 |---|---|---|
 | Persistent memory | Partial (session history + AGENTS.md) | Yes (automatic, layered) |
-| Scheduled jobs | Partial (desktop app only; CLI has none) | Yes |
+| Scheduled jobs | Partial (desktop app Automations; CLI has none) | Yes |
 | Messaging app access | No | Yes |
-| Web UI | No | Yes (self-hosted) |
-| Provider-agnostic | Yes (10+ providers) | Yes (10+ providers) |
+| Web UI | No (CLI + desktop app) | Yes (self-hosted) |
+| Provider-agnostic | Yes (12+ providers) | Yes |
 | Self-hosted | Yes | Yes |
 | Open source | Yes (Apache 2.0) | Yes |
+| Background/cloud agent mode | Yes (Codex Cloud) | Yes (self-hosted cron) |
+| Self-improving skills | No | Yes |
 
 ### vs. OpenCode
 
-OpenCode is an open-source TUI agentic coding assistant, provider-agnostic across 75+ providers.
-It has a WebUI embedded in its binary and an official desktop app. It uses SQLite for session
-history and AGENTS.md for project context.
+OpenCode is an open-source TUI agentic coding assistant supporting 75+ providers. It has a WebUI
+embedded in its binary, an official desktop app, SQLite session history, and AGENTS.md project
+context. It supports CLAUDE.md as a fallback for users migrating from Claude Code. There are 30+
+community plugins, and community messaging integrations exist for Telegram, Slack, Discord, and
+Microsoft Teams — though none are first-party and all require manual setup.
 
-No native scheduled jobs (a community background plugin exists), no first-party messaging
-integration (community Telegram bots exist but require manual setup), and no automatic
-cross-session semantic memory. Good for interactive terminal coding sessions.
+OpenCode Go ($10/month) and OpenCode Zen (curated model service) are subscription tiers. The
+GitHub Copilot official integration launched January 2026. There is no native scheduling; a
+community background plugin exists. No automatic cross-session semantic memory.
 
 | | OpenCode | Hermes |
 |---|---|---|
 | Persistent memory | Partial (session history + AGENTS.md) | Yes (automatic, layered) |
 | Scheduled jobs | No (community plugin only) | Yes |
-| Messaging app access | No (community Telegram bot only) | Yes (first-party, 10+ platforms) |
+| Messaging app access | Community integrations only (Telegram/Slack/Discord/Teams) | Yes (first-party, many platforms) |
 | Web UI | Yes (embedded + desktop app) | Yes (self-hosted) |
 | Mobile access | No | Yes |
-| Skills system | No | Yes |
+| Skills / plugins | Yes (30+ community plugins) | Yes (auto-generated, first-party) |
 | Provider-agnostic | Yes (75+ providers) | Yes |
 | Open source | Yes | Yes |
+| Self-hosted autonomous execution | No | Yes |
 
-### vs. Cursor / Windsurf / Copilot
+### vs. Cursor
 
-Category 2 tools -- exceptional at in-editor autocomplete, inline diffs, and code review.
-Not competing for the same job as Hermes, and they work well alongside it.
+Cursor has changed substantially. The "no memory, no scheduling, no messaging" description was
+accurate in 2024 and is wrong now.
 
-Windsurf was earliest with workspace-scoped memory (Cascade Memories); Copilot has been
-shipping repo-level memory since late 2025. Cursor has no native cross-session memory as of
-early 2026. None have scheduling or messaging access.
+Memories (per-project cross-session knowledge base) shipped in beta with v1.0 in June 2025.
+Automations launched March 5, 2026 — time-based, event-based (GitHub/Linear/PagerDuty), and
+communication-based (Slack) triggers that fire background agents on cloud VMs. The web app,
+mobile agent, and Slack bot give it multi-surface reach. Cursor v3.0 (April 2, 2026) is
+explicitly agent-first with Design Mode and 30+ marketplace plugins. Cursor acquired Supermaven
+for autocomplete. As of early 2026 it's valued at $29.3B with $2B ARR. It is not a narrow editor
+tool anymore.
+
+Hermes still has a different profile: it's self-hosted and server-resident, the same persistent
+identity follows you across every surface without cloud intermediation, and it works with any
+model family rather than being cloud-VM-based. For workflows that require data sovereignty,
+self-hosted scheduling, or deep Python/ML tooling on your own hardware, Cursor's cloud-agent
+architecture is a fundamental mismatch. For teams that want editor-native agents with strong
+IDE integration, Cursor's recent evolution is significant.
 
 | | Cursor | Windsurf | Copilot | Hermes |
 |---|---|---|---|---|
-| In-editor autocomplete | Excellent | Excellent | Excellent | No |
+| In-editor autocomplete | Excellent (Supermaven) | Excellent (Cascade) | Excellent | No |
 | Inline diff / refactor | Yes | Yes | Yes | Via shell |
-| Cross-session memory | No | Yes (workspace) | Partial (repo, early access) | Yes |
-| Scheduled background jobs | No | No | No | Yes |
-| Messaging app / mobile | No | No | No | Yes |
+| Cross-session memory | Yes (Memories, per-project) | Yes (Cascade Memories, workspace) | Yes (Agentic Memory, repo-scoped, 28-day expiry) | Yes (automatic, persistent) |
+| Scheduled background jobs | Yes (Automations, cloud VM) | No | Via Coding Agent (issue-driven) | Yes (self-hosted cron) |
+| Messaging app / multi-surface | Yes (Slack bot, web app, mobile) | No | Via Copilot CLI / fleet | Yes (many platforms) |
+| Background/cloud agent mode | Yes (Automations on cloud VMs) | No | Yes (Coding Agent, GA Mar 2026) | Yes (self-hosted) |
 | Terminal tool use | Limited | Limited | Limited | Full |
 | Self-hosted | No | No | No | Yes |
-| Provider-agnostic | Partial | Partial | No | Yes |
+| Self-hosted autonomous execution | No | No | No | Yes |
+| Provider-agnostic | Partial | Partial | No (GitHub models) | Yes |
 | Open source | No | No | No | Yes |
+| Memory inspectability | Partial | Yes (stored locally) | Limited | Yes (markdown files) |
 
-### vs. Claude.ai / ChatGPT
+### vs. Claude.ai and ChatGPT
 
-Category 1. For drafting, Q&A, and brainstorming in the moment, both are excellent.
+These are no longer simple chat tools. The description of "no memory, no scheduling, no
+messaging" is inaccurate for both.
 
-Claude.ai memory has been improving -- it now generates memory from chat history, not just
-user-curated entries. Claude.ai can also execute code and read/write files in a sandboxed
-environment via Artifacts. These are real capabilities, just not the same as direct filesystem
-or shell access on your own server.
+Claude Cowork (in Claude Desktop) launched scheduled tasks on February 25, 2026 — hourly,
+daily, weekly, weekdays, and on-demand. It runs in an isolated VM with file and shell access.
+Claude has 50+ service connectors as of February 2026 including Slack (launched January 26,
+2026), Gmail, Google Calendar, Google Drive, Microsoft 365, Notion, Asana, Linear, and Jira.
+Memory auto-generates from chat history, not just user-curated entries. Code execution and
+file access in Artifacts is sandboxed, not the same as shell access on your own server.
 
-| | Claude.ai / ChatGPT | Hermes |
-|---|---|---|
-| Memory across conversations | Yes (improving; auto-generated from history) | Yes (deep, automatic) |
-| Runs shell commands | No | Yes |
-| Code execution | Sandboxed (Artifacts) | Yes (full shell) |
-| Reads / writes files | Sandboxed (Artifacts) | Yes (full filesystem) |
-| Schedules background jobs | No | Yes |
-| Web UI | Yes | Yes |
-| Messaging apps | No | Yes |
-| Self-hosted | No | Yes |
-| Provider-agnostic | No | Yes |
-| Open source | No | Yes |
+ChatGPT has Agent Mode (launched July 17, 2025), Scheduled Tasks (January 2025, recurring
+automated prompts), a computer-using agent, Projects, 50+ connectors including Gmail, GitHub,
+and Google Drive, dual-mode memory (auto + manual), and ChatGPT Pulse for Pro users (daily
+research briefings). It is not a passive Q&A interface.
+
+Where Claude.ai and ChatGPT differ from Hermes: neither is self-hosted, neither is
+provider-agnostic, and neither gives you execution on your own hardware. Connectors and
+scheduling exist, but they run on Anthropic's or OpenAI's infrastructure. Your memory, session
+history, and agent execution live on their servers, not yours. For many use cases that's fine
+— they are capable and well-supported. For privacy-conscious users, regulated environments, or
+workflows that require persistent server-side execution on controlled hardware, it's a
+disqualifying constraint.
+
+| | Claude.ai | ChatGPT | Hermes |
+|---|---|---|---|
+| Memory across conversations | Yes (auto-generated from history) | Yes (dual-mode: auto + manual) | Yes (deep, automatic) |
+| Scheduled tasks | Yes (Cowork: hourly/daily/weekly) | Yes (since Jan 2025) | Yes (any cron, self-hosted) |
+| Service connectors / messaging | Yes (50+ via Cowork) | Yes (50+ connectors) | Yes (many platforms, direct) |
+| Runs shell commands | Sandboxed (Cowork VM) | Sandboxed | Yes (full shell) |
+| Code execution | Sandboxed | Sandboxed | Yes (full shell) |
+| Reads / writes files | Sandboxed | Sandboxed | Yes (full filesystem) |
+| Web UI | Yes (Anthropic-hosted) | Yes (OpenAI-hosted) | Yes (self-hosted) |
+| Self-hosted | No | No | Yes |
+| Provider-agnostic | No | No | Yes |
+| Open source | No | No | Yes |
+| Self-hosted autonomous execution | No | No | Yes |
+| Memory inspectability | Limited | Limited | Yes (markdown files) |
 
 ---
 
-## The Compounding Advantage
+## The compounding advantage
 
-What matters most about Hermes is that it improves over time. That is the point.
+What distinguishes Hermes from most of the tools above is that it gets meaningfully better at
+your specific workflow over time without manual configuration.
 
 Every time Hermes encounters a new environment, it saves facts to memory. Every time it solves
 a problem a new way, it saves the approach as a skill. Every time you correct it, it updates its
-profile of you. Every session, every scheduled job, every tool call, the agent gets more
-calibrated to you and your workflow.
+profile of you. Every session, every scheduled job, every tool call adds to a body of knowledge
+that is specific to you, stored on your hardware, and available to every future interaction.
 
-A Claude Code session on day one and day one hundred are identical. A Hermes agent on day one
-and day one hundred is smarter about you -- it knows your stack, your conventions, your
-preferences, and the solutions that have worked before.
+A Claude Code session on day one and day one hundred are identical — it starts fresh. A Hermes
+agent on day one and day one hundred knows your stack, your conventions, your preferences, and
+the solutions that have worked before. That's the actual compounding.
 
 ---
 
-## Who Hermes Is For
+## Who Hermes is for
 
-**Solo developers and power users** who don't want to re-explain their stack every session and
-want an AI that actually knows their environment.
+Solo developers and power users who don't want to re-explain their stack every session and want
+an AI that actually knows their environment.
 
-**Teams on a shared server** where multiple people want Claude-quality AI access without each
-paying for a separate subscription or running local tooling.
+Teams on a shared server where multiple people want capable AI access without each paying for
+a separate subscription or running separate local tooling.
 
-**Automation-heavy workflows** where you want an AI running tasks on a schedule, delivering
-results to your phone, without babysitting it.
+Automation-heavy workflows where you want an AI running tasks on a schedule, delivering results
+to your phone, without babysitting it.
 
-**Privacy-conscious users** who want their conversations, memory, and files on their own
-hardware.
+Privacy-conscious users who want their conversations, memory, and files on their own hardware.
 
-**Multi-model users** who want to switch between OpenAI, Anthropic, Google, DeepSeek, and
-others based on cost, capability, or rate limits, without rebuilding their workflow each time.
+Multi-model users who want to switch between OpenAI, Anthropic, Google, DeepSeek, and others
+based on cost, capability, or rate limits, without rebuilding their workflow each time.
 
 ---
 
-## Scope and Limits
+## What Hermes is not
 
-**Hermes lives in the terminal, browser, and messaging apps.** For in-editor autocomplete and
-inline diffs, use Cursor or Windsurf alongside it -- they do that job better.
+Hermes is not the best in-editor autocomplete tool. Cursor and Windsurf do that job better.
+Use one alongside Hermes.
 
-**You run Hermes on your own server.** That means initial setup, but your data stays on your
+It is not zero-setup. You are running a server. That means initial configuration, and it means
+you're responsible for uptime, upgrades, and backups. The tradeoff is data sovereignty and
+control; that only makes sense if you actually want it.
+
+It does not make weaker models magical. Memory and skills help, but the underlying model still
+determines reasoning quality. Hermes with a weak model is a well-organized weak model.
+
+It still needs guardrails, approvals, and observability for high-stakes automations. Autonomous
+execution on a schedule with shell access is powerful and requires judgment about what to
+approve. Terminal commands can require confirmation before running; use that for anything
+consequential.
+
+If you need the absolute lowest-friction path to a one-off answer or a quick edit, a chat
+interface or an in-editor tool is the right call. Hermes is for continuity and autonomy, not
+minimum-friction one-shots.
+
+---
+
+## Scope and limits
+
+Hermes lives in the terminal, browser, and messaging apps. For in-editor autocomplete and inline
+diffs, use Cursor or Windsurf — they do that job better and work well alongside Hermes.
+
+You run Hermes on your own server. That means initial setup, but your data stays on your
 hardware and you control the schedule, the models, and the costs.
 
-**Hermes is an orchestration and memory layer.** It makes whatever model you point it at more
-useful over time. The models do the reasoning; Hermes makes sure that reasoning accumulates into
+Hermes is an orchestration and memory layer. It makes whatever model you point at it more useful
+over time. The models do the reasoning; Hermes makes sure that reasoning accumulates into
 something durable.
 
 ---
 
-## Quick Reference
+## Security and control
 
-| | OpenClaw | Claude Code | Codex CLI | OpenCode | Cursor | Claude.ai | Hermes |
-|---|---|---|---|---|---|---|---|
-| Persistent memory (auto) | Yes | Partial† | Partial | Partial | No | Yes (improving) | **Yes** |
-| Scheduled / background jobs | Yes | Partial‡ | Partial§ | No | No | No | **Yes (self-hosted)** |
-| Messaging app access | Yes (15+ platforms) | Partial (Telegram/Discord preview; Slack native) | No | No | No | No | **Yes (10+ platforms)** |
-| Web UI | Dashboard only | Yes (Anthropic cloud) | No | Yes | No | Yes | **Yes (self-hosted)** |
-| Skills system | Yes (marketplace) | Yes (Hooks + Plugins) | No | No | No | No | **Yes** |
-| Self-improving skills | Partial | No | No | No | No | No | **Yes** |
-| Browser / computer control | Yes (Chrome CDP) | No | No | No | No | No | Via shell |
-| Python / ML ecosystem | No (Node.js) | No | No | No | No | No | **Yes** |
-| In-editor autocomplete | No | No | No | No | Yes | No | No |
-| Orchestrates other agents | No | No | No | No | No | No | **Yes** |
-| Provider-agnostic | Yes | No (Claude only) | Yes | Yes | Partial | No | **Yes** |
-| Self-hosted | Yes | No | Yes | Yes | No | No | **Yes** |
-| Open source | Yes (MIT) | No | Yes | Yes | No | No | **Yes** |
-| Always-on / autonomous | Yes | No | No | No | No | No | **Yes** |
+Memory is stored locally on your server as readable, editable files: user profile, agent memory,
+and skills are all markdown. Session history is in SQLite on your machine. You can inspect,
+edit, or delete any of it directly.
 
-† Claude Code has CLAUDE.md / MEMORY.md project context and rolling auto-memory, but not full automatic cross-session recall  
-‡ Claude Code scheduling: cloud-managed (Anthropic infrastructure) or desktop-app only; no self-hosted cron  
-§ Codex scheduling: desktop app Automations only; CLI has no native scheduling
+If you want external memory providers, eight are supported: Mem0, Honcho, Hindsight, RetainDB,
+ByteRover, Supermemory, Holographic, and others. These are optional and configurable.
+
+Execution runs in configurable backends: local shell, Docker, SSH, Daytona, Singularity, or
+Modal. You choose what execution environment Hermes operates in and what it can reach.
+
+Terminal commands can require confirmation before running. For any automation that touches
+production systems or makes external calls, enable approval controls.
+
+Secrets stay on your hardware. Hermes does not phone home; it calls whatever model APIs you
+configure directly.
+
+Multiple profiles give isolation between users or projects. A shared server can have separate
+profiles with separate memory, separate skills, and separate history.
+
+---
+
+## Quick reference
+
+| | OpenClaw | Claude Code | Codex | OpenCode | Cursor | Copilot | Claude.ai | ChatGPT | Hermes |
+|---|---|---|---|---|---|---|---|---|---|
+| Persistent memory (auto) | Yes | Partial† | Partial | Partial | Yes (per-project) | Yes (repo-scoped‡) | Yes | Yes | Yes |
+| Scheduled / background jobs | Yes | Partial§ | Partial¶ | No | Yes (Automations) | Via Coding Agent | Yes (Cowork) | Yes | Yes (self-hosted) |
+| Messaging / multi-surface | Yes (24+ platforms) | Partial (preview) | No | Community only | Yes (Slack/web/mobile) | Via CLI/fleet | Yes (50+ connectors) | Yes (50+ connectors) | Yes (many platforms) |
+| Web UI | Chat UI + control dashboard | Anthropic-hosted | No | Yes | Yes + mobile | github.com | Yes (Claude Desktop) | Yes | Yes (self-hosted) |
+| Skills system | Yes (ClawHub marketplace) | Yes (Hooks + Plugins) | Partial (Skills) | Community plugins | Yes (marketplace) | No | No | No | Yes (auto-generated) |
+| Self-improving skills | Partial | No | No | No | No | No | No | No | Yes |
+| Browser / computer control | Yes (Chrome CDP) | No | No | No | No | No | No | Yes (CUA) | Via shell |
+| In-editor autocomplete | No | No | Via extension | No | Excellent | Excellent | No | No | No |
+| Orchestrates other agents | No | No | No | No | No | No | No | No | Yes |
+| Provider-agnostic | Yes | No (Claude only) | Yes | Yes | Partial | No | No | No | Yes |
+| Self-hosted | Yes | No | Yes (CLI) | Yes | No | No | No | No | Yes |
+| Self-hosted autonomous execution | Yes | No | No | No | No | No | No | No | Yes |
+| Background/cloud agent mode | Yes | Yes (cloud) | Yes (Codex Cloud) | No | Yes (cloud VMs) | Yes (Coding Agent) | Yes (Cowork VM) | Yes (Agent Mode) | Yes (self-hosted) |
+| Memory inspectability | Limited | Partial | Partial | Partial | Partial | Limited | Limited | Limited | Yes (markdown files) |
+| Open source | Yes (MIT) | No | Yes (Apache 2.0) | Yes | No | No | No | No | Yes |
+| Always-on autonomous execution | Yes | No | No | No | No | No | No | No | Yes |
+
+† Claude Code: CLAUDE.md / MEMORY.md project context plus auto-memory since v2.1.59+; no automatic cross-project accumulation
+‡ Copilot Agentic Memory: public preview Jan 15, 2026; enabled by default Mar 4, 2026; repo-scoped, auto-expires after 28 days
+§ Claude Code scheduling: cloud-managed (Anthropic infrastructure) or desktop-app only; no self-hosted cron
+¶ Codex scheduling: desktop app Automations only; CLI has no native scheduling

README.md

@@ -1,21 +1,24 @@
 # Hermes Web UI
 
-[Hermes Agent](https://hermes-agent.nousresearch.com/) is a sophisticated autonomous agent that lives on your server, accessed via a terminal or messaging apps, remembers what it learns, and gets more capable the longer it runs.
+[Hermes Agent](https://hermes-agent.nousresearch.com/) is a sophisticated autonomous agent that lives on your server, accessed via a terminal or messaging apps, that remembers what it learns and gets more capable the longer it runs.
 
 Hermes WebUI is a lightweight, dark-themed web app interface in your browser for [Hermes Agent](https://hermes-agent.nousresearch.com/).
 Full parity with the CLI experience - everything you can do from a terminal,
 you can do from this UI. No build step, no framework, no bundler. Just Python
 and vanilla JS.
 
-Layout: three-panel Claude-style. Left sidebar for sessions and tools,
-center for chat, right for workspace file browsing.
+Layout: three-panel. Left sidebar for sessions and navigation, center for chat,
+right for workspace file browsing. Model, profile, and workspace controls live in
+the **composer footer** — always visible while composing. A circular context ring
+shows token usage at a glance. All settings and session tools are in the
+**Hermes Control Center** (launcher at the sidebar bottom).
 
-<img alt="Hermes Web UI — three-panel layout" width="1417" height="867" alt="image" src="https://github.com/user-attachments/assets/51adff98-53ee-4800-8508-78b6c34dd3dc" />
+<img width="2448" height="1748" alt="Hermes Web UI — three-panel layout" src="https://github.com/user-attachments/assets/6bf8af4c-209d-441e-8b92-6515d7a0c369" />
 
 <table>
   <tr>
     <td width="50%" align="center">
-      <img alt="Light mode with full profile support" src="https://github.com/user-attachments/assets/9b68142f-d974-4493-a8d1-fd73e622c7fd" />
+      <img width="2940" height="1848" alt="Light mode with full profile support" src="https://github.com/user-attachments/assets/4ef3a59c-7a66-4705-b4e7-cb9148fe4c47" />
       <br /><sub>Light mode with full profile support</sub>
     </td>
     <td width="50%" align="center">
@@ -92,20 +95,31 @@ ecosystem. See [HERMES.md](HERMES.md) for the full side-by-side.
 
 ## Quick start
 
-First, you need to install and configure [Hermes Agent](https://hermes-agent.nousresearch.com/). Once installed:
+Run the repo bootstrap:
 
 ```bash
 git clone https://github.com/nesquena/hermes-webui.git hermes-webui
 cd hermes-webui
+python3 bootstrap.py
+```
+
+Or keep using the shell launcher:
+
+```bash
 ./start.sh
 ```
 
-That is it. The script will:
+The bootstrap will:
 
-1. Locate your Hermes agent checkout automatically.
-2. Find (or create) a Python environment with the required dependencies.
-3. Start the server.
-4. Print the URL (and SSH tunnel command if you are on a remote machine).
+1. Detect Hermes Agent and, if missing, attempt the official installer (`curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash`).
+2. Find or create a Python environment with the WebUI dependencies.
+3. Start the web server and wait for `/health`.
+4. Open the browser unless you pass `--no-browser`.
+5. Drop you into a first-run onboarding wizard inside the WebUI.
+
+> Native Windows is not supported for this bootstrap yet. Use Linux, macOS, or WSL2.
+
+If provider setup is still incomplete after install, the onboarding wizard will point you to finish it with `hermes model` instead of trying to replicate the full CLI setup in-browser.
 
 ---
 
@@ -113,14 +127,23 @@ That is it. The script will:
 
 **Pre-built images** (amd64 + arm64) are published to GHCR on every release:
 
+Make sure the `HERMES_WEBUI_STATE_DIR` (by default `~/.hermes/webui-mvp`, as detailed in the `.env.example` file) folder exist with the UID/GID of the owner of the `.hermes` folder. 
+The container will also mount your configured "workspace" (also from the example .env.example) as `/workspace`. adapt the location as needed.
+
+
 ```bash
 docker pull ghcr.io/nesquena/hermes-webui:latest
-docker run -d -p 8787:8787 -v ~/.hermes:/root/.hermes ghcr.io/nesquena/hermes-webui:latest
+docker run -d \
+-e WANTED_UID=`id -u` -e WANTED_GID=`id -g` \
+-v ~/.hermes:/home/hermeswebui/.hermes -e HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui-mvp \
+-v ~/workspace:/workspace \
+-p 8787:8787 ghcr.io/nesquena/hermes-webui:latest
 ```
 
 Or run with Docker Compose (recommended):
 
 ```bash
+# Check the docker-compose.yml and make sure to adapt as needed, at minimum WANTED_UID/WANTED_GID
 docker compose up -d
 ```
 
@@ -128,7 +151,11 @@ Or build locally:
 
 ```bash
 docker build -t hermes-webui .
-docker run -d -p 8787:8787 -v ~/.hermes:/root/.hermes hermes-webui
+docker run -d \
+-e WANTED_UID=`id -u` -e WANTED_GID=`id -g` \
+-v ~/.hermes:/home/hermeswebui/.hermes -e HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui-mvp \
+-v ~/workspace:/workspace \
+-p 8787:8787 hermes-webui
 ```
 
 Open http://localhost:8787 in your browser.
@@ -136,15 +163,119 @@ Open http://localhost:8787 in your browser.
 To enable password protection:
 
 ```bash
-docker run -d -p 8787:8787 -e HERMES_WEBUI_PASSWORD=your-secret -v ~/.hermes:/root/.hermes ghcr.io/nesquena/hermes-webui:latest
+docker run -d \
+-e WANTED_UID=`id -u` -e WANTED_GID=`id -g` \
+-v ~/.hermes:/home/hermeswebui/.hermes -e HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui-mvp \
+-v ~/workspace:/workspace \
+-p 8787:8787 -e HERMES_WEBUI_PASSWORD=your-secret ghcr.io/nesquena/hermes-webui:latest
 ```
 
-Session data persists in a named volume (`hermes-data`) across restarts.
-
 > **Note:** By default, Docker Compose binds to `127.0.0.1` (localhost only).
 > To expose on a network, change the port to `"8787:8787"` in `docker-compose.yml`
 > and set `HERMES_WEBUI_PASSWORD` to enable authentication.
 
+### Two-container setup (Agent + WebUI)
+
+If you run the Hermes Agent in its own Docker container and want the WebUI
+in a separate container:
+
+```bash
+docker compose -f docker-compose.two-container.yml up -d
+```
+
+This starts both containers with shared volumes:
+
+- **`hermes-home`** — shared `~/.hermes` for config, sessions, skills, memory
+- **`hermes-agent-src`** — the agent's source code, mounted into the WebUI
+  container so it can install the agent's Python dependencies at startup
+
+> **Volume type:** The compose files use named Docker volumes by default.
+> If you prefer bind mounts to an existing directory (e.g. for sharing state
+> with an agent container you already run), both containers must mount the
+> same host path — the agent writes to `/root/.hermes`, the WebUI reads from
+> `/home/hermeswebui/.hermes`. See `docker-compose.two-container.yml` for
+> a bind-mount example.
+
+The WebUI's init script automatically installs hermes-agent and all its
+dependencies (openai, anthropic, etc.) into its own Python environment on
+first boot. Subsequent restarts reuse the installed packages.
+
+> **How it works:** The WebUI imports hermes-agent's Python modules directly
+> (not via HTTP). The shared volume makes the agent source available, and
+> the init script runs `uv pip install` to set up the dependencies. Both
+> containers share the same `~/.hermes` directory for config and state.
+
+See `docker-compose.two-container.yml` for the full configuration.
+
+### Running alongside hermes-dashboard (three-container setup)
+
+To run the Hermes Agent, Hermes Dashboard, and the WebUI together on a
+shared volume, use the three-container Compose file:
+
+```bash
+docker compose -f docker-compose.three-container.yml up -d
+```
+
+This brings up:
+- **`hermes-agent`** — gateway API on port 8642
+- **`hermes-dashboard`** — monitoring UI on port 9119
+- **`hermes-webui`** — browser chat interface on port 8787
+
+All three services share the same `hermes-home` named volume so config,
+sessions, skills, and memory are consistent across all surfaces.
+
+#### Why UIDs must match
+
+The `hermes-home` volume is a bind-mount in practice — all three containers
+write to the same filesystem tree under `~/.hermes`. If the containers run
+as different UIDs, whichever container creates a file first becomes its
+owner, and the others hit `PermissionError` on subsequent writes.
+
+The fix is to make all containers run as **your host user's UID and GID**.
+
+#### Variable name asymmetry
+
+> ⚠️ **The two image families use different environment variable names** for
+> the UID/GID setting:
+>
+> | Image | Variable |
+> |---|---|
+> | `nousresearch/hermes-agent` (agent + dashboard) | `HERMES_UID` / `HERMES_GID` |
+> | `ghcr.io/nesquena/hermes-webui` | `WANTED_UID` / `WANTED_GID` |
+>
+> You must set **both pairs** when using a `.env` file.
+
+#### Recommended setup
+
+For a standard Linux user (UID ≥ 1000):
+
+```bash
+# Create a .env file with your host UID/GID
+echo "UID=$(id -u)" >> .env
+echo "GID=$(id -g)" >> .env
+# hermes-agent / hermes-dashboard
+echo "HERMES_UID=$(id -u)" >> .env
+echo "HERMES_GID=$(id -g)" >> .env
+```
+
+For NAS/Unraid deployments where a fixed service account is preferred, use
+`10000:10000` (or your NAS service UID) instead of `$(id -u)`.
+
+If you get `PermissionError` on an **existing** `~/.hermes` directory, run
+the one-time ownership fix:
+
+```bash
+chown -R $(id -u):$(id -g) ~/.hermes
+```
+
+#### Volume mount mode
+
+The dashboard container needs **read-write** access to the shared volume
+(it writes session logs and dashboard state). Do **not** add `:ro` to the
+`hermes-home` volume in `hermes-dashboard`'s `volumes:` entry.
+
+See `docker-compose.three-container.yml` for the full reference configuration.
+
 ---
 
 ## What start.sh discovers automatically
@@ -167,6 +298,7 @@ If discovery finds everything, nothing else is required.
 export HERMES_WEBUI_AGENT_DIR=/path/to/hermes-agent
 export HERMES_WEBUI_PYTHON=/path/to/python
 export HERMES_WEBUI_PORT=9000
+export HERMES_WEBUI_AUTO_INSTALL=1  # enable auto-install of agent deps (disabled by default)
 ./start.sh
 ```
 
@@ -222,8 +354,8 @@ WireGuard. Install it on your server and your phone, and they join the same
 private network -- no port forwarding, no SSH tunnels, no public exposure.
 
 The Hermes Web UI is fully responsive with a mobile-optimized layout
-(hamburger sidebar, bottom navigation bar, touch-friendly controls), so it
-works well as a daily-driver agent interface from your phone.
+(hamburger sidebar, sidebar top tabs in the drawer, touch-friendly controls),
+so it works well as a daily-driver agent interface from your phone.
 
 **Setup:**
 
@@ -284,8 +416,8 @@ Or using the agent venv explicitly:
 ```
 
 Tests run against an isolated server on port 8788 with a separate state directory.
-Production data and real cron jobs are never touched. Current count: **424 tests**
-across 22 test files.
+Production data and real cron jobs are never touched. Current count: **1898 tests**
+across 53 test files.
 
 ---
 
@@ -297,7 +429,7 @@ across 22 test files.
 - Send a message while one is processing -- it queues automatically
 - Edit any past user message inline and regenerate from that point
 - Retry the last assistant response with one click
-- Cancel a running task from the activity bar
+- Cancel a running task directly from the composer footer (Stop button next to Send)
 - Tool call cards inline -- each shows the tool name, args, and result snippet; expand/collapse all toggle for multi-tool turns
 - Subagent delegation cards -- child agent activity shown with distinct icon and indented border
 - Mermaid diagram rendering inline (flowcharts, sequence diagrams, gantt charts)
@@ -314,6 +446,7 @@ across 22 test files.
 
 ### Sessions
 - Create, rename, duplicate, delete, search by title and message content
+- Session actions via `⋯` dropdown per session — pin, move to project, archive, duplicate, delete
 - Pin/star sessions to the top of the sidebar (gold indicator)
 - Archive sessions (hide without deleting, toggle to show)
 - Session projects -- named groups with colors for organizing sessions
@@ -345,10 +478,11 @@ across 22 test files.
 - Hidden when browser doesn't support Web Speech API (Chrome, Edge, Safari)
 
 ### Profiles
-- Profile picker in the topbar -- purple chip with dropdown showing all profiles
+- Profile chip in the **composer footer** -- dropdown showing all profiles with gateway status and model info
 - Gateway status dots (green = running), model info, skill count per profile
 - Profiles management panel -- create, switch, and delete profiles from the sidebar
 - Clone config from active profile on create
+- Optional custom endpoint fields on create -- Base URL and API key written into the profile's `config.yaml` at creation time, so Ollama, LMStudio, and other local endpoints can be configured without editing files manually
 - Seamless switching -- no server restart; reloads config, skills, memory, cron, models
 - Per-session profile tracking (records which profile was active at creation)
 
@@ -362,23 +496,24 @@ across 22 test files.
 - CDN resources pinned with SRI integrity hashes
 
 ### Themes
-- 6 built-in themes: Dark (default), Light, Slate, Solarized Dark, Monokai, Nord
+- 7 built-in themes: Dark (default), Light, Slate, Solarized Dark, Monokai, Nord, OLED
 - Switch via Settings panel dropdown (instant live preview) or `/theme` command
 - Persists across reloads (server-side in settings.json + localStorage for flicker-free loading)
 - Custom themes: define a `:root[data-theme="name"]` CSS block and it works — see [THEMES.md](THEMES.md)
 
 ### Settings and configuration
-- Settings panel (gear icon) -- default model, default workspace, send key, theme
+- **Hermes Control Center** (sidebar launcher button) -- Conversation tab (export/import/clear), Preferences tab (model, send key, theme, language, all toggles), System tab (version, password)
 - Send key: Enter (default) or Ctrl/Cmd+Enter
 - Show/hide CLI sessions toggle (enabled by default)
 - Token usage display toggle (off by default, also via `/usage` command)
+- Control Center always opens on the Conversation tab; resets on close
 - Unsaved changes guard -- discard/save prompt when closing with unpersisted changes
 - Cron completion alerts -- toast notifications and unread badge on Tasks tab
 - Background agent error alerts -- banner when a non-active session encounters an error
 
 ### Slash commands
 - Type `/` in the composer for autocomplete dropdown
-- Built-in: `/help`, `/clear`, `/model <name>`, `/workspace <name>`, `/new`, `/usage`, `/theme`, `/compact`
+- Built-in: `/help`, `/clear`, `/compress [focus topic]`, `/compact` (alias), `/model <name>`, `/workspace <name>`, `/new`, `/usage`, `/theme`
 - Arrow keys navigate, Tab/Enter select, Escape closes
 - Unrecognized commands pass through to the agent
 
@@ -393,10 +528,10 @@ across 22 test files.
 
 ### Mobile responsive
 - Hamburger sidebar -- slide-in overlay on mobile (<640px)
-- Bottom navigation bar -- 5-tab iOS-style fixed bar
+- Sidebar top tabs stay available on mobile; no fixed bottom nav stealing chat height
 - Files slide-over panel from right edge
 - Touch targets minimum 44px on all interactive elements
-- Composer positioned above bottom nav
+- Full-height chat/composer on phones without bottom-nav spacing
 - Desktop layout completely unchanged
 
 ---
@@ -404,31 +539,33 @@ across 22 test files.
 ## Architecture
 
 ```
-server.py               HTTP routing shell + auth middleware (~83 lines)
+server.py               HTTP routing shell + auth middleware (~154 lines)
 api/
-  auth.py               Optional password authentication, signed cookies (~149 lines)
-  config.py             Discovery, globals, model detection, reloadable config (~726 lines)
-  helpers.py            HTTP helpers, security headers (~71 lines)
-  models.py             Session model + CRUD + CLI bridge (~338 lines)
-  profiles.py           Profile state management, hermes_cli wrapper (~366 lines)
-  routes.py             All GET + POST route handlers (~1314 lines)
-  streaming.py          SSE engine, run_agent, cancel support (~332 lines)
-  upload.py             Multipart parser, file upload handler (~78 lines)
+  auth.py               Optional password authentication, signed cookies (~201 lines)
+  config.py             Discovery, globals, model detection, reloadable config (~1110 lines)
+  helpers.py            HTTP helpers, security headers (~175 lines)
+  models.py             Session model + CRUD + CLI bridge (~377 lines)
+  onboarding.py         First-run onboarding wizard, OAuth provider support (~507 lines)
+  profiles.py           Profile state management, hermes_cli wrapper (~411 lines)
+  routes.py             All GET + POST route handlers (~2250 lines)
+  state_sync.py         /insights sync — message_count to state.db (~113 lines)
+  streaming.py          SSE engine, run_agent, cancel support (~660 lines)
+  updates.py            Self-update check and release notes (~257 lines)
+  upload.py             Multipart parser, file upload handler (~82 lines)
   workspace.py          File ops, workspace helpers, git detection (~288 lines)
 static/
-  index.html            HTML template (~388 lines)
-  style.css             All CSS incl. mobile responsive (~726 lines)
-  ui.js                 DOM helpers, renderMd, tool cards, context indicator (~1063 lines)
-  workspace.js          File preview, file ops, git badge (~247 lines)
-  sessions.js           Session CRUD, collapsible groups, search (~589 lines)
-  messages.js           send(), SSE handlers, rAF throttle (~352 lines)
-  panels.js             Cron, skills, memory, profiles, settings (~1146 lines)
-  commands.js           Slash command autocomplete (~170 lines)
-  boot.js               Mobile nav, voice input, boot IIFE (~338 lines)
+  index.html            HTML template (~600 lines)
+  style.css             All CSS incl. mobile responsive, themes (~1050 lines)
+  ui.js                 DOM helpers, renderMd, tool cards, context indicator (~1740 lines)
+  workspace.js          File preview, file ops, git badge (~286 lines)
+  sessions.js           Session CRUD, collapsible groups, search, reload recovery (~800 lines)
+  messages.js           send(), SSE handlers, live streaming, session recovery (~655 lines)
+  panels.js             Cron, skills, memory, profiles, settings (~1438 lines)
+  commands.js           Slash command autocomplete (~267 lines)
+  boot.js               Mobile nav, voice input, boot IIFE (~524 lines)
 tests/
   conftest.py           Isolated test server (port 8788)
-  test_sprint{1-23}.py  22 test files, 426 test functions
-  test_regressions.py   Permanent regression gate (23 tests)
+  61 test files          961 test functions
 Dockerfile              python:3.12-slim container image
 docker-compose.yml      Compose with named volume and optional auth
 .github/workflows/      CI: multi-arch Docker build + GitHub Release on tag
@@ -449,6 +586,119 @@ State lives outside the repo at `~/.hermes/webui-mvp/` by default
 - `SPRINTS.md` -- forward sprint plan with CLI + Claude parity targets
 - `THEMES.md` -- theme system documentation, custom theme guide
 
+## Contributors
+
+Hermes WebUI is built with help from the open-source community. Every PR — whether merged directly or incorporated via rebase — shapes the project, and we're grateful to everyone who has taken the time to contribute.
+
+### Major contributions
+
+**[@aronprins](https://github.com/aronprins)** — v0.50.0 UI overhaul (PR #242)
+The biggest single contribution to the project: a complete UI redesign that moved model/profile/workspace controls into the composer footer, replaced the gear-icon settings panel with the Hermes Control Center (tabbed modal), removed the activity bar in favor of inline composer status, redesigned the session list with a `⋯` action dropdown, and added the workspace panel state machine. 26 commits, thoroughly designed and iterated through multiple review rounds.
+
+**[@iRonin](https://github.com/iRonin)** — Security hardening sprint (PRs #196–#204)
+Six consecutive security and reliability PRs: session memory leak fix (expired token pruning), Content-Security-Policy + Permissions-Policy headers, 30-second slow-client connection timeout, optional HTTPS/TLS support via environment variables, upstream branch tracking fix for self-update, and CLI session support in the file browser API. This is the kind of focused, high-quality security work that makes a self-hosted tool trustworthy.
+
+**[@DavidSchuchert](https://github.com/DavidSchuchert)** — German translation (PR #190)
+Complete German locale (`de`) covering all UI strings, settings labels, commands, and system messages — and in doing so, stress-tested the i18n system and exposed several elements that weren't yet translatable, which got fixed as part of the same PR.
+
+**[@Jordan-SkyLF](https://github.com/Jordan-SkyLF)** — Live streaming, session recovery, workspace fallback (PRs #366, #367)
+Three interlocking improvements: workspace fallback resolution so the server recovers gracefully when the configured workspace is deleted or unavailable; live reasoning cards that upgrade the generic thinking spinner to a real-time reasoning display as the model thinks; and durable session state recovery via `localStorage` so in-flight tool cards, partial assistant output, and the live SSE stream all survive a full page reload or session switch.
+
+### Feature contributions
+
+**[@gabogabucho](https://github.com/gabogabucho)** — Spanish locale + onboarding wizard (PRs #275, #285)
+Full Spanish (`es`) locale covering all 175 UI strings, plus the one-shot bootstrap onboarding wizard that guides new users through provider setup on first launch — the feature most responsible for new users actually getting started.
+
+**[@bergeouss](https://github.com/bergeouss)** — Real-time gateway session sync (PR #274)
+Bridged the gateway session database (Telegram, Discord, Slack, etc.) into the WebUI sidebar with live SSE polling. Gateway sessions now appear alongside WebUI sessions in real time, without any changes to hermes-agent.
+
+**[@ccqqlo](https://github.com/ccqqlo)** — Terminal approval UX + custom model discovery + mobile close button (PRs #224, #225, #238, #333)
+A run of focused quality-of-life improvements: terminal tool approval prompts that stay visible long enough to actually be read, restored custom model API key discovery, and the redundant mobile close button fix that had been confusing users on narrow screens.
+
+**[@kevin-ho](https://github.com/kevin-ho)** — OLED theme (PR #168)
+Added the 7th built-in theme: pure black backgrounds with warm accents tuned to reduce burn-in risk. Small diff, big impact for anyone on an OLED display.
+
+**[@Bobby9228](https://github.com/Bobby9228)** — Mobile Profiles button + Android Chrome fixes (PRs #253, #263, #265)
+Added the Profiles entry to the mobile navigation flow, making profile switching reachable on phones, plus a set of Android Chrome-specific fixes for the profile dropdown.
+
+**[@franksong2702](https://github.com/franksong2702)** — Session title guard + breadcrumb nav (PRs #301, #302)
+Two clean bug fixes / features: the session title guard that stops `title_from()` from overwriting user-renamed sessions after every turn, and clickable breadcrumb navigation in the workspace file preview panel.
+
+**[@betamod](https://github.com/betamod)** — Security hardening (PR #171)
+A comprehensive security audit PR covering CSRF protection, SSRF guards, XSS escaping improvements, and the env race condition between concurrent agent sessions — foundational security work that shipped in v0.39.0.
+
+**[@TaraTheStar](https://github.com/TaraTheStar)** — Bot name + thinking blocks + login refactor (PRs #132, #176, #181)
+Made the assistant display name configurable throughout the UI, added thinking/reasoning block display in chat, and refactored the login page to use template variables instead of inline string replacement.
+
+**[@thadreber-web](https://github.com/thadreber-web)** — CLI session bridge (PR #56)
+The original CLI session bridge: reads CLI sessions from the agent's SQLite state store and surfaces them in the WebUI sidebar. This was the first bridge between the CLI and WebUI session worlds.
+
+**[@deboste](https://github.com/deboste)** — Reverse proxy auth + mobile responsive layout + model routing (PRs #3, #4, #5)
+Three of the very first community PRs: fixed EventSource/fetch to use the URL origin for reverse proxy setups, corrected model provider routing from config, and added mobile responsive layout with dvh viewport fix. Early foundation work.
+
+### Bug fix and security contributions
+
+**[@Hinotoi-agent](https://github.com/Hinotoi-agent)** — Profile .env secret isolation (PR #351)
+Fixed API key leakage between profiles on switch — switching from a profile with `OPENAI_API_KEY` to one without it left the key in the process environment for the duration of the session, effectively leaking credentials. A subtle and important security fix.
+
+**[@lawrencel1ng](https://github.com/lawrencel1ng)** — Bandit security fixes B310/B324/B110 + QuietHTTPServer (PR #354)
+Systematic bandit security scan fixes: URL scheme validation before `urlopen`, MD5 `usedforsecurity=False`, and 40+ bare `except: pass` blocks replaced with proper logging — plus `QuietHTTPServer` to stop client-disconnect log spam from SSE streams.
+
+**[@lx3133584](https://github.com/lx3133584)** — CSRF fix for reverse proxy on non-standard ports (PR #360)
+Fixed CSRF rejection for deployments behind Nginx Proxy Manager or similar on non-standard ports — a real-world blocker for anyone hosting on a port other than 80/443.
+
+**[@DelightRun](https://github.com/DelightRun)** — session_search fix for WebUI sessions (PR #356)
+The `session_search` tool silently returned "Session database not available" in every WebUI session. Tracked down the missing `SessionDB` injection in the streaming path and fixed it.
+
+**[@shaoxianbilly](https://github.com/shaoxianbilly)** — Unicode filename downloads (PR #378)
+Fixed `UnicodeEncodeError` crashes when downloading workspace files with Chinese, Japanese, or other non-ASCII names. Implemented proper `Content-Disposition` header with RFC 5987 `filename*=UTF-8''...` encoding.
+
+**[@huangzt](https://github.com/huangzt)** — Cancel interrupts agent (PR #244)
+Made the Cancel button actually interrupt the running agent and clean up UI state, rather than just hiding the button while the agent kept running.
+
+**[@tgaalman](https://github.com/tgaalman)** — Thinking card fix (PR #169)
+Fixed top-level reasoning fields being missed in the thinking card display — an edge case in how Claude's extended thinking blocks surface in the API response.
+
+**[@smurmann](https://github.com/smurmann)** — Custom provider routing fix (PR #189)
+Fixed model routing for slash-prefixed custom provider models, which were being misrouted in the model selector. A precise fix for a real edge case in multi-provider setups.
+
+**[@jeffscottward](https://github.com/jeffscottward)** — Claude Haiku model ID fix (PR #145)
+Caught and corrected the Claude Haiku model ID (`3-5` → `4-5`) immediately after the Anthropic release — the kind of quick community catch that keeps the model dropdown accurate.
+
+**[@kcclaw001](https://github.com/kcclaw001)** — Credential redaction in API responses (PR #243)
+Added credential redaction to all API response paths so API keys, tokens, and other secrets in session data or error messages are masked before reaching the browser.
+
+**[@mbac](https://github.com/mbac)** — Phantom "Custom" provider group fix (PR #191)
+Removed the phantom "Custom" optgroup that appeared in the model dropdown even when no custom provider was configured — a small but consistently confusing UI noise issue.
+
+**[@andrewy-wizard](https://github.com/andrewy-wizard)** — Chinese localization (PR #177)
+Added Simplified Chinese (`zh`) locale to the WebUI. One of the first non-English locales and the most-used non-English locale in the codebase.
+
+**[@mmartial](https://github.com/mmartial)** — Docker UID/GID matching (PR #237)
+Added Docker support for running as an arbitrary UID/GID matching the host user, eliminating permission issues with bind-mounted volumes — essential for Docker deployments where the host user isn't UID 1000.
+
+**[@vCillusion](https://github.com/vCillusion)** — pip package resolution fix (PR #76)
+Fixed agent dependency resolution to prefer packages from the venv's site-packages over the agent directory itself, preventing shadowing bugs when developing locally.
+
+**[@carlytwozero](https://github.com/carlytwozero)** — API key pass-through for non-Anthropic providers (PR #78)
+Fixed `api_key` not being passed to `AIAgent` for non-Anthropic `/anthropic` providers — a quiet regression that silently broke any non-default provider.
+
+**[@mangodxd](https://github.com/mangodxd)** — Type hints cleanup (PR #115)
+Added missing type hints across 10 files and corrected 9 inaccurate existing ones — the kind of maintenance work that makes the codebase easier to reason about.
+
+**[@Argonaut790](https://github.com/Argonaut790)** — HTML entity decode + Traditional Chinese locale (PR #239)
+Fixed double-escaping of HTML entities in `renderMd()` — LLM output containing `&lt;code&gt;` was being escaped a second time, rendering as literal text instead of the intended markdown. The same PR also completed the Simplified Chinese translation (40+ missing keys) and added a full Traditional Chinese (`zh-Hant`) locale.
+
+**[@indigokarasu](https://github.com/indigokarasu)** — Visual redesign proposal: icon rail + design token system + 7 themes (PR #213)
+A CSS-only redesign of the full UI — proper design tokens (`--bg-primary`, `--text-info`, spacing scale), an icon rail sidebar replacing the emoji tab strip, consistent form cards, breadcrumb nav, and 7 built-in themes as custom properties. The PR didn't merge as-is but directly shaped the design language and theme architecture that shipped in v0.50.0.
+
+**[@zenc-cp](https://github.com/zenc-cp)** — Anti-hallucination guard for ReAct loop (PR #133)
+Added a streaming token buffer and post-run message scrub to `streaming.py` to detect and strip fake tool execution JSON that weaker models write inline instead of calling tools properly. A three-layer approach: ephemeral anti-hallucination prompt, live token filtering, and session history cleanup. The pattern influenced later streaming.py improvements.
+
+---
+
+Want to contribute? See [ARCHITECTURE.md](ARCHITECTURE.md) for the codebase layout and [TESTING.md](TESTING.md) for how to run the test suite. The best contributions are focused, well-tested, and solve a real problem — exactly what every person on this list did.
+
 ## Repo
 
 ```

ROADMAP.md

@@ -3,8 +3,8 @@
 > Goal: Full 1:1 parity with the Hermes CLI experience via a clean dark web UI.
 > Everything you can do from the CLI terminal, you can do from this UI.
 >
-> Last updated: v0.35 (April 5, 2026)
-> Tests: 433 total (433 passing, 0 failures)
+> Last updated: v0.50.185 (April 24, 2026) — 2107 tests collected
+> Tests: 2107 collected (`pytest tests/ --collect-only -q`)
 > Source: <repo>/
 
 ---
@@ -32,14 +32,24 @@
 | Sprint 13 | Alerts + polish | Cron completion alerts (polling + badge), background error banner, session duplicate, browser tab title | 221 |
 | Sprint 14 | Visual polish + workspace ops | Mermaid diagrams, message timestamps, file rename, folder create, session tags, session archive | 233 |
 | Sprint 15 | Session projects + code copy | Session projects/folders, code block copy button, tool card expand/collapse toggle | 237 |
-| Sprint 16 | Session sidebar visual polish | SVG action icons, overlay hover actions, pin indicator, project border, safe HTML rendering | 289 |
+| Sprint 16 | Session sidebar visual polish | SVG action icons, session action dropdown, pin indicator, project border, safe HTML rendering | 289 |
 | Sprint 17 | Workspace polish + slash commands + settings | Breadcrumb navigation, slash command autocomplete, send key setting (#26) | 318 |
 | Sprint 18 | Thinking display + workspace tree | File preview auto-close, thinking/reasoning cards, expandable directory tree (#22) | 318 |
 | Sprint 19 | Auth + security hardening | Password auth (off by default), login page, security headers, 20MB body limit (#23) | 328 |
 | Sprint 20 | Voice input + send button | Voice input (Web Speech API), send button icon-circle with pop-in animation | 415 |
-| Sprint 21 | Mobile responsive + Docker | Hamburger sidebar, bottom nav, files slide-over, Docker support (#21, #7) | 415 |
+| Sprint 21 | Mobile responsive + Docker | Hamburger sidebar, mobile nav, files slide-over, Docker support (#21, #7) | 415 |
 | Sprint 22 | Multi-profile support | Profile picker, management panel, seamless switching, per-session tracking (#28) | 415 |
 | Sprint 23 | Agentic transparency | Token/cost display, subagent cards, skill picker in cron, skill linked files, workspace tree persistence, timestamp fixes | 424 |
+| v0.44.0 patch | Fix batch: approval card, login CSP, update diagnostics, Lucide icons | PRs #221 #225 #226 #227 #228 | 579 |
+| v0.45.0 | Custom endpoint in new profile form | Base URL + API key fields; server-side URL validation; config.yaml merge; 9 new tests (PR #233, fixes #170) | 604 |
+| v0.46.0 | Security, Docker UID/GID, model discovery, i18n, cancel fix | Credential redaction in API responses (PR #243); Docker UID/GID matching (PR #237); custom model API key discovery (PR #238); HTML entity decode + zh/zh-Hant i18n (PR #239); cancel interrupts agent (PR #244); +20 tests | 624 |
+| v0.47.0 | Dialogs, session menu, skills command, mobile fixes, mobile QA | Shared app dialogs (#251); session ⋯ menu (#252); mobile QA suite (#254); custom provider slash routing fix (#255); Android Chrome mobile fixes (#256); /skills command (#257); +21 tests | 645 |
+| v0.47.1 | Spanish locale | Full Spanish (es) locale, 175 keys, key-parity tests (#275 @gabogabucho); +3 tests | 648 |
+| v0.48.0 | Gateway session sync | Real-time Telegram/Discord/Slack sessions in sidebar via SSE + DB polling (#274 @bergeouss); +10 tests | 658 |
+| v0.48.1 | Table inline formatting | `inlineMd()` in table cells — **bold**, *italic*, `code`, links render correctly (PR #278); 0 new tests | 658 |
+| v0.48.2 | Provider mismatch warning | Toast warning + auth_mismatch error type for provider/model mismatches (#283, fixes #266); +21 tests | 679 |
+| v0.49.1 | Docker docs + mobile Profiles button | Two-container Docker compose (#291/#288); Profiles added to the mobile navigation flow with correct panel wiring and SVG sizing (#297/#265 @gabogabucho); +3 tests | 700 |
+| v0.49.0 | First-run onboarding wizard + self-update hardening | One-shot bootstrap + guided setup wizard; provider config persisted to config.yaml + .env; OpenRouter/Anthropic/OpenAI/Custom; wizard hidden after completion (#285); self-update stderr/split-ref/conflict fixes (#287); skip flaky redaction test (#289); +18 tests | 697 |
 | v0.32 | Auto-compaction handling | Compression detection, /compact command, real context window indicator | 424 |
 | v0.33 | /insights sync | Opt-in state.db sync so `hermes /insights` includes WebUI sessions | 424 |
 | v0.34 | Sprint 26 — Pluggable themes | Dark, Light, Slate, Solarized, Monokai, Nord; settings unsaved-changes guard; /theme command | 433 |
@@ -47,6 +57,35 @@
 | v0.34.2 | Theme text colors | 5 new per-theme typography variables (--strong, --em, --code-text, --code-inline-bg, --pre-text) | 433 |
 | v0.34.3 | Light theme final polish | 46 light-scoped selector overrides for sidebar, roles, chips, interactive elements | 433 |
 | v0.35 | Security hardening | Env race fix, random signing key, upload path traversal, PBKDF2 password hash | 433 |
+| v0.36–v0.37 | Model routing, personality config, tool card reload, duplicate model fixes | Model routing by provider prefix, personality via config.yaml, tool cards reload on page refresh | 466 |
+| v0.38.0–v0.38.6 | Model selector, custom endpoints, OLED theme, reasoning display, insights sync | Custom endpoint URL fix, OLED theme, top-level reasoning field fix, message_count sync to state.db | 466 |
+| v0.39.0 | Security hardening (Sprint 29) | CSRF, PBKDF2, rate limiting, session ID validation, SSRF, ENV_LOCK, XSS, HMAC, skills traversal, secure cookie, error sanitization, startup warning | 499 |
+| v0.40–v0.44.2 | Approval card + Lucide icons + sprint auth | Approval prompt surfaced in UI, emoji icons → Lucide SVG, login CSP inline fix, update diagnostics | 579 |
+| v0.45–v0.46 | Custom endpoints + security + i18n + cancel | Custom endpoint Base URL + API key on profile create, credential redaction (PR #243), Docker UID/GID (PR #237), HTML entity decode + zh/zh-Hant i18n, cancel interrupts agent | 624 |
+| v0.47–v0.47.1 | Dialogs + session menu + skills + mobile QA + Spanish | Shared app dialogs, session ⋯ menu, /skills command, mobile QA suite, Android Chrome fixes, Spanish locale (@gabogabucho) | 648 |
+| v0.48–v0.48.2 | Gateway session sync + table formatting + provider warnings | Real-time Telegram/Discord/Slack sessions in sidebar (@bergeouss), inlineMd() in table cells, provider/model mismatch toast | 679 |
+| v0.49–v0.49.1 | Onboarding wizard + Docker two-container | One-shot bootstrap + guided setup wizard, OpenRouter/Anthropic/OpenAI/Custom provider config, two-container Docker compose, mobile Profiles button | 700 |
+| v0.50.0 | v0.50.0 UI overhaul (Sprint 34) | Composer-centric controls, Hermes Control Center modal, workspace panel state machine, collapsible date groups, rAF streaming throttle, context ring indicator (@aronprins) | 742 |
+| v0.50.5–v0.50.10 | Think-tag edge cases + onboarding hardening + mobile fixes | MiniMax M2.5 leading-whitespace think-tag fix, skip-onboarding env var, OAuth provider path, Docker bridge networks fix, model dropdown dedup, title auto-generation fix, mobile close button | 802 |
+| v0.50.11–v0.50.12 | Chat table styles + URL autolink + profile env isolation | .msg-body table borders, plain URL auto-linking, profile .env secret isolation on switch (prevents API key leakage across profiles, @Hinotoi-agent) | 815 |
+| v0.50.13–v0.50.15 | session_search + security sweep + KaTeX math | SessionDB injection for session_search in WebUI (@DelightRun), bandit B310/B324/B110 + QuietHTTPServer (@lawrencel1ng), KaTeX math rendering with fence-before-math fix | 871 |
+| v0.50.16–v0.50.17 | CSRF reverse proxy + Docker uv pre-install | Scheme-aware CSRF port normalization for non-standard ports (@lx3133584), Docker uv pre-installed at build time as root (fixes air-gapped startup, @mmartial-pattern) | 900 |
+| v0.50.18–v0.50.19 | Workspace fallback + Unicode filenames | Cascading workspace path recovery (@Jordan-SkyLF), Unicode Content-Disposition headers with RFC 5987 filename* (@shaoxianbilly), silent auth error surfacing, stale model cleanup | 924 |
+| v0.50.20–v0.50.21 | Silent errors + live model fetching + durable streaming recovery | apperror on empty agent response, /api/models/live endpoint with SSRF guard, live reasoning cards, tool_complete SSE events, SESSION_QUEUES, localStorage reload recovery (@Jordan-SkyLF) | 961 |
+| v0.50.22–v0.50.36-local.1 | Upstream sync + minimal local patch retention | Synced to upstream `v0.50.36`; retained first-password session continuity in Settings/onboarding; removed local Assistant Reply Language enhancement; added legacy settings cleanup regression coverage | 1059 |
+| v0.50.37–v0.50.40 | Sprint 40 — rendering fixes + KaTeX CSP + MEDIA images | Think-tag edge cases, renderMd link double-linking fix, MEDIA: inline image rendering, KaTeX CSP font-src fix | 1117 |
+| v0.50.41–v0.50.43 | Sprint 41/42 — context ring, session polish, renderMd hardening | Context indicator live usage, session display fixes, renderMd bold+code stash, outer link pass ordering, _ob_stash, autolink double-link fixes (@multiple contributors) | 1150 |
+| v0.50.44 | Renderer formatting bug fixes (#486, #487) | CSS: inline code sizing in table cells; JS: markdown image syntax ![alt](url) → <img> in renderMd + inlineMd; _img_stash for autolink protection | 1195 |
+| v0.50.45–v0.50.100 | Upstream sync + contributor sprint | Sidebar declutter, SKIP_ONBOARDING, runtime route details, subpath mount, bug batch (light theme/panel/model cache/Docker), Docker UID/GID auto-detect, chat transcript redesign, favicon SVG+PNG+ICO, Docker UID-mismatch crash fix, auto-title markdown strip | 1777 |
+| v0.50.101–v0.50.139 | Contributor sprint wave | Custom providers, Russian locale, collapsed timestamps, IME composition fixes, model-switch toast, approval queue multi-slot, live model fetching SSRF guard, orphaned tool-message sanitization, profile polish sprint (model routing, workspace cross-profile, legacy session backfill), font-size CSS fix | 1777 |
+| v0.50.140–v0.50.147 | Bug batch + appearance | Font size setting visibly scales UI text (#843), slash command echoed as user message (#840), scroll selected item into view (#838), tasks refresh button (#835), font size toggle (#833), stale model fix (#829), session search clear on boot (#822), gateway SSE polling fallback (#635) | 1858 |
+| v0.50.148–v0.50.150 | Session index + read-path + profile | Prune stale _index.json ghost rows after session-id rotation (#847 @franksong2702), GET /api/session side-effect-free model resolution (#848 @franksong2702), profile switching cookie persist + syncTopbar fix (#849 @migueltavares) | 1858 |
+| v0.50.151 | credential_pool + Ollama Cloud | Providers added via auth store credential_pool now visible in model dropdown; Ollama Cloud support; ambient gh-cli token suppression; _apply_provider_prefix helper (#820 @starship-s) | 1898 |
+| v0.50.152 | Image rendering + auto-title | image_generate MEDIA: token renders all https:// URLs as img regardless of extension (closes #853); auto-title strips Qwen3-style plain-text thinking preambles (closes #857) | 1898 |
+| v0.50.153 | Portal model routing | Live-fetched models from portal providers (Nous, OpenCode) now get @provider: prefix so they route correctly instead of falling through to OpenRouter (closes #854) | 1898 |
+| v0.50.154 | Thinking card mirror fix | _streamDisplay() early return removed — thinking card and main response now show distinct content when provider double-emits (closes #852) | 1898 |
+| v0.50.155 | Honcho session stability | gateway_session_key=session_id passed to AIAgent so Honcho per-session strategy maintains one Honcho session per WebUI chat instead of one per turn (closes #855) | 1903 |
+| v0.50.156 | Auto-install security gate | auto_install_agent_deps() is now opt-in; set HERMES_WEBUI_AUTO_INSTALL=1 to enable; _trusted_agent_dir() checks ownership/permission bits before running pip (⚠️ breaking: default changed) | 1903 |
 
 ---
 
@@ -54,14 +93,14 @@
 
 | Layer | Location | Status |
 |-------|----------|--------|
-| Python server | <repo>/server.py (~81 lines) + api/ modules (~3210 lines) | Thin shell + auth middleware + business logic in api/ |
-| HTML template | <repo>/static/index.html (~364 lines) | Served from disk |
-| CSS | <repo>/static/style.css (~670 lines) | Served from disk, incl. mobile responsive |
-| JavaScript | <repo>/static/{ui,workspace,sessions,messages,panels,boot,commands}.js | 7 modules, ~3610 lines total |
+| Python server | <repo>/server.py (~165 lines) + api/ modules (~5000 lines) | Thin shell + QuietHTTPServer + auth middleware + business logic in api/ |
+| HTML template | <repo>/static/index.html (~600 lines) | Served from disk |
+| CSS | <repo>/static/style.css (~1050 lines) | Served from disk, incl. mobile responsive, KaTeX, table styles |
+| JavaScript | <repo>/static/{ui,workspace,sessions,messages,panels,boot,commands,icons,i18n,login}.js | 10 modules, ~7100 lines total |
 | Docker | Dockerfile, docker-compose.yml, .dockerignore | python:3.12-slim, multi-arch (amd64+arm64) |
 | CI/CD | .github/workflows/release.yml | Auto-release + GHCR publish on tag push |
 | Runtime state | ~/.hermes/webui-mvp/sessions/ | Session JSON files |
-| Test server | Port 8788, state dir ~/.hermes/webui-mvp-test/ | Isolated, wiped per run |
+| Test server | Port 8788 (conftest.py), port 8789 (browser sanity) | Isolated, wiped per run |
 | Production server | Port 8787 | SSH tunnel from Mac |
 
 ---
@@ -71,11 +110,12 @@
 ### Chat and Agent
 - [x] Send messages, get SSE-streaming responses
 - [x] Switch models per session (10 models, grouped by provider)
+- [x] Composer-scoped model picker in footer (moved from sidebar to align with per-conversation model selection)
 - [x] Multi-provider API support: use any Hermes agent API provider (OpenAI, Anthropic, Google, etc.) directly, not just OpenRouter (Sprint 11)
 - [x] Custom endpoint model discovery: auto-detect models from Ollama, LM Studio, and other local LLM servers via base_url (PR #18)
 - [x] Upload files to workspace (drag-drop, click, clipboard paste)
 - [x] File tray with remove button
-- [x] Tool progress shown in activity bar above composer
+- [x] Tool progress shown inline in the conversation via live tool cards
 - [x] Approval card for dangerous commands (Allow once/session/always, Deny)
 - [x] Approval polling + SSE-pushed approval events
 - [x] INFLIGHT guard: switch sessions mid-request without losing response
@@ -87,23 +127,25 @@
 - [x] Token/cost estimate per message (Sprint 23)
 
 ### Tool Visibility
-- [x] Tool progress in activity bar (moved out of composer footer)
+- [x] Tool progress in live tool cards (kept out of the composer/footer chrome)
 - [x] Approval card with all 4 choices
 - [x] Tool call cards inline (collapsed, show name/args/result)
 
 ### Workspace / Files
+- [x] Workspace panel defaults closed and opens only for active browsing or preview
 - [x] Browse workspace directory tree with type icons
 - [x] Preview text/code files (read-only)
 - [x] Preview markdown files (rendered, tables supported)
 - [x] Preview image files (PNG, JPG, GIF, SVG, WEBP inline)
 - [x] Edit files inline (Edit button, Enter to save, Escape to cancel)
 - [x] Create new file (+ button in panel header)
-- [x] Delete file (hover trash, confirm dialog)
+- [x] Delete file (hover trash, confirmation modal)
 - [x] File name truncation with tooltip for long names
 - [x] Right panel resizable (drag inner edge)
 - [x] Syntax highlighted code preview (Prism.js)
 - [x] Rename file (Sprint 14)
 - [x] Create folder (Sprint 14)
+- [x] Shared app modal for confirm/input flows (Sprint 33)
 
 ### Sessions
 - [x] Create session (+ button or Cmd/Ctrl+K)
@@ -193,7 +235,7 @@
 - [x] Voice input via Web Speech API (Sprint 20)
 
 ### Mobile
-- [x] Mobile responsive layout — hamburger sidebar, bottom nav, files slide-over (Sprint 21)
+- [x] Mobile responsive layout — hamburger sidebar, sidebar tabs on phones, files slide-over (Sprint 21 + later mobile nav simplification)
 
 ### Profiles
 - [x] Multi-profile support — create, switch, delete profiles (Sprint 22, Issue #28)
@@ -204,14 +246,14 @@
 - [x] Streaming performance -- rAF-throttled token rendering (Sprint 24, PR #81)
 - [x] Workspace git detection -- branch name and dirty status badge (Sprint 24, PR #82)
 - [x] Collapsible date groups -- click group headers to collapse (Sprint 24, PR #80)
-- [x] Context usage indicator -- token count and cost in composer footer (Sprint 24, PR #83)
+- [x] Context usage indicator -- compact circular badge in composer footer (Sprint 24, PR #83; refreshed April 10, 2026)
 - [ ] LLM-generated session titles -- auto-title via small model instead of first-message substring (PR #75)
 - [ ] Workspace git detection -- show branch name, dirty status in workspace header (PR #75)
 - [ ] Clarify dialog -- agent can ask clarifying questions that block until user responds (PR #75)
 - [ ] Gateway approval polling -- support blocking approvals from messaging gateway (PR #75)
 - [ ] Unified session storage -- SessionDB shared between webui and CLI (PR #75)
 - [ ] TTS playback of responses (deferred)
-- [x] Background task cancel (activity bar Cancel button)
+- [x] Background task cancel (composer footer stop button)
 - [ ] Code execution cell (deferred)
 - [ ] Desktop application (Sprint 25, PLANNED)
 - [x] Pluggable UI themes -- Dark, Light, Slate, Solarized, Monokai, Nord (Sprint 26, v0.34)

SPRINTS.md

@@ -1,32 +1,45 @@
 # Hermes Web UI -- Forward Sprint Plan
 
-> Current state: v0.36 | 433 tests | Daily driver ready
-> This document plans the path from here to two targets:
+> Current state: v0.50.156 | 1903 tests | Full daily driver — CLI parity achieved
 >
-> Target A: 1:1 feature parity with the Hermes CLI (everything you can do from the
->           terminal, you can do from the browser)
+> NOTE: This file is preserved as a historical planning record. Current sprint state
+> and version history live in CHANGELOG.md and ROADMAP.md.
 >
-> Target B: 1:1 parity with Claude's reproducible features (the full Claude
->           browser UI experience, minus things only Anthropic can build)
+> Target A (CLI parity): ✅ Complete — all core tools, workspace, cron, skills,
+>           memory, sessions, profiles, model routing, streaming, voice, mobile.
 >
-> Sprints are ordered by impact. Each builds on the one before.
-> Past sprint history lives in CHANGELOG.md.
+> Target B (Claude parity): ~90% — thinking display, math rendering (KaTeX),
+>           tool cards, workspace preview, onboarding, settings panel all done.
+>           Remaining: full subagent transparency UI, file diff viewer.
+>
+> Last meaningful update: v0.50.21 (April 13, 2026). See CHANGELOG.md for full history.
 
 ---
 
-## Where we are now (v0.21)
+## Where we are now (v0.50.21 — updated April 2026)
 
-**CLI parity: ~90% complete.** Core agent loop, all tools visible, workspace
-file ops with tree view, cron/skills/memory CRUD, session management, streaming,
-cancel, multi-provider models, custom endpoint discovery, slash commands,
-thinking/reasoning display, password auth -- all solid. Gaps are subagent
-visibility, toolset control, and code execution.
+> The sections below describe the state as of v0.36 for historical reference.
+> See ROADMAP.md for the current sprint history table (v0.36 → v0.50.21).
 
-**Claude parity: ~70% complete.** Chat, streaming, file browser, session
-management, tool cards, syntax highlighting, model switching, projects,
-settings, Mermaid diagrams, mobile layout, breadcrumb workspace nav, slash
-commands, thinking display, auth -- all present. Gaps are artifacts, voice,
-TTS, sharing, mobile-optimized layout.
+**CLI parity: ✅ Complete** as of v0.50.x. Core agent loop, all tools visible, workspace
+file ops with tree view and git detection, cron/skills/memory CRUD, session
+management, streaming with rAF throttle, cancel, multi-provider models, custom
+endpoint discovery, slash commands (help/clear/model/workspace/new/usage/theme/compact),
+thinking/reasoning display, password auth, multi-profile support with seamless
+switching, CLI session bridge (read and import from state.db), context
+auto-compaction handling, self-update checker. Remaining gaps: subagent
+session tree, toolset control per session, code execution cells.
+
+**Claude parity: ~85% complete.** Chat, streaming, file browser, session
+management with projects and tags, tool cards with subagent delegation,
+syntax highlighting, model switching, Mermaid diagrams, mobile responsive
+layout (hamburger sidebar, bottom nav, files slide-over), breadcrumb
+workspace nav with tree view, slash commands, thinking/reasoning display,
+auth with signed cookies, 6 pluggable UI themes (dark/light/slate/solarized/
+monokai/nord), voice input (Web Speech API), collapsible date groups,
+context usage indicator, token/cost display, git branch badge, Docker
+support. Remaining gaps: artifacts (HTML/SVG preview), TTS playback,
+sharing/public URLs, code execution inline.
 
 ---
 
@@ -248,7 +261,7 @@ inconsistently across platforms. These were the most common visual complaints.
   button now only appears in the hover overlay like all other actions.
 
 ### Track B: Features
-- **SVG action icons.** Replaced all emoji HTML entities (★, 📂, 📦, ⊕, 🗑)
+- **SVG action icons.** Replaced old symbol and emoji HTML entities
   with monochrome SVG line icons that inherit `currentColor`. Consistent
   rendering across macOS, Linux, and Windows. Icons: pin (star), folder,
   archive (box), duplicate (overlapping squares), trash (bin with lines).
@@ -754,7 +767,7 @@ Both architectures in one .app. No separate downloads needed.
 - JS bridge fires when approval card appears/disappears
 
 **Menu bar mode (optional, v2):**
-- A small status bar item (⚗️ icon in menu bar) that opens a compact popover
+- A small status bar item (beaker icon in menu bar) that opens a compact popover
 - Popover shows current session status, last message, quick-compose field
 - Useful for running Hermes in the background without a full window
 
@@ -1155,7 +1168,8 @@ New test cases in `tests/test_sprint26.py`:
 
 ---
 
-*Last updated: April 5, 2026*
-*Current version: v0.36 | 433 tests*
+*Last updated: April 12, 2026*
+*Current version: v0.49.1 | 700 tests*
 *Next sprint: Sprint 24 (Web Polish + Bug Fix Pass)*
 *Horizon sprint: Sprint 25 (macOS Desktop Application)*
+*Docs sweep policy: update markdown proactively during PR reviews and after significant releases*

TESTING.md

@@ -1,15 +1,17 @@
 # Hermes Web UI: Browser Testing Plan
 
 > This document is for manual browser testing by you or by a Claude browser agent.
-> It covers user-facing features of the UI through Sprint 26 (v0.34.3).
+> It covers user-facing features of the UI through v0.50.21 and later releases.
 > Each section is written as a step-by-step test procedure with expected outcomes.
 > A browser agent (e.g. Claude with Chrome access) can execute this plan directly.
 >
-> Prerequisites: SSH tunnel is active on port 8786. Open http://localhost:8786 in browser.
-> Server health check: curl http://127.0.0.1:8786/health should return {"status":"ok"}.
+> Prerequisites: SSH tunnel is active on port 8787. Open http://localhost:8787 in browser.
+> Server health check: curl http://127.0.0.1:8787/health should return {"status":"ok"}.
 >
-> Automated tests: 433 total (433 passing, 0 failures)
+> Automated coverage: 2239 tests collected via `pytest tests/ --collect-only -q`. Includes onboarding coverage for bootstrap/static wizard presence, real provider config persistence (`config.yaml` + `.env`), the `/api/onboarding/*` backend, the onboarding skip/existing-config guard, and CSS regression coverage for smooth thinking/tool card disclosure animation.
 > Run: `pytest tests/ -v --timeout=60`
+>
+> Local regression focus: verify that a previously closed workspace panel stays visually closed from first paint through boot completion on desktop refresh; there should be no brief open-then-close flash.
 
 ---
 
@@ -32,9 +34,11 @@ SETUP: Clear localStorage (DevTools > Application > Local Storage > delete herme
 STEPS:
   1. Navigate to http://localhost:8787
 EXPECT:
-  - Dark background, Hermes logo in sidebar header
+  - Dark background
+  - Sidebar begins directly with the icon tab row; there is no dedicated branding header
   - Center area shows "What can I help with?" heading with suggestion buttons
   - Session list in sidebar is empty or shows existing sessions
+  - Sidebar footer shows a single "Hermes WebUI" control-center button
   - No session is highlighted active
   - Send button is present but there is no input focus by default
 FAIL: Page shows error, blank white screen, or auto-creates a new session without user action.
@@ -73,11 +77,11 @@ STEPS:
 EXPECT:
   - User message appears immediately in chat
   - Thinking dots (three animated dots) appear below
-  - Status bar shows "Hermes is thinking..."
   - Send button becomes disabled (grayed out)
+  - A red stop button appears in the composer footer while the turn is running
   - Within 10-30 seconds, Hermes responds with a three-word greeting
   - Thinking dots disappear
-  - Send button re-enables
+  - Send button re-enables and the stop button disappears
   - Session title in sidebar updates to reflect the first message
 FAIL: Message never appears, thinking dots never go away, Send button stays disabled forever.
 
@@ -144,7 +148,7 @@ FAIL: New session created, error thrown, or UI breaks.
 ### T3.1: Model Dropdown Shows All Options
 SETUP: Any active session.
 STEPS:
-  1. Look at the sidebar bottom: "Model" label and a dropdown
+  1. Look at the composer footer: to the right of the attach/mic controls there is a model dropdown
   2. Click the dropdown to expand it
 EXPECT:
   - Provider groups visible: OpenAI, Anthropic, Other
@@ -153,18 +157,30 @@ EXPECT:
   - Other group: Gemini 2.5 Pro, DeepSeek V3, Llama 4 Scout
 FAIL: Only 2 options visible, no groups, or missing models.
 
-### T3.2: Model Chip Reflects Selection
+### T3.2: Model Dropdown Reflects Active Conversation
 SETUP: Active session.
 STEPS:
   1. Change model dropdown to "Claude Sonnet 4.6"
 EXPECT:
-  - The blue chip in the topbar right updates to "Sonnet 4.6" immediately
-  - NOT "GPT-5.4 Mini" (this was Bug B3, now fixed)
+  - The composer footer dropdown stays on "Claude Sonnet 4.6"
+  - Sending the next message uses that session model rather than an older one from another conversation
 STEPS (continued):
   2. Change model to "Gemini 2.5 Pro"
 EXPECT:
-  - Chip updates to "Gemini 2.5 Pro" (not "GPT-5.4 Mini")
-FAIL: Chip shows wrong model name for any non-Sonnet selection.
+  - The dropdown updates to "Gemini 2.5 Pro"
+  - Switching away and back to the conversation restores the same model in the footer selector
+FAIL: Dropdown shows the wrong active model after a session switch, or sending uses a stale model.
+
+### T3.3: Context Badge Shares Footer Space Cleanly
+SETUP: Active session with at least one completed response.
+STEPS:
+  1. Look at the right side of the composer footer
+EXPECT:
+  - A compact circular context badge appears next to the send button when usage data is available
+  - The number in the center shows the used percentage
+  - Hovering or focusing the badge shows a tooltip with percent used, token count, auto-compress threshold, and estimated cost when available
+  - The model dropdown remains usable without overlapping the send button or pushing controls out of view
+FAIL: Linear meter still shown, tooltip missing/incomplete, controls overlap, or footer wraps in a broken way.
 
 ---
 
@@ -228,8 +244,18 @@ FAIL: File not removed, error.
 
 ## Section 5: Workspace File Browser
 
-### T5.1: File Tree Loads on Session Start
+### T5.0: Panel Is Closed By Default
 SETUP: Active session with workspace set.
+EXPECT:
+  - Right workspace panel is hidden on initial load
+  - Center chat column uses the freed width
+  - "Files" toggle is visible in the topbar
+FAIL: Right panel starts open without any browsing or preview action.
+
+### T5.1: File Tree Loads When Files Panel Is Opened
+SETUP: Active session with workspace set.
+STEPS:
+  1. Click the "Files" toggle in the topbar
 EXPECT:
   - Right panel shows "WORKSPACE" header
   - File tree lists files and directories in the workspace
@@ -263,10 +289,11 @@ STEPS:
   1. Click the X button in the panel header
 EXPECT:
   - Preview closes
-  - File tree is visible again
+  - If the panel auto-opened for that preview, the entire right panel closes again
+  - If the panel was manually opened for browsing first, the file tree is visible again
   - Preview area is hidden
   - Reopening the same file shows fresh content (no stale cached text)
-FAIL: X button does nothing, tree does not reappear.
+FAIL: X button does nothing, panel stays stuck open, or the file tree does not reappear after manual browse mode.
 
 ### T5.5: Preview an Image File (Sprint 2)
 SETUP: Upload a PNG, JPG, or any image file to the workspace, OR the workspace already contains one.
@@ -377,7 +404,8 @@ FAIL: Command blocked after Allow once, card stays, error.
 ### T8.1: Download Conversation as Markdown
 SETUP: A session with at least 2 messages (1 user + 1 assistant).
 STEPS:
-  1. Click the "Transcript" download button in the sidebar bottom
+  1. Click the "Hermes" button in the sidebar footer
+  2. In the Control Center modal, click "Transcript"
 EXPECT:
   - Browser downloads a .md file named hermes-{session_id}.md
   - Opening the file shows the conversation in markdown format:
@@ -468,6 +496,7 @@ FAIL: No log output, log shows Apache-style text instead of JSON, log file not c
 SETUP: Message is sending (thinking dots visible).
 EXPECT:
   - Send button is visually grayed out
+  - Stop button is visible in the composer footer
   - Pressing Enter does NOT send another message
   - Clicking Send button does nothing
 FAIL: Multiple messages sent while one is in flight.
@@ -831,7 +860,7 @@ FAIL: No icon ever appears, icon always visible (not hover-only).
 ### T21.2: Delete a File with Confirmation
 STEPS:
   1. Hover over a file and click its trash icon
-  2. A browser confirm dialog appears: "Delete [filename]?"
+  2. An in-app confirmation modal appears: "Delete [filename]?"
   3. Click OK
 EXPECT:
   - Toast: "Deleted [filename]"
@@ -842,7 +871,7 @@ FAIL: File not deleted, no confirmation dialog, error.
 ### T21.3: Cancel Delete Does Nothing
 STEPS:
   1. Hover over a file and click its trash icon
-  2. Click Cancel on the confirm dialog
+  2. Click Cancel on the confirmation modal
 EXPECT:
   - File remains in the tree
   - No toast, no error
@@ -851,7 +880,7 @@ FAIL: File deleted despite cancel.
 ### T21.4: Create a New File
 STEPS:
   1. Click the + button in the workspace panel header
-  2. A prompt dialog appears: "New file name (e.g. notes.md):"
+  2. An in-app input modal appears: "New file name (e.g. notes.md):"
   3. Type "test-sprint4.md" and click OK
 EXPECT:
   - Toast: "Created test-sprint4.md"
@@ -925,7 +954,7 @@ FAIL: Invalid path added, no error.
 ### T22.4: Remove a Workspace
 STEPS:
   1. Click the X button next to any non-default workspace
-  2. Confirm the dialog
+  2. Confirm the modal
 EXPECT:
   - Workspace disappears from the list
   - Toast: "Workspace removed"
@@ -981,7 +1010,7 @@ STEPS:
   1. Hover over an assistant message
   2. Click the clipboard icon
 EXPECT:
-  - Icon briefly shows a checkmark (✓) then reverts to clipboard
+  - Icon briefly shows a check icon, then reverts to the copy icon
   - Paste (Cmd+V) elsewhere shows the full text of that message
 FAIL: No visual feedback, clipboard empty or wrong content.
 
@@ -994,23 +1023,23 @@ STEPS:
   1. Click any .py, .js, or .txt file in the workspace file tree
 EXPECT:
   - File content shows in read-only monospace view
-  - An "✎ Edit" button is visible in the preview path bar
+  - An Edit button with a pencil icon is visible in the preview path bar
   - Content is NOT editable (clicking in it does nothing)
 FAIL: Content immediately editable, no Edit button.
 
 ### T24.2: Edit Button Enters Edit Mode
 STEPS:
-  1. Click "✎ Edit" on a code file preview
+  1. Click the Edit button on a code file preview
 EXPECT:
   - Read-only view replaced by an editable textarea
   - Content of the file is pre-populated in the textarea
-  - Button changes to "💾 Save"
+  - Button changes to "Save" with a disk icon
 FAIL: Nothing changes, button doesn't change.
 
 ### T24.3: Save Writes Changes to Disk
 STEPS:
   1. In edit mode, change some text
-  2. Click "💾 Save"
+  2. Click the Save button
 EXPECT:
   - Read-only view returns, showing the updated content
   - Toast: "Saved"
@@ -1022,8 +1051,8 @@ STEPS:
   1. Enter edit mode on a file
   2. Make any change (type a character)
 EXPECT:
-  - Button shows "💾 Save*" (asterisk indicates unsaved changes)
-FAIL: No asterisk, button stays as "💾 Save".
+  - Button shows "Save*" with the disk icon still visible (asterisk indicates unsaved changes)
+FAIL: No asterisk, button stays as "Save".
 
 ### T24.5: Markdown File Edit-Save Roundtrip
 STEPS:
@@ -1070,7 +1099,7 @@ against each criterion below. A Claude browser agent can verify these with brows
 
 ### T25.1: Sidebar Nav Tabs are Icon-Only
 EXPECT:
-  - Five icon-only tabs in the sidebar nav row: 💬 ⏱️ 📚 🧠 📁
+  - Five icon-only tabs in the sidebar nav row: message, clock, book, brain, folder
   - No text labels visible by default (text removed to prevent overflow)
   - Hovering a tab shows a tooltip with the label (Chat/Tasks/Skills/Memory/Spaces)
   - Active tab has a blue underline, icon brighter blue
@@ -1194,7 +1223,7 @@ STEPS:
   3. Click Create job
 EXPECT:
   - Form closes
-  - Toast: "Job created ✓"
+  - Toast: "Job created"
   - New job appears in the cron list with status "active"
 FAIL: Error shown, job not created, form stays open.
 
@@ -1225,7 +1254,8 @@ FAIL: Job created, form doesn't close.
 ### T28.1: JSON Export Button Downloads File
 SETUP: Active session with at least a few messages.
 STEPS:
-  1. Click the "JSON" button in the sidebar footer (next to Transcript)
+  1. Click the "Hermes" button in the sidebar footer
+  2. In the Control Center modal, click "JSON"
 EXPECT:
   - Browser downloads a file named hermes-{session_id}.json
   - Opening the file shows valid JSON with: session_id, title, messages array,
@@ -1289,7 +1319,7 @@ STEPS (continued from T29.1):
   1. Change the name field to "Renamed Job"
   2. Click Save
 EXPECT:
-  - Form closes, toast "Job updated ✓"
+  - Form closes, toast "Job updated"
   - Job header shows new name
 FAIL: Save fails, name unchanged.
 
@@ -1297,7 +1327,7 @@ FAIL: Save fails, name unchanged.
 SETUP: A cron job you can safely delete (or a test job created for this).
 STEPS:
   1. Expand the job, click "Delete"
-  2. Confirm the dialog
+  2. Confirm the modal
 EXPECT:
   - Toast: "Job deleted"
   - Job disappears from the list
@@ -1326,7 +1356,7 @@ tags: [test]
 # Test"
   2. Click Save skill
 EXPECT:
-  - Toast "Skill created ✓", form closes
+  - Toast "Skill created", form closes
   - Skill appears in the skills list
 FAIL: Error, skill not in list.
 
@@ -1354,7 +1384,7 @@ STEPS:
   1. In edit mode, add a line to the textarea
   2. Click Save
 EXPECT:
-  - Toast "Memory saved ✓", form closes
+  - Toast "Memory saved", form closes
   - Memory panel reloads showing the updated content
 FAIL: Save fails, content unchanged.
 
@@ -1467,14 +1497,16 @@ FAIL: Both messages removed, wrong message sent, crash.
 ### T34.1: Clear Button Appears When Session Has Messages
 SETUP: Session with at least one message.
 EXPECT:
-  - A "🗑 Clear" chip appears in the topbar right side (next to the workspace chip)
-  - Button NOT visible when session has no messages / empty state
+  - The "Hermes" button is visible in the sidebar footer
+  - Opening the Control Center shows a "Clear" action in the Conversation section
+  - The Clear action is disabled when there is no active session or no messages
 FAIL: Button always visible, never visible.
 
 ### T34.2: Clear Wipes Messages and Resets Title
 STEPS:
-  1. Click the Clear button in the topbar
-  2. Confirm the dialog
+  1. Click the "Hermes" button in the sidebar footer
+  2. Click "Clear" in the Conversation section
+  3. Confirm the modal
 EXPECT:
   - All messages disappear from the chat area
   - Empty state ("What can I help with?") reappears
@@ -1485,7 +1517,7 @@ FAIL: Session deleted, messages remain, title not reset.
 
 ### T34.3: Cancel Clear Does Nothing
 STEPS:
-  1. Click Clear, then click Cancel in the confirm dialog
+  1. Click Clear, then click Cancel in the confirmation modal
 EXPECT:
   - All messages still present
   - No toast, no change
@@ -1609,7 +1641,7 @@ Each has automated API-level tests in `tests/test_sprint{N}.py`.
 - Switch model. Send a message. Verify response uses selected model.
 
 ### Sprint 12: Settings + Pin + Import
-- Click gear icon. Settings overlay opens.
+- Click the "Hermes WebUI" button in the sidebar footer. Control Center overlay opens with vertical section tabs on the left.
 - Change default model, save. Restart server. Verify setting persisted.
 - Pin a session (star icon in hover overlay). Verify it floats to top of list.
 - Export session as JSON. Import it back. Verify messages restored.
@@ -1637,11 +1669,12 @@ Each has automated API-level tests in `tests/test_sprint{N}.py`.
 
 ### Sprint 16: Sidebar Visual Polish
 - Session titles use full sidebar width (no truncated space for hidden icons).
-- Hover a session → action buttons appear from right with gradient fade.
+- Hover a session → a dotted actions trigger appears on the right.
+- Click the dotted trigger → a dropdown opens with pin, project, archive, duplicate, and delete actions.
 - All icons are monochrome SVGs (not emoji). Consistent across platforms.
 - Pinned sessions show small gold star inline. Unpinned = no star, full title width.
-- Active session has gold highlight (not blue). Overlay gradient matches.
-- Double-click to rename → overlay hides during rename.
+- Active session has gold highlight (not blue).
+- Double-click to rename → session actions hide during rename.
 
 ### Sprint 17: Workspace + Slash Commands + Send Key
 - Navigate into a subdirectory. Breadcrumb bar appears with clickable segments.
@@ -1655,6 +1688,13 @@ Each has automated API-level tests in `tests/test_sprint{N}.py`.
 - Click a directory toggle arrow (▸) → expands in-place showing children.
 - Click again (▾) → collapses. Double-click navigates into it (breadcrumb view).
 - If model returns thinking blocks (Claude extended thinking), verify collapsible gold card appears above response.
+- Verify the thinking card has a tinted background, visible border, and rounded corners like a tool card, but in the gold thinking palette.
+- Open and close a thinking card. Verify the caret rotation and the content reveal both animate smoothly instead of snapping open.
+
+### UI Polish: Tool Card Disclosure Animation
+- Trigger a response with at least one completed tool call card.
+- Open and close the tool call card. Verify the caret rotates smoothly and the args/result section animates open and closed instead of appearing instantly.
+- If a turn has 2+ tool cards, use "Expand all / Collapse all" and verify the same smooth animation applies to every card in the group.
 
 ### Sprint 19: Auth + Security
 - No password set: everything works as normal. No login page.
@@ -1684,12 +1724,13 @@ Each has automated API-level tests in `tests/test_sprint{N}.py`.
 - Open on mobile viewport (<640px): hamburger icon visible in topbar.
 - Tap hamburger → sidebar slides in from left with backdrop overlay.
 - Tap outside sidebar → closes. Tap a session → closes and loads session.
-- Bottom navigation bar: 5 tabs (Chat, Tasks, Skills, Memory, Spaces).
-- Tap "Tasks" in bottom nav → sidebar opens showing Tasks panel.
-- Tap "Chat" in bottom nav → sidebar closes (chat is in main area).
+- Sidebar top nav remains visible inside the mobile drawer; includes Chat/Tasks/Skills/Memory/Spaces/Profile tabs.
+- Tap "Tasks" in the drawer nav → Tasks panel opens in the sidebar drawer.
+- Tap "Chat" in the drawer nav → sidebar closes and chat is unobstructed in the main area.
 - Files button in topbar → right panel slides in from right.
+- No fixed mobile bottom nav; chat transcript and composer use the reclaimed vertical space.
 - All touch targets are at least 44px (session items, buttons, icons).
-- Desktop viewport (>640px): no hamburger, no bottom nav, no mobile elements.
+- Desktop viewport (>640px): no hamburger or mobile overlay; desktop layout unchanged.
 - Docker: `docker compose up -d` starts server on port 8787.
 - Docker: session data persists across container restarts (named volume).
 
@@ -1702,14 +1743,47 @@ Each has automated API-level tests in `tests/test_sprint{N}.py`.
 - "Use" button switches profile. Delete button removes non-default profiles.
 - "+ New profile" form: name validation (lowercase + hyphens), clone config checkbox.
 - Create profile → appears in list and dropdown.
-- Delete profile → confirm dialog. Auto-switches to default if deleting active.
+- Delete profile → confirmation modal. Auto-switches to default if deleting active.
 - Attempt switch while agent busy → blocked with toast message.
 - With hermes-agent not installed → only default profile shown, graceful fallback.
 
 ---
 
-*Last updated: Sprint 26 / v0.34.3, April 5, 2026*
-*Total automated tests: 433 (433 passing, 0 failures)*
+## Slash command parity (manual checklist)
+
+For each batch-1 command, run via webui slash menu AND via `hermes` CLI in the
+same `HERMES_HOME` (when applicable) and verify identical effect.
+
+- [ ] `/help` — dropdown lists 25+ commands; selecting `/help` posts an assistant message listing them.
+- [ ] `/new` (and alias `/reset`) — starts fresh session.
+- [ ] `/clear` — clears current transcript display (webui-only meaning, distinct from CLI's "clear screen").
+- [ ] `/title <name>` — renames active session, topbar + sidebar update; `/title` alone shows current title.
+- [ ] `/status` — assistant message shows session_id, model, workspace, message count.
+- [ ] `/usage` — assistant message shows token counts; the "show token usage" setting is unchanged (toggle still in Settings panel).
+- [ ] `/stop` — interrupts a running stream; with no active stream toasts "No active task to stop."
+- [ ] `/retry` — removes last user+assistant exchange, refills composer with last user text, resends. Final transcript has only ONE copy of the resent message.
+- [ ] `/undo` — removes last user+assistant exchange; toast confirms; repeated until empty toasts "Nothing to undo."
+- [ ] `/model <name>` — switches model dropdown.
+- [ ] `/personality` — lists personalities; `/personality <name>` switches.
+- [ ] `/skills [query]` — lists matching skills.
+- [ ] `/theme <name>` — switches webui theme.
+- [ ] `/workspace <name>` — switches workspace.
+
+Unknown / deferred:
+
+- [ ] `/yolo`, `/reasoning`, `/voice`, `/branch`, `/insights`, `/debug`, `/reload`, etc. — toast "Web UI 暂未实现该命令: /<name>". MUST NOT be sent as plain text to the LLM.
+- [ ] `/compact` — toast "/compress is not available in the web UI yet — use the CLI for now." (was sending free text to LLM before this batch.)
+- [ ] Made-up command (e.g. `/fhfajl`) — fall through to send as text (existing behavior preserved for typos vs. real commands).
+
+Bridged CLI sessions:
+
+- [ ] Open a CLI-bridged session in webui sidebar (if `show_cli_sessions` setting enabled).
+- [ ] `/retry`, `/undo` toast "该命令仅支持 Web UI 原生会话…" and do nothing.
+
+---
+
+*Last updated: v0.50.91, April 19, 2026*
+*Total automated tests collected: 2107*
 *Regression gate: tests/test_regressions.py*
 *Run: pytest tests/ -v --timeout=60*
 *Source: <repo>/*

THEMES.md

@@ -1,6 +1,6 @@
 # Hermes Web UI — Themes
 
-Hermes Web UI supports pluggable color themes. Five themes ship built-in, and
+Hermes Web UI supports pluggable color themes. Seven themes ship built-in, and
 you can create your own with pure CSS — no Python changes needed.
 
 ---
@@ -27,6 +27,7 @@ preview is instant — the UI updates as you click through options.
 | **Solarized Dark** | Ethan Schoonover's classic dark palette. Teal background, warm accents. |
 | **Monokai** | Warm dark theme inspired by the Monokai editor scheme. Green/pink accents. |
 | **Nord** | Arctic blue-gray palette from the Nord color system. Calm and minimal. |
+| **OLED** | True black (#000) backgrounds for OLED displays. Minimizes glow and burn-in risk. |
 | **Custom themes** | Any string accepted by `settings.json`, `POST /api/settings`, and `/theme` if added to the picker/command list. Pure CSS variables only. |
 
 ---

api/agent_sessions.py

@@ -0,0 +1,55 @@
+"""Shared helpers for reading Hermes Agent sessions from state.db."""
+import logging
+import sqlite3
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+
+def read_importable_agent_session_rows(db_path: Path, limit: int = 200, log=None) -> list[dict]:
+    """Return non-WebUI agent sessions that have readable message rows.
+
+    Hermes Agent can create rows in ``state.db.sessions`` before a session has
+    any messages. WebUI cannot import those rows, so both the regular
+    ``/api/sessions`` path and the gateway SSE watcher must filter them the
+    same way.
+    """
+    db_path = Path(db_path)
+    if not db_path.exists():
+        return []
+
+    log = log or logger
+    with sqlite3.connect(str(db_path)) as conn:
+        conn.row_factory = sqlite3.Row
+        cur = conn.cursor()
+
+        # Older Hermes Agent versions may not have source tracking. Without a
+        # source column we cannot safely distinguish WebUI rows from agent rows.
+        cur.execute("PRAGMA table_info(sessions)")
+        session_cols = {row[1] for row in cur.fetchall()}
+        if 'source' not in session_cols:
+            log.warning(
+                "agent session listing skipped: state.db at %s has no 'source' column "
+                "(older hermes-agent?). Agent sessions unavailable. "
+                "Upgrade hermes-agent to fix this.",
+                db_path,
+            )
+            return []
+
+        cur.execute(
+            """
+            SELECT s.id, s.title, s.model, s.message_count,
+                   s.started_at, s.source,
+                   COUNT(m.id) AS actual_message_count,
+                   MAX(m.timestamp) AS last_activity
+            FROM sessions s
+            LEFT JOIN messages m ON m.session_id = s.id
+            WHERE s.source IS NOT NULL AND s.source != 'webui'
+            GROUP BY s.id
+            HAVING COUNT(m.id) > 0
+            ORDER BY COALESCE(MAX(m.timestamp), s.started_at) DESC
+            LIMIT ?
+            """,
+            (int(limit),),
+        )
+        return [dict(row) for row in cur.fetchall()]

api/auth.py

@@ -6,12 +6,17 @@ or configuring a password in the Settings panel.
 import hashlib
 import hmac
 import http.cookies
+import json
+import logging
 import os
 import secrets
+import tempfile
 import time
 
 from api.config import STATE_DIR, load_settings
 
+logger = logging.getLogger(__name__)
+
 # ── Public paths (no auth required) ─────────────────────────────────────────
 PUBLIC_PATHS = frozenset({
     '/login', '/health', '/favicon.ico',
@@ -21,20 +26,86 @@ PUBLIC_PATHS = frozenset({
 COOKIE_NAME = 'hermes_session'
 SESSION_TTL = 86400  # 24 hours
 
-# Active sessions: token -> expiry timestamp
-_sessions = {}
+_SESSIONS_FILE = STATE_DIR / '.sessions.json'
+
+
+def _load_sessions() -> dict[str, float]:
+    """Load persisted sessions from STATE_DIR, pruning expired entries.
+
+    Returns an empty dict on any read or parse error so startup is never
+    blocked by a corrupt or missing sessions file.
+    """
+    try:
+        if _SESSIONS_FILE.exists():
+            data = json.loads(_SESSIONS_FILE.read_text(encoding='utf-8'))
+            if not isinstance(data, dict):
+                raise ValueError('malformed sessions file — expected dict')
+            now = time.time()
+            return {t: exp for t, exp in data.items()
+                    if isinstance(t, str) and isinstance(exp, (int, float)) and exp > now}
+    except Exception as e:
+        logger.debug("Failed to load sessions file, starting fresh: %s", e)
+    return {}
+
+
+def _save_sessions(sessions: dict[str, float]) -> None:
+    """Atomically persist sessions to STATE_DIR/.sessions.json (0600).
+
+    Uses a temp file + os.replace() so a crash mid-write never leaves a
+    truncated file.  Mirrors the same pattern as .signing_key persistence.
+    """
+    try:
+        STATE_DIR.mkdir(parents=True, exist_ok=True)
+        fd, tmp = tempfile.mkstemp(dir=STATE_DIR, suffix='.sessions.tmp')
+        try:
+            with os.fdopen(fd, 'w', encoding='utf-8') as f:
+                json.dump(sessions, f)
+            os.chmod(tmp, 0o600)
+            os.replace(tmp, _SESSIONS_FILE)
+        except Exception:
+            try:
+                os.unlink(tmp)
+            except OSError:
+                pass
+            raise
+    except Exception as e:
+        logger.debug("Failed to persist sessions: %s", e)
+
+
+# Active sessions: token -> expiry timestamp (persisted across restarts via STATE_DIR)
+_sessions = _load_sessions()
+
+# ── Login rate limiter ──────────────────────────────────────────────────────
+_login_attempts = {}  # ip -> [timestamp, ...]
+_LOGIN_MAX_ATTEMPTS = 5
+_LOGIN_WINDOW = 60  # seconds
+
+def _check_login_rate(ip: str) -> bool:
+    """Return True if the IP is allowed to attempt login."""
+    now = time.time()
+    attempts = _login_attempts.get(ip, [])
+    # Prune old attempts
+    attempts = [t for t in attempts if now - t < _LOGIN_WINDOW]
+    _login_attempts[ip] = attempts
+    return len(attempts) < _LOGIN_MAX_ATTEMPTS
+
+def _record_login_attempt(ip: str) -> None:
+    now = time.time()
+    attempts = _login_attempts.get(ip, [])
+    attempts.append(now)
+    _login_attempts[ip] = attempts
 
 
 def _signing_key():
     """Return a random signing key, generating and persisting one on first call."""
     key_file = STATE_DIR / '.signing_key'
-    if key_file.exists():
-        try:
+    try:
+        if key_file.exists():
             raw = key_file.read_bytes()
             if len(raw) >= 32:
                 return raw[:32]
-        except Exception:
-            pass
+    except Exception:
+        logger.debug("Failed to read or access signing key file, using in-memory key")
     # Generate a new random key
     key = secrets.token_bytes(32)
     try:
@@ -42,7 +113,7 @@ def _signing_key():
         key_file.write_bytes(key)
         key_file.chmod(0o600)
     except Exception:
-        pass  # key works for this process even if persist fails
+        logger.debug("Failed to persist signing key, using in-memory key only")
     return key
 
 
@@ -84,16 +155,28 @@ def create_session() -> str:
     """Create a new auth session. Returns signed cookie value."""
     token = secrets.token_hex(32)
     _sessions[token] = time.time() + SESSION_TTL
-    sig = hmac.new(_signing_key(), token.encode(), hashlib.sha256).hexdigest()[:16]
+    _save_sessions(_sessions)
+    sig = hmac.new(_signing_key(), token.encode(), hashlib.sha256).hexdigest()[:32]
     return f"{token}.{sig}"
 
 
+def _prune_expired_sessions():
+    """Remove all expired session entries to prevent unbounded memory growth."""
+    now = time.time()
+    expired = [t for t, exp in _sessions.items() if now > exp]
+    if expired:
+        for token in expired:
+            _sessions.pop(token, None)
+        _save_sessions(_sessions)
+
+
 def verify_session(cookie_value) -> bool:
     """Verify a signed session cookie. Returns True if valid and not expired."""
     if not cookie_value or '.' not in cookie_value:
         return False
+    _prune_expired_sessions()  # lazy cleanup on every verification attempt
     token, sig = cookie_value.rsplit('.', 1)
-    expected_sig = hmac.new(_signing_key(), token.encode(), hashlib.sha256).hexdigest()[:16]
+    expected_sig = hmac.new(_signing_key(), token.encode(), hashlib.sha256).hexdigest()[:32]
     if not hmac.compare_digest(sig, expected_sig):
         return False
     expiry = _sessions.get(token)
@@ -107,7 +190,9 @@ def invalidate_session(cookie_value) -> None:
     """Remove a session token."""
     if cookie_value and '.' in cookie_value:
         token = cookie_value.rsplit('.', 1)[0]
-        _sessions.pop(token, None)
+        if token in _sessions:
+            _sessions.pop(token, None)
+            _save_sessions(_sessions)
 
 
 def parse_cookie(handler) -> str | None:
@@ -157,6 +242,9 @@ def set_auth_cookie(handler, cookie_value) -> None:
     cookie[COOKIE_NAME]['samesite'] = 'Lax'
     cookie[COOKIE_NAME]['path'] = '/'
     cookie[COOKIE_NAME]['max-age'] = str(SESSION_TTL)
+    # Set Secure flag when connection is HTTPS
+    if getattr(handler.request, 'getpeercert', None) is not None or handler.headers.get('X-Forwarded-Proto', '') == 'https':
+        cookie[COOKIE_NAME]['secure'] = True
     handler.send_header('Set-Cookie', cookie[COOKIE_NAME].OutputString())
 
 

api/background.py

@@ -0,0 +1,87 @@
+"""Background and ephemeral task tracking for /background and /btw commands."""
+from __future__ import annotations
+
+import logging
+import threading
+import time
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+_lock = threading.Lock()
+
+# parent_session_id -> list of task dicts
+_BACKGROUND_TASKS: dict[str, list[dict[str, Any]]] = {}
+
+# btw ephemeral session tracking: parent_sid -> {ephemeral_sid, stream_id, question}
+_BTW_TRACKING: dict[str, dict[str, Any]] = {}
+
+
+def track_background(parent_sid: str, bg_sid: str, stream_id: str,
+                     task_id: str, prompt: str) -> None:
+    with _lock:
+        _BACKGROUND_TASKS.setdefault(parent_sid, []).append({
+            "task_id": task_id,
+            "bg_session_id": bg_sid,
+            "stream_id": stream_id,
+            "prompt": prompt,
+            "status": "running",
+            "started_at": time.time(),
+            "answer": None,
+            "completed_at": None,
+        })
+
+
+def track_btw(parent_sid: str, ephemeral_sid: str, stream_id: str,
+              question: str) -> None:
+    with _lock:
+        _BTW_TRACKING[parent_sid] = {
+            "ephemeral_session_id": ephemeral_sid,
+            "stream_id": stream_id,
+            "question": question,
+        }
+
+
+def complete_background(parent_sid: str, task_id: str, answer: str) -> None:
+    with _lock:
+        for t in _BACKGROUND_TASKS.get(parent_sid, []):
+            if t["task_id"] == task_id and t["status"] == "running":
+                t["status"] = "done"
+                t["answer"] = answer
+                t["completed_at"] = time.time()
+                break
+
+
+def get_results(parent_sid: str) -> list[dict[str, Any]]:
+    """Return completed background task results and remove only the done ones
+    from tracking.  Tasks still in ``status="running"`` MUST stay in the list
+    so that ``complete_background()`` can still find them when the worker
+    thread finishes — otherwise the first poll during a long-running task
+    silently drops it and the result is lost forever.
+    """
+    with _lock:
+        tasks = _BACKGROUND_TASKS.get(parent_sid, [])
+        done = [t for t in tasks if t["status"] == "done"]
+        still_running = [t for t in tasks if t["status"] != "done"]
+        if still_running:
+            _BACKGROUND_TASKS[parent_sid] = still_running
+        else:
+            _BACKGROUND_TASKS.pop(parent_sid, None)
+        return [{
+            "task_id": t["task_id"],
+            "prompt": t["prompt"],
+            "answer": t["answer"],
+            "completed_at": t["completed_at"],
+        } for t in done]
+
+
+def get_background_tasks(parent_sid: str) -> list[dict[str, Any]]:
+    """Return all background tasks (running and done) for a parent session."""
+    with _lock:
+        return list(_BACKGROUND_TASKS.get(parent_sid, []))
+
+
+def cleanup_btw(parent_sid: str) -> dict[str, Any] | None:
+    """Remove and return btw tracking for a parent session."""
+    with _lock:
+        return _BTW_TRACKING.pop(parent_sid, None)

api/clarify.py

@@ -0,0 +1,128 @@
+"""Clarify prompt state for the WebUI.
+
+This mirrors the approval flow structure, but the response is a free-form
+clarification string instead of an approval decision.
+"""
+
+from __future__ import annotations
+
+import threading
+from typing import Optional
+
+
+_lock = threading.Lock()
+_pending: dict[str, dict] = {}
+_gateway_queues: dict[str, list] = {}
+_gateway_notify_cbs: dict[str, object] = {}
+
+
+class _ClarifyEntry:
+    """One pending clarify request inside a session."""
+
+    __slots__ = ("event", "data", "result")
+
+    def __init__(self, data: dict):
+        self.event = threading.Event()
+        self.data = data
+        self.result: Optional[str] = None
+
+
+def register_gateway_notify(session_key: str, cb) -> None:
+    """Register a per-session callback for sending clarify requests to the UI."""
+    with _lock:
+        _gateway_notify_cbs[session_key] = cb
+
+
+def _clear_queue_locked(session_key: str) -> list[_ClarifyEntry]:
+    entries = _gateway_queues.pop(session_key, [])
+    _pending.pop(session_key, None)
+    return entries
+
+
+def unregister_gateway_notify(session_key: str) -> None:
+    """Unregister the per-session callback and unblock any waiting clarify prompt."""
+    with _lock:
+        _gateway_notify_cbs.pop(session_key, None)
+        entries = _clear_queue_locked(session_key)
+    for entry in entries:
+        entry.event.set()
+
+
+def clear_pending(session_key: str) -> int:
+    """Clear any pending clarify prompts for the session without removing the callback."""
+    with _lock:
+        entries = _clear_queue_locked(session_key)
+    for entry in entries:
+        entry.event.set()
+    return len(entries)
+
+
+def submit_pending(session_key: str, data: dict) -> _ClarifyEntry:
+    """Queue a pending clarify request and notify the UI callback if registered."""
+    with _lock:
+        queue = _gateway_queues.setdefault(session_key, [])
+        # De-duplicate while unresolved: if the most recent pending clarify is
+        # semantically identical, reuse it instead of stacking duplicates.
+        if queue:
+            last = queue[-1]
+            if (
+                str(last.data.get("question", "")) == str(data.get("question", ""))
+                and list(last.data.get("choices_offered") or [])
+                == list(data.get("choices_offered") or [])
+            ):
+                entry = last
+                cb = _gateway_notify_cbs.get(session_key)
+                # Keep _pending aligned to the oldest unresolved entry.
+                _pending[session_key] = queue[0].data
+                if cb:
+                    try:
+                        cb(dict(entry.data))
+                    except Exception:
+                        pass
+                return entry
+
+        entry = _ClarifyEntry(data)
+        queue.append(entry)
+        _pending[session_key] = queue[0].data
+        cb = _gateway_notify_cbs.get(session_key)
+    if cb:
+        try:
+            cb(data)
+        except Exception:
+            pass
+    return entry
+
+
+def get_pending(session_key: str) -> dict | None:
+    """Return the oldest pending clarify request for this session, if any."""
+    with _lock:
+        queue = _gateway_queues.get(session_key) or []
+        if queue:
+            return dict(queue[0].data)
+        pending = _pending.get(session_key)
+        return dict(pending) if pending else None
+
+
+def has_pending(session_key: str) -> bool:
+    with _lock:
+        return bool(_gateway_queues.get(session_key))
+
+
+def resolve_clarify(session_key: str, response: str, resolve_all: bool = False) -> int:
+    """Resolve the oldest pending clarify request for a session."""
+    with _lock:
+        queue = _gateway_queues.get(session_key)
+        if not queue:
+            _pending.pop(session_key, None)
+            return 0
+        entries = list(queue) if resolve_all else [queue.pop(0)]
+        if queue:
+            _pending[session_key] = queue[0].data
+        else:
+            _clear_queue_locked(session_key)
+    count = 0
+    for entry in entries:
+        entry.result = response
+        entry.event.set()
+        count += 1
+    return count

api/commands.py

@@ -0,0 +1,56 @@
+"""Expose hermes-agent's COMMAND_REGISTRY to the webui frontend.
+
+This module is the single integration point with hermes_cli.commands.
+If hermes-agent is unavailable the endpoint degrades to an empty list
+so the frontend can still load with WEBUI_ONLY commands.
+"""
+from __future__ import annotations
+import logging
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+# Commands that are gateway_only in the agent registry -- webui never
+# wants to expose them (sethome, restart, update etc.) even if a future
+# agent version drops the gateway_only flag. /commands is the agent's
+# own command-listing command; webui has its own /help that calls
+# cmdHelp() locally, so /commands would be redundant and confusing.
+_NEVER_EXPOSE: frozenset[str] = frozenset({
+    'sethome', 'restart', 'update', 'commands',
+})
+
+
+def list_commands(_registry=None) -> list[dict[str, Any]]:
+    """Return COMMAND_REGISTRY entries as JSON-friendly dicts.
+
+    Returns empty list if hermes_cli is not installed (graceful
+    degradation -- the frontend has its own fallback minimum set).
+
+    Args:
+        _registry: Optional injected registry for testing. When None
+            (production), imports COMMAND_REGISTRY from hermes_cli.
+    """
+    if _registry is None:
+        try:
+            from hermes_cli.commands import COMMAND_REGISTRY as _registry
+        except ImportError:
+            logger.warning("hermes_cli.commands not importable -- /api/commands returns []")
+            return []
+
+    out: list[dict[str, Any]] = []
+    for cmd in _registry:
+        if cmd.gateway_only:
+            continue
+        if cmd.name in _NEVER_EXPOSE:
+            continue
+        out.append({
+            'name': cmd.name,
+            'description': cmd.description,
+            'category': cmd.category,
+            'aliases': list(cmd.aliases),
+            'args_hint': cmd.args_hint,
+            'subcommands': list(cmd.subcommands),
+            'cli_only': bool(cmd.cli_only),
+            'gateway_only': bool(cmd.gateway_only),
+        })
+    return out

api/gateway_watcher.py

@@ -0,0 +1,227 @@
+"""
+Hermes Web UI -- Gateway session watcher.
+
+Background daemon thread that polls state.db every 5 seconds for changes
+to gateway sessions (telegram, discord, slack, etc.). When changes are
+detected, it pushes notifications to all subscribed SSE clients.
+
+This enables real-time session list updates in the sidebar without
+requiring any changes to hermes-agent.
+"""
+import hashlib
+import json
+import logging
+import os
+import queue
+import threading
+import time
+from pathlib import Path
+
+from api.config import HOME
+from api.agent_sessions import read_importable_agent_session_rows
+
+logger = logging.getLogger(__name__)
+
+
+# ── State hash tracking ─────────────────────────────────────────────────────
+
+def _snapshot_hash(sessions: list) -> str:
+    """Create a lightweight hash of session IDs and timestamps for change detection."""
+    key = '|'.join(
+        f"{s['session_id']}:{s.get('updated_at', 0)}:{s.get('message_count', 0)}"
+        for s in sorted(sessions, key=lambda x: x['session_id'])
+    )
+    return hashlib.md5(key.encode(), usedforsecurity=False).hexdigest()
+
+
+# ── DB resolution (shared pattern with state_sync.py) ──────────────────────
+
+def _get_state_db_path() -> Path:
+    """Resolve state.db path for the active profile."""
+    try:
+        from api.profiles import get_active_hermes_home
+        hermes_home = Path(get_active_hermes_home()).expanduser().resolve()
+    except Exception:
+        hermes_home = Path(os.getenv('HERMES_HOME', str(HOME / '.hermes'))).expanduser().resolve()
+    return hermes_home / 'state.db'
+
+
+def _get_agent_sessions_from_db() -> list:
+    """Read all non-webui sessions from state.db.
+    Returns list of session dicts, or empty list on any error.
+    """
+    db_path = _get_state_db_path()
+    if not db_path.exists():
+        return []
+
+    try:
+        sessions = []
+        for row in read_importable_agent_session_rows(db_path, limit=200, log=logger):
+            sessions.append({
+                'session_id': row['id'],
+                'title': row['title'] or 'Agent Session',
+                'model': row['model'] or None,
+                'message_count': row['message_count'] or row['actual_message_count'] or 0,
+                'created_at': row['started_at'],
+                'updated_at': row['last_activity'] or row['started_at'],
+                'source': row['source'] or 'cli',
+            })
+        return sessions
+    except Exception:
+        return []
+
+
+# ── GatewayWatcher ──────────────────────────────────────────────────────────
+
+class GatewayWatcher:
+    """Background thread that polls state.db for agent session changes.
+
+    Usage:
+        watcher = GatewayWatcher()
+        watcher.start()
+        q = watcher.subscribe()
+        # ... receive change events via q.get() ...
+        watcher.unsubscribe(q)
+        watcher.stop()
+    """
+
+    POLL_INTERVAL = 5  # seconds between polls
+    SUBSCRIBER_TIMEOUT = 30  # seconds before sending keepalive comment
+
+    def __init__(self):
+        self._subscribers: list[queue.Queue] = []
+        self._sub_lock = threading.Lock()
+        self._stop_event = threading.Event()
+        self._thread: threading.Thread | None = None
+        self._last_hash: str = ''
+        self._last_sessions: list = []
+
+    def start(self):
+        """Start the watcher daemon thread."""
+        if self._thread and self._thread.is_alive():
+            return
+        self._stop_event.clear()
+        self._thread = threading.Thread(target=self._poll_loop, daemon=True, name='gateway-watcher')
+        self._thread.start()
+
+    def is_alive(self) -> bool:
+        """Return True when the poll thread is running.
+
+        Public accessor used by ``/api/sessions/gateway/stream`` probe mode and
+        the live SSE handler to detect a watcher instance whose poll thread
+        died silently (e.g. uncaught exception in ``_poll_loop``).  Callers
+        use this to decide whether to return 503 and trigger the client-side
+        polling fallback, instead of handing out an SSE connection that would
+        never emit events.
+        """
+        t = self._thread
+        return t is not None and t.is_alive()
+
+    def stop(self):
+        """Stop the watcher thread."""
+        self._stop_event.set()
+        # Wake up any subscribers
+        with self._sub_lock:
+            for q in self._subscribers:
+                try:
+                    q.put(None)  # sentinel
+                except Exception:
+                    logger.debug("Failed to send sentinel to subscriber")
+        if self._thread:
+            self._thread.join(timeout=3)
+            self._thread = None
+
+    def subscribe(self) -> queue.Queue:
+        """Subscribe to change events. Returns a queue.Queue.
+        Events are dicts: {'type': 'sessions_changed', 'sessions': [...]}
+        A None sentinel means the watcher is stopping.
+        """
+        q = queue.Queue(maxsize=10)
+        with self._sub_lock:
+            self._subscribers.append(q)
+        return q
+
+    def unsubscribe(self, q: queue.Queue):
+        """Remove a subscriber queue."""
+        with self._sub_lock:
+            try:
+                self._subscribers.remove(q)
+            except ValueError:
+                pass
+
+    def _notify_subscribers(self, sessions: list):
+        """Push change event to all subscribers."""
+        event = {
+            'type': 'sessions_changed',
+            'sessions': sessions,
+        }
+        with self._sub_lock:
+            dead = []
+            for q in self._subscribers:
+                try:
+                    q.put_nowait(event)
+                except queue.Full:
+                    dead.append(q)  # remove slow consumers
+                except Exception:
+                    dead.append(q)
+            for q in dead:
+                try:
+                    self._subscribers.remove(q)
+                except ValueError:
+                    pass
+                # Send a None sentinel so the SSE handler unblocks, closes,
+                # and lets the browser's EventSource auto-reconnect.
+                try:
+                    q.put_nowait(None)
+                except Exception:
+                    logger.debug("Failed to send sentinel to dead subscriber")
+
+    def _poll_loop(self):
+        """Main polling loop. Runs in a daemon thread."""
+        while not self._stop_event.is_set():
+            try:
+                sessions = _get_agent_sessions_from_db()
+                current_hash = _snapshot_hash(sessions)
+
+                if current_hash != self._last_hash:
+                    self._last_hash = current_hash
+                    self._last_sessions = sessions
+                    self._notify_subscribers(sessions)
+            except Exception:
+                logger.debug("Error in gateway watcher poll loop", exc_info=True)
+
+            # Sleep in small increments so we can stop promptly
+            for _ in range(self.POLL_INTERVAL * 10):
+                if self._stop_event.is_set():
+                    return
+                time.sleep(0.1)
+
+
+# ── Module-level singleton ─────────────────────────────────────────────────
+
+_watcher: GatewayWatcher | None = None
+_watcher_lock = threading.Lock()
+
+
+def start_watcher():
+    """Start the global gateway watcher (idempotent)."""
+    global _watcher
+    with _watcher_lock:
+        if _watcher is None:
+            _watcher = GatewayWatcher()
+            _watcher.start()
+
+
+def stop_watcher():
+    """Stop the global gateway watcher."""
+    global _watcher
+    with _watcher_lock:
+        if _watcher is not None:
+            _watcher.stop()
+            _watcher = None
+
+
+def get_watcher() -> GatewayWatcher | None:
+    """Get the global watcher instance (or None if not started)."""
+    with _watcher_lock:
+        return _watcher

api/helpers.py

@@ -2,6 +2,7 @@
 Hermes Web UI -- HTTP helper functions.
 """
 import json as _json
+import re as _re
 from pathlib import Path
 from api.config import IMAGE_EXTS, MD_EXTS
 
@@ -18,6 +19,15 @@ def bad(handler, msg, status: int=400):
     return j(handler, {'error': msg}, status=status)
 
 
+def _sanitize_error(e: Exception) -> str:
+    """Strip filesystem paths from exception messages before returning to client."""
+    import re
+    msg = str(e)
+    # Remove absolute paths (Unix and Windows)
+    msg = re.sub(r'(?:(?:/[a-zA-Z0-9_.-]+)+|(?:[A-Z]:\\[^\s]+))', '<path>', msg)
+    return msg
+
+
 def safe_resolve(root: Path, requested: str) -> Path:
     """Resolve a relative path inside root, raising ValueError on traversal."""
     resolved = (root / requested).resolve()
@@ -30,16 +40,54 @@ def _security_headers(handler):
     handler.send_header('X-Content-Type-Options', 'nosniff')
     handler.send_header('X-Frame-Options', 'DENY')
     handler.send_header('Referrer-Policy', 'same-origin')
+    handler.send_header(
+        'Content-Security-Policy',
+        "default-src 'self' https://*.cloudflareaccess.com; "
+        "script-src 'self' 'unsafe-inline' https://cdn.jsdelivr.net https://static.cloudflareinsights.com; "
+        "style-src 'self' 'unsafe-inline' https://cdn.jsdelivr.net; "
+        "img-src 'self' data: https: blob:; font-src 'self' data: https://cdn.jsdelivr.net; connect-src 'self'; "
+        "manifest-src 'self' https://*.cloudflareaccess.com; "
+        "base-uri 'self'; form-action 'self'"
+    )
+    handler.send_header(
+        'Permissions-Policy',
+        'camera=(), microphone=(self), geolocation=()'
+    )
 
 
-def j(handler, payload, status: int=200) -> None:
-    """Send a JSON response."""
+def _accepts_gzip(handler) -> bool:
+    """Check if the client accepts gzip encoding."""
+    headers = getattr(handler, 'headers', None)
+    if not headers:
+        return False
+    ae = headers.get('Accept-Encoding', '')
+    return 'gzip' in ae
+
+
+def j(handler, payload, status: int=200, extra_headers: dict=None) -> None:
+    """Send a JSON response.
+
+    *extra_headers*: optional dict of additional headers to include
+    (e.g., {'Set-Cookie': '...'}).  Headers are sent before end_headers().
+    """
     body = _json.dumps(payload, ensure_ascii=False, indent=2).encode('utf-8')
     handler.send_response(status)
     handler.send_header('Content-Type', 'application/json; charset=utf-8')
+
+    # Gzip-compress responses over 1KB when the client accepts it.
+    # Typical JSON API responses compress 70-80%, giving a big speedup
+    # for large payloads (session history, message lists).
+    if _accepts_gzip(handler) and len(body) > 1024:
+        import gzip
+        body = gzip.compress(body, compresslevel=4)
+        handler.send_header('Content-Encoding', 'gzip')
+
     handler.send_header('Content-Length', str(len(body)))
     handler.send_header('Cache-Control', 'no-store')
     _security_headers(handler)
+    if extra_headers:
+        for k, v in extra_headers.items():
+            handler.send_header(k, v)
     handler.end_headers()
     handler.wfile.write(body)
 
@@ -59,6 +107,88 @@ def t(handler, payload, status: int=200, content_type: str='text/plain; charset=
 MAX_BODY_BYTES = 20 * 1024 * 1024  # 20MB limit for non-upload POST bodies
 
 
+# ── Credential redaction ──────────────────────────────────────────────────────
+
+def _build_redact_fn():
+    """Return redact_sensitive_text from hermes-agent if available, else a fallback."""
+    try:
+        from agent.redact import redact_sensitive_text
+        return redact_sensitive_text
+    except ImportError:
+        pass
+
+    # Minimal fallback covering the most common credential prefixes
+    _CRED_RE = _re.compile(
+        r"(?<![A-Za-z0-9_-])("
+        r"sk-[A-Za-z0-9_-]{10,}"          # OpenAI / Anthropic / OpenRouter
+        r"|ghp_[A-Za-z0-9]{10,}"          # GitHub PAT (classic)
+        r"|github_pat_[A-Za-z0-9_]{10,}"  # GitHub PAT (fine-grained)
+        r"|gho_[A-Za-z0-9]{10,}"          # GitHub OAuth token
+        r"|ghu_[A-Za-z0-9]{10,}"          # GitHub user-to-server token
+        r"|ghs_[A-Za-z0-9]{10,}"          # GitHub server-to-server token
+        r"|ghr_[A-Za-z0-9]{10,}"          # GitHub refresh token
+        r"|AKIA[A-Z0-9]{16}"              # AWS Access Key ID
+        r"|xox[baprs]-[A-Za-z0-9-]{10,}" # Slack tokens
+        r"|hf_[A-Za-z0-9]{10,}"          # HuggingFace token
+        r"|SG\.[A-Za-z0-9_-]{10,}"       # SendGrid API key
+        r")(?![A-Za-z0-9_-])"
+    )
+    _AUTH_HDR_RE = _re.compile(r"(Authorization:\s*Bearer\s+)(\S+)", _re.IGNORECASE)
+    _ENV_RE = _re.compile(
+        r"([A-Z0-9_]{0,50}(?:API_?KEY|TOKEN|SECRET|PASSWORD|PASSWD|CREDENTIAL|AUTH)[A-Z0-9_]{0,50})"
+        r"\s*=\s*(['\"]?)(\S+)\2"
+    )
+    _PRIVKEY_RE = _re.compile(
+        r"-----BEGIN[A-Z ]*PRIVATE KEY-----[\s\S]*?-----END[A-Z ]*PRIVATE KEY-----"
+    )
+
+    def _mask(token: str) -> str:
+        return f"{token[:6]}...{token[-4:]}" if len(token) >= 18 else "***"
+
+    def _fallback_redact(text: str) -> str:
+        if not isinstance(text, str) or not text:
+            return text
+        text = _CRED_RE.sub(lambda m: _mask(m.group(1)), text)
+        text = _AUTH_HDR_RE.sub(lambda m: m.group(1) + _mask(m.group(2)), text)
+        text = _ENV_RE.sub(
+            lambda m: f"{m.group(1)}={m.group(2)}{_mask(m.group(3))}{m.group(2)}", text
+        )
+        text = _PRIVKEY_RE.sub("[REDACTED PRIVATE KEY]", text)
+        return text
+
+    return _fallback_redact
+
+
+_redact_text = _build_redact_fn()
+
+
+def _redact_value(v):
+    """Recursively redact credentials from strings, dicts, and lists."""
+    if isinstance(v, str):
+        return _redact_text(v)
+    if isinstance(v, dict):
+        return {k: _redact_value(val) for k, val in v.items()}
+    if isinstance(v, list):
+        return [_redact_value(item) for item in v]
+    return v
+
+
+def redact_session_data(session_dict: dict) -> dict:
+    """Redact credentials from message content and tool_call data before API response.
+
+    Applies to: messages[], tool_calls[], and title.
+    The underlying session file is not modified; redaction is response-layer only.
+    """
+    result = dict(session_dict)
+    if isinstance(result.get('title'), str):
+        result['title'] = _redact_text(result['title'])
+    if 'messages' in result:
+        result['messages'] = _redact_value(result['messages'])
+    if 'tool_calls' in result:
+        result['tool_calls'] = _redact_value(result['tool_calls'])
+    return result
+
+
 def read_body(handler) -> dict:
     """Read and JSON-parse a POST request body (capped at 20MB)."""
     length = int(handler.headers.get('Content-Length', 0))
@@ -69,3 +199,50 @@ def read_body(handler) -> dict:
         return _json.loads(raw)
     except Exception:
         return {}
+
+
+# ── Profile cookie helpers (issue #798) ─────────────────────────────────────
+
+PROFILE_COOKIE_NAME = 'hermes_profile'
+
+
+def get_profile_cookie(handler) -> str | None:
+    """Extract the hermes_profile cookie value from the request, or None."""
+    cookie_header = handler.headers.get('Cookie', '')
+    if not cookie_header:
+        return None
+    import http.cookies as _hc
+    cookie = _hc.SimpleCookie()
+    try:
+        cookie.load(cookie_header)
+    except _hc.CookieError:
+        return None
+    morsel = cookie.get(PROFILE_COOKIE_NAME)
+    if morsel and morsel.value:
+        # Validate against profile-name pattern before trusting
+        from api.profiles import _PROFILE_ID_RE
+        val = morsel.value
+        if val == 'default' or _PROFILE_ID_RE.fullmatch(val):
+            return val
+    return None
+
+
+def build_profile_cookie(name: str) -> str:
+    """Build a Set-Cookie header value for the hermes_profile cookie.
+
+    Always persist the selected profile in the cookie, including 'default'.
+    Clearing the cookie causes the backend to fall back to process-global
+    _active_profile, which can unexpectedly switch clients back to another
+    profile.
+
+    Set HttpOnly because the UI reads the active profile from
+    /api/profile/active JSON and does not need to access this cookie via
+    document.cookie.
+    """
+    import http.cookies as _hc
+    cookie = _hc.SimpleCookie()
+    cookie[PROFILE_COOKIE_NAME] = name
+    cookie[PROFILE_COOKIE_NAME]['path'] = '/'
+    cookie[PROFILE_COOKIE_NAME]['httponly'] = True
+    cookie[PROFILE_COOKIE_NAME]['samesite'] = 'Lax'
+    return cookie[PROFILE_COOKIE_NAME].OutputString()

api/metering.py

@@ -0,0 +1,187 @@
+"""
+Hermes Web UI -- Streaming performance metering.
+
+Tracks Tokens Per Second (TPS) across all active WebUI sessions, and the
+HIGH/LOW TPS values observed over the past 60 minutes.  Metering data is
+emitted via SSE events so the header label can update live during a stream.
+
+Architecture
+────────────
+Each streaming session is tracked independently.  TPS per session is:
+
+    session_tps = total_tokens / (last_token_ts - first_token_ts)
+
+The global tps is the average of all currently active sessions' TPS values.
+This correctly represents the system's real-time capacity regardless of how
+many sessions are running or how long each has been streaming.
+
+For HIGH/LOW tracking, every stats snapshot records the current global tps
+(only when > 0 — idle periods are skipped) into a rolling 60-minute history.
+The max/min of that history gives the peak throughput observed over the past hour.
+
+The ticker in streaming.py calls get_interval() — it returns 1.0 when sessions
+are actively receiving tokens so the header updates at 1 Hz, and 10.0 when idle
+so the ticker exits and no idle readings are emitted.
+
+Usage from api/streaming.py
+─────────────────────────────
+  from api.metering import meter
+
+  meter().begin_session(stream_id)                     # stream starts
+  meter().record_token(stream_id, running_output)     # per output token
+  meter().record_reasoning(stream_id, running_reasoning_len)  # per reasoning token
+
+The SSE `metering` event payload:
+  {
+    "tps": 47.3,    # average TPS across active sessions (real-time)
+    "high": 52.1,   # highest average TPS observed in the past 60 minutes
+    "low":  31.4,   # lowest average TPS (excl. readings < 1 tps, to ignore idle)
+    "active": 1,    # sessions currently streaming
+  }
+"""
+
+from __future__ import annotations
+
+import threading
+import time
+from dataclasses import dataclass
+
+_HOUR_SECS = 3600.0   # rolling window for HIGH/LOW tracking
+_STALE_SECS = 60.0    # consider a session inactive after this
+
+
+@dataclass
+class _SessionMeter:
+    output_tokens: int = 0
+    reasoning_tokens: int = 0
+    first_token_ts: float = 0.0   # time.monotonic() of first token received
+    last_token_ts: float = 0.0    # time.monotonic() of last token received
+
+    def total_tokens(self) -> int:
+        return self.output_tokens + self.reasoning_tokens
+
+    def tps(self) -> float:
+        if self.first_token_ts == 0.0 or self.last_token_ts <= self.first_token_ts:
+            return 0.0
+        return self.total_tokens() / (self.last_token_ts - self.first_token_ts)
+
+
+class GlobalMeter:
+    """Thread-safe global streaming meter.
+
+    Tracks per-session TPS, averages them for a global tps, and maintains a
+    60-minute rolling history of global tps snapshots for HIGH/LOW reporting.
+    """
+
+    __slots__ = (
+        '_lock',
+        '_sessions',        # stream_id -> _SessionMeter
+        '_readings',        # [(monotonic_ts, tps), ...] rolling 60-minute history
+        '_window_start',    # monotonic ts of current window
+    )
+
+    def __init__(self) -> None:
+        self._lock = threading.Lock()
+        self._sessions: dict[str, _SessionMeter] = {}
+        self._readings: list[tuple[float, float]] = []
+        self._window_start: float = time.monotonic()
+
+    # ── Public API ────────────────────────────────────────────────────────────
+
+    def begin_session(self, stream_id: str) -> None:
+        with self._lock:
+            self._sessions[stream_id] = _SessionMeter()
+
+    def get_interval(self) -> float:
+        """Return 1.0 when sessions are actively receiving tokens, 10.0 when idle.
+
+        Used by the streaming ticker to run at 1 Hz during work and exit when
+        there is nothing to measure.
+        """
+        now = time.monotonic()
+        with self._lock:
+            # Only count sessions that have received at least one token recently.
+            active_sids = {
+                sid for sid, s in self._sessions.items()
+                if s.first_token_ts > 0 and (now - s.last_token_ts) <= _STALE_SECS
+            }
+            return 1.0 if active_sids else 10.0
+
+    def record_token(self, stream_id: str, running_output_tokens: int) -> None:
+        now = time.monotonic()
+        with self._lock:
+            s = self._sessions.get(stream_id)
+            if s is None:
+                return
+            if s.first_token_ts == 0.0:
+                s.first_token_ts = now
+            s.last_token_ts = now
+            s.output_tokens = running_output_tokens
+
+    def record_reasoning(self, stream_id: str, running_reasoning_tokens: int) -> None:
+        now = time.monotonic()
+        with self._lock:
+            s = self._sessions.get(stream_id)
+            if s is None:
+                return
+            if s.first_token_ts == 0.0:
+                s.first_token_ts = now
+            s.last_token_ts = now
+            s.reasoning_tokens = running_reasoning_tokens
+
+    def end_session(self, stream_id: str, final_output_tokens: int, input_tokens: int = 0) -> None:
+        with self._lock:
+            self._sessions.pop(stream_id, None)
+
+    def get_stats(self) -> dict:
+        now = time.monotonic()
+        with self._lock:
+            # Prune stale sessions
+            stale = [
+                sid for sid, s in self._sessions.items()
+                if s.first_token_ts > 0 and (now - s.last_token_ts) > _STALE_SECS
+            ]
+            for sid in stale:
+                self._sessions.pop(sid, None)
+
+            # Reset window if everything went stale
+            if not self._sessions:
+                self._window_start = now
+
+            # Compute global tps: average of per-session TPS values
+            active = [s for s in self._sessions.values() if s.first_token_ts > 0]
+            if active:
+                global_tps = sum(s.tps() for s in active) / len(active)
+            else:
+                global_tps = 0.0
+
+            # Prune readings older than 1 hour
+            cutoff = now - _HOUR_SECS
+            self._readings = [(ts, v) for ts, v in self._readings if ts > cutoff]
+
+            # Only record this snapshot for HIGH/LOW if there is active work.
+            # This prevents idle periods from flooding the history and keeps
+            # HIGH/LOW meaningful for the past hour of actual throughput.
+            if global_tps > 0:
+                self._readings.append((now, global_tps))
+
+            # HIGH/LOW from the past hour (skip near-zero idle readings)
+            active_readings = [v for _, v in self._readings if v >= 1.0]
+            high = max(active_readings) if active_readings else 0.0
+            low = min(active_readings) if active_readings else 0.0
+
+            return {
+                'tps': round(global_tps, 1),
+                'high': round(high, 1),
+                'low': round(low, 1),
+                'active': len(self._sessions),
+            }
+
+
+# ── Module-level singleton ─────────────────────────────────────────────────────
+
+_meter = GlobalMeter()
+
+
+def meter() -> GlobalMeter:
+    return _meter

api/models.py

@@ -1,8 +1,9 @@
-"""
-Hermes Web UI -- Session model and in-memory session store.
-"""
+"""Hermes Web UI -- Session model and in-memory session store."""
 import collections
 import json
+import logging
+import os
+import threading
 import time
 import uuid
 from pathlib import Path
@@ -10,27 +11,296 @@ from pathlib import Path
 import api.config as _cfg
 from api.config import (
     SESSION_DIR, SESSION_INDEX_FILE, SESSIONS, SESSIONS_MAX,
-    LOCK, DEFAULT_WORKSPACE, DEFAULT_MODEL, PROJECTS_FILE, HOME
+    LOCK, STREAMS, STREAMS_LOCK, DEFAULT_WORKSPACE, DEFAULT_MODEL, PROJECTS_FILE, HOME,
+    get_effective_default_model,
 )
 from api.workspace import get_last_workspace
+from api.agent_sessions import read_importable_agent_session_rows
+
+logger = logging.getLogger(__name__)
+
+# ---------------------------------------------------------------------------
+# Stale temp-file cleanup
+# ---------------------------------------------------------------------------
+# Both Session.save() and _write_session_index() use the atomic-write pattern:
+#   write to  <path>.tmp.<pid>.<tid>  →  os.replace() to final path
+# If the process crashes between write and replace the .tmp file is left
+# behind.  Because the name embeds pid + tid, leftover files can never be
+# reused by a different process/thread, so they are safe to remove on the
+# next startup.  _cleanup_stale_tmp_files() is called from the full-rebuild
+# path of _write_session_index (i.e. at first index access / startup) and
+# removes any *.tmp.* file whose mtime is older than one hour.
+# ---------------------------------------------------------------------------
+
+_STALE_TMP_AGE_SECONDS = 3600  # 1 hour
+
+# Serializes index writers so concurrent Session.save() calls cannot race on
+# stale baselines while still allowing LOCK to be released before disk I/O.
+_INDEX_WRITE_LOCK = threading.RLock()
 
 
-def _write_session_index():
-    """Rebuild the session index file for O(1) future reads."""
-    entries = []
-    for p in SESSION_DIR.glob('*.json'):
-        if p.name.startswith('_'): continue
+def _cleanup_stale_tmp_files() -> None:
+    """Best-effort removal of stale ``*.tmp.*`` files from SESSION_DIR.
+
+    Only files whose mtime is older than ``_STALE_TMP_AGE_SECONDS`` are
+    removed so that in-flight writes from a long-running sibling process
+    are not disturbed.  Errors are logged and swallowed — this must never
+    prevent startup.
+    """
+    cutoff = time.time() - _STALE_TMP_AGE_SECONDS
+    try:
+        for p in SESSION_DIR.glob('*.tmp.*'):
+            try:
+                if p.stat().st_mtime < cutoff:
+                    p.unlink(missing_ok=True)
+                    logger.debug("Cleaned up stale tmp file: %s", p.name)
+            except OSError:
+                pass  # best-effort
+    except Exception:
+        pass  # SESSION_DIR may not exist yet; that's fine
+
+
+def _index_entry_exists(session_id: str, in_memory_ids=None) -> bool:
+    """Return True if an index entry still has backing state.
+
+    A session can legitimately exist either as a persisted JSON file or as an
+    in-memory Session object that has not been flushed yet.  This helper is used
+    to prune stale `_index.json` rows left behind after session-id rotation or
+    file removal.
+    """
+    if not session_id:
+        return False
+    if in_memory_ids is None:
+        with LOCK:
+            in_memory_ids = set(SESSIONS.keys())
+    if session_id in in_memory_ids:
+        return True
+    p = SESSION_DIR / f'{session_id}.json'
+    return p.exists()
+
+
+def _write_session_index(updates=None):
+    """Update the session index file.
+
+    When *updates* is provided (a list of Session objects whose compact
+    entries should be refreshed), this does a targeted in-place update of
+    the existing index — O(1) for single-session changes.  When *updates*
+    is None, a full rebuild is performed (used on startup / first call).
+
+    LOCK protects in-memory state snapshots and payload construction only;
+    disk I/O (write/flush/fsync/replace) always runs outside LOCK.
+    """
+    _tmp = SESSION_INDEX_FILE.with_suffix(f'.tmp.{os.getpid()}.{threading.current_thread().ident}')
+
+    with _INDEX_WRITE_LOCK:
+        # Lazy full-rebuild path — used when index doesn't exist yet.
+        if updates is None or not SESSION_INDEX_FILE.exists():
+            _cleanup_stale_tmp_files()  # best-effort sweep on startup / first call
+            entries = []
+            for p in SESSION_DIR.glob('*.json'):
+                if p.name.startswith('_'):
+                    continue
+                try:
+                    s = Session.load(p.stem)
+                    if s:
+                        entries.append(s.compact())
+                except Exception:
+                    logger.debug("Failed to load session from %s", p)
+
+            with LOCK:
+                existing_ids = {e.get('session_id') for e in entries}
+                for s in SESSIONS.values():
+                    if s.session_id not in existing_ids:
+                        entries.append(s.compact())
+                entries.sort(key=lambda s: s.get('updated_at', 0), reverse=True)
+                _payload = json.dumps(entries, ensure_ascii=False, indent=2)
+
+            try:
+                with open(_tmp, 'w', encoding='utf-8') as f:
+                    f.write(_payload)
+                    f.flush()
+                    os.fsync(f.fileno())
+                os.replace(_tmp, SESSION_INDEX_FILE)
+            except Exception:
+                # Best-effort cleanup of stale tmp on failure
+                try:
+                    _tmp.unlink(missing_ok=True)
+                except Exception:
+                    pass
+                raise
+            return
+
+        # Fast path: patch existing index with updated sessions.
+        # This avoids loading every session file on every single save().
+        _fallback = False
         try:
-            s = Session.load(p.stem)
-            if s: entries.append(s.compact())
+            with LOCK:
+                existing = json.loads(SESSION_INDEX_FILE.read_text(encoding='utf-8'))
+                in_memory_ids = set(SESSIONS.keys())
+
+                # Avoid N filesystem exists() checks under LOCK by collecting
+                # on-disk IDs once.
+                on_disk_ids = {
+                    p.stem
+                    for p in SESSION_DIR.glob('*.json')
+                    if not p.name.startswith('_')
+                }
+
+                existing = [
+                    e for e in existing
+                    if (e.get('session_id') in in_memory_ids or e.get('session_id') in on_disk_ids)
+                ]
+
+                # Build lookup of updated entries
+                updated_map = {s.session_id: s.compact() for s in updates}
+                existing_ids = {e.get('session_id') for e in existing}
+                # Add any updated entries not yet in the index
+                for sid, entry in updated_map.items():
+                    if sid not in existing_ids:
+                        existing.append(entry)
+                # Replace matching entries in-place
+                for i, e in enumerate(existing):
+                    sid = e.get('session_id')
+                    if sid in updated_map:
+                        existing[i] = updated_map[sid]
+                existing.sort(key=lambda s: s.get('updated_at', 0), reverse=True)
+                _payload = json.dumps(existing, ensure_ascii=False, indent=2)
+
+            try:
+                with open(_tmp, 'w', encoding='utf-8') as f:
+                    f.write(_payload)
+                    f.flush()
+                    os.fsync(f.fileno())
+                os.replace(_tmp, SESSION_INDEX_FILE)
+            except Exception:
+                try:
+                    _tmp.unlink(missing_ok=True)
+                except Exception:
+                    pass
+                raise
         except Exception:
-            pass
-    with LOCK:
-        for s in SESSIONS.values():
-            if not any(e['session_id'] == s.session_id for e in entries):
-                entries.append(s.compact())
-    entries.sort(key=lambda s: s['updated_at'], reverse=True)
-    SESSION_INDEX_FILE.write_text(json.dumps(entries, ensure_ascii=False, indent=2), encoding='utf-8')
+            _fallback = True
+
+    if _fallback:
+        # Corrupt or missing index — fall back to full rebuild (called outside LOCK to avoid deadlock)
+        _write_session_index(updates=None)
+
+
+def _active_stream_ids():
+    with STREAMS_LOCK:
+        return set(STREAMS.keys())
+
+
+def _is_streaming_session(active_stream_id, active_stream_ids):
+    return bool(active_stream_id and active_stream_id in active_stream_ids)
+
+def _session_sort_timestamp(session):
+    if isinstance(session, dict):
+        return session.get('last_message_at') or session.get('updated_at') or 0
+    return _last_message_timestamp(getattr(session, 'messages', None)) or getattr(session, 'updated_at', 0) or 0
+
+
+def _message_timestamp(message):
+    if not isinstance(message, dict):
+        return None
+    raw = message.get('_ts') or message.get('timestamp')
+    try:
+        return float(raw) if raw is not None else None
+    except (TypeError, ValueError):
+        return None
+
+
+def _last_message_timestamp(messages):
+    if not isinstance(messages, list):
+        return None
+    for message in reversed(messages):
+        if isinstance(message, dict) and message.get('role') == 'tool':
+            continue
+        ts = _message_timestamp(message)
+        if ts:
+            return ts
+    return None
+
+
+def _find_top_level_json_key(text, key):
+    """Return the byte offset of a top-level JSON object key, if present."""
+    depth = 0
+    i = 0
+    n = len(text)
+    while i < n:
+        ch = text[i]
+        if ch == '"':
+            start = i
+            i += 1
+            escaped = False
+            chars = []
+            while i < n:
+                c = text[i]
+                if escaped:
+                    chars.append(c)
+                    escaped = False
+                elif c == '\\':
+                    escaped = True
+                elif c == '"':
+                    break
+                else:
+                    chars.append(c)
+                i += 1
+            if i >= n:
+                return None
+            if depth == 1 and ''.join(chars) == key:
+                j = i + 1
+                while j < n and text[j] in ' \t\r\n':
+                    j += 1
+                if j < n and text[j] == ':':
+                    return start
+        elif ch in '{[':
+            depth += 1
+        elif ch in '}]':
+            depth -= 1
+        i += 1
+    return None
+
+
+def _read_metadata_json_prefix(path, max_prefix_bytes=65536):
+    """Read only the metadata portion before the top-level messages array."""
+    buf = ''
+    with open(path, 'r', encoding='utf-8') as f:
+        while len(buf.encode('utf-8')) < max_prefix_bytes:
+            chunk = f.read(4096)
+            if not chunk:
+                return None
+            buf += chunk
+            messages_pos = _find_top_level_json_key(buf, 'messages')
+            if messages_pos is None:
+                continue
+            prefix = buf[:messages_pos].rstrip()
+            if prefix.endswith(','):
+                prefix = prefix[:-1].rstrip()
+            return f'{prefix}\n}}'
+    return None
+
+
+def _lookup_index_message_count(session_id):
+    """Return the indexed message count without loading the full session file."""
+    try:
+        entries = json.loads(SESSION_INDEX_FILE.read_text(encoding='utf-8'))
+    except Exception:
+        return None
+    if not isinstance(entries, list):
+        return None
+    for entry in entries:
+        if entry.get('session_id') != session_id:
+            continue
+        count = entry.get('message_count')
+        if isinstance(count, int) and count >= 0:
+            return count
+        try:
+            count = int(count)
+        except (TypeError, ValueError):
+            return None
+        return count if count >= 0 else None
+    return None
 
 
 class Session:
@@ -40,6 +310,13 @@ class Session:
                  tool_calls=None, pinned: bool=False, archived: bool=False,
                  project_id: str=None, profile=None,
                  input_tokens: int=0, output_tokens: int=0, estimated_cost=None,
+                 personality=None,
+                 active_stream_id: str=None,
+                 pending_user_message: str=None,
+                 pending_attachments=None,
+                 pending_started_at=None,
+                 compression_anchor_visible_idx=None,
+                 compression_anchor_message_key=None,
                  **kwargs):
         self.session_id = session_id or uuid.uuid4().hex[:12]
         self.title = title
@@ -56,35 +333,113 @@ class Session:
         self.input_tokens = input_tokens or 0
         self.output_tokens = output_tokens or 0
         self.estimated_cost = estimated_cost
+        self.personality = personality
+        self.active_stream_id = active_stream_id
+        self.pending_user_message = pending_user_message
+        self.pending_attachments = pending_attachments or []
+        self.pending_started_at = pending_started_at
+        self.compression_anchor_visible_idx = compression_anchor_visible_idx
+        self.compression_anchor_message_key = compression_anchor_message_key
+        self._metadata_message_count = None
 
     @property
     def path(self):
         return SESSION_DIR / f'{self.session_id}.json'
 
-    def save(self) -> None:
-        self.updated_at = time.time()
-        self.path.write_text(
-            json.dumps(self.__dict__, ensure_ascii=False, indent=2),
-            encoding='utf-8',
-        )
-        _write_session_index()
+    def save(self, touch_updated_at: bool = True, skip_index: bool = False) -> None:
+        if touch_updated_at:
+            self.updated_at = time.time()
+        # Write metadata fields first so load_metadata_only() can read them
+        # without parsing the full messages array (which may be 400KB+).
+        # Fields are listed in the order they should appear in the JSON file.
+        METADATA_FIELDS = [
+            'session_id', 'title', 'workspace', 'model', 'created_at', 'updated_at',
+            'pinned', 'archived', 'project_id', 'profile',
+            'input_tokens', 'output_tokens', 'estimated_cost',
+            'personality', 'active_stream_id',
+            'pending_user_message', 'pending_attachments', 'pending_started_at',
+            'compression_anchor_visible_idx', 'compression_anchor_message_key',
+        ]
+        meta = {k: getattr(self, k, None) for k in METADATA_FIELDS}
+        meta['messages'] = self.messages
+        meta['tool_calls'] = self.tool_calls
+        # Fields not in METADATA_FIELDS (e.g. last_usage, message_count) go at the end
+        extra = {k: v for k, v in self.__dict__.items()
+                 if k not in METADATA_FIELDS and k not in ('messages', 'tool_calls')
+                 and not k.startswith('_')}
+        payload = json.dumps({**meta, **extra}, ensure_ascii=False, indent=2)
+        tmp = self.path.with_suffix(f'.tmp.{os.getpid()}.{threading.current_thread().ident}')
+        try:
+            with open(tmp, 'w', encoding='utf-8') as f:
+                f.write(payload)
+                f.flush()
+                os.fsync(f.fileno())
+            os.replace(tmp, self.path)
+        except Exception:
+            try:
+                tmp.unlink(missing_ok=True)
+            except Exception:
+                pass
+            raise
+        if not skip_index:
+            _write_session_index(updates=[self])
 
     @classmethod
     def load(cls, sid):
+        # Validate session ID format to prevent path traversal
+        if not sid or not all(c in '0123456789abcdefghijklmnopqrstuvwxyz_' for c in sid):
+            return None
         p = SESSION_DIR / f'{sid}.json'
         if not p.exists():
             return None
         return cls(**json.loads(p.read_text(encoding='utf-8')))
 
-    def compact(self) -> dict:
+    @classmethod
+    def load_metadata_only(cls, sid):
+        """Load only the compact metadata fields, skipping the messages array.
+
+        Session JSON files have metadata fields (session_id, title, model, etc.)
+        at the top level, before the large messages array. Read only up to the
+        top-level "messages" field and synthesize a small metadata-only object.
+        Falls back to load() for legacy or unexpected file layouts.
+        """
+        if not sid or not all(c in '0123456789abcdefghijklmnopqrstuvwxyz_' for c in sid):
+            return None
+        p = SESSION_DIR / f'{sid}.json'
+        if not p.exists():
+            return None
+        try:
+            prefix = _read_metadata_json_prefix(p)
+            if not prefix:
+                return cls.load(sid)
+            parsed = json.loads(prefix)
+            needed = {'session_id', 'title', 'created_at', 'updated_at'}
+            if not needed.issubset(parsed.keys()):
+                return cls.load(sid)
+            parsed['messages'] = []
+            parsed['tool_calls'] = []
+            session = cls(**parsed)
+            session._metadata_message_count = _lookup_index_message_count(sid)
+            return session
+        except Exception:
+            # Corrupt prefix or decode error — fall back to full load
+            return cls.load(sid)
+
+    def compact(self, include_runtime=False, active_stream_ids=None) -> dict:
+        active_stream_ids = active_stream_ids if active_stream_ids is not None else set()
         return {
             'session_id': self.session_id,
             'title': self.title,
             'workspace': self.workspace,
             'model': self.model,
-            'message_count': len(self.messages),
+            'message_count': (
+                self._metadata_message_count
+                if self._metadata_message_count is not None
+                else len(self.messages)
+            ),
             'created_at': self.created_at,
             'updated_at': self.updated_at,
+            'last_message_at': _last_message_timestamp(self.messages) or self.updated_at,
             'pinned': self.pinned,
             'archived': self.archived,
             'project_id': self.project_id,
@@ -92,14 +447,33 @@ class Session:
             'input_tokens': self.input_tokens,
             'output_tokens': self.output_tokens,
             'estimated_cost': self.estimated_cost,
+            'personality': self.personality,
+            'compression_anchor_visible_idx': self.compression_anchor_visible_idx,
+            'compression_anchor_message_key': self.compression_anchor_message_key,
+            'active_stream_id': self.active_stream_id,
+            'is_streaming': _is_streaming_session(
+                self.active_stream_id, active_stream_ids
+            ) if include_runtime else False,
         }
 
-def get_session(sid):
+def get_session(sid, metadata_only=False):
+    """Load a session, optionally with metadata only (skipping the messages array).
+
+    Metadata-only loads intentionally do not populate the full-session cache.
+    Otherwise a later full load could return a compact object with an empty
+    messages list. Use this when you only need compact() metadata and not the
+    actual message history (e.g., for fast sidebar switching).
+    """
     with LOCK:
         if sid in SESSIONS:
             SESSIONS.move_to_end(sid)  # LRU: mark as recently used
             return SESSIONS[sid]
-    s = Session.load(sid)
+    if metadata_only:
+        s = Session.load_metadata_only(sid)
+        if s:
+            return s
+    else:
+        s = Session.load(sid)
     if s:
         with LOCK:
             SESSIONS[sid] = s
@@ -109,14 +483,28 @@ def get_session(sid):
         return s
     raise KeyError(sid)
 
-def new_session(workspace=None, model=None):
-    # Use _cfg.DEFAULT_MODEL (not the import-time snapshot) so save_settings() changes take effect
-    try:
-        from api.profiles import get_active_profile_name
-        _profile = get_active_profile_name()
-    except ImportError:
-        _profile = None
-    s = Session(workspace=workspace or get_last_workspace(), model=model or _cfg.DEFAULT_MODEL, profile=_profile)
+def new_session(workspace=None, model=None, profile=None):
+    """Create a new in-memory session and persist it.
+
+    *profile* — when supplied by the caller (e.g. from the request body sent
+    by the active browser tab), it is used directly so that concurrent clients
+    on different profiles don't fight over a shared process-global.  If not
+    supplied, we fall back to the process-level active profile (the pre-#798
+    behaviour, preserved for calls that originate outside a request context).
+    """
+    if profile is None:
+        # Fallback: read process-level global (single-client or startup path)
+        try:
+            from api.profiles import get_active_profile_name
+            profile = get_active_profile_name()
+        except ImportError:
+            profile = None
+    effective_model = model or get_effective_default_model()
+    s = Session(
+        workspace=workspace or get_last_workspace(),
+        model=effective_model,
+        profile=profile,
+    )
     with LOCK:
         SESSIONS[s.session_id] = s
         SESSIONS.move_to_end(s.session_id)
@@ -126,18 +514,49 @@ def new_session(workspace=None, model=None):
     return s
 
 def all_sessions():
+    active_stream_ids = _active_stream_ids()
     # Phase C: try index first for O(1) read; fall back to full scan
     if SESSION_INDEX_FILE.exists():
         try:
             index = json.loads(SESSION_INDEX_FILE.read_text(encoding='utf-8'))
+            index = [
+                s for s in index
+                if _index_entry_exists(s.get('session_id'))
+            ]
+            backfilled = []
+            for i, s in enumerate(index):
+                if 'last_message_at' not in s:
+                    full = Session.load(s.get('session_id'))
+                    if full:
+                        index[i] = full.compact()
+                        backfilled.append(full)
+            if backfilled:
+                try:
+                    _write_session_index(updates=backfilled)
+                except Exception:
+                    logger.debug("Failed to persist last_message_at backfill")
+            for s in index:
+                s['is_streaming'] = _is_streaming_session(
+                    s.get('active_stream_id'),
+                    active_stream_ids,
+                )
             # Overlay any in-memory sessions that may be newer than the index
             index_map = {s['session_id']: s for s in index}
             with LOCK:
                 for s in SESSIONS.values():
-                    index_map[s.session_id] = s.compact()
-            result = sorted(index_map.values(), key=lambda s: (s.get('pinned', False), s['updated_at']), reverse=True)
+                    index_map[s.session_id] = s.compact(
+                        include_runtime=True,
+                        active_stream_ids=active_stream_ids,
+                    )
+            result = sorted(index_map.values(), key=lambda s: (s.get('pinned', False), _session_sort_timestamp(s)), reverse=True)
             # Hide empty Untitled sessions from the UI (created by tests, page refreshes, etc.)
-            result = [s for s in result if not (s.get('title','Untitled')=='Untitled' and s.get('message_count',0)==0)]
+            # Exempt sessions younger than 60 s so a brand-new session stays visible (#789)
+            _now = time.time()
+            result = [s for s in result if not (
+                s.get('title', 'Untitled') == 'Untitled'
+                and s.get('message_count', 0) == 0
+                and (_now - s.get('updated_at', _now)) > 60
+            )]
             # Backfill: sessions created before Sprint 22 have no profile tag.
             # Attribute them to 'default' so the client profile filter works correctly.
             for s in result:
@@ -145,7 +564,7 @@ def all_sessions():
                     s['profile'] = 'default'
             return result
         except Exception:
-            pass  # fall through to full scan
+            logger.debug("Failed to load session index, falling back to full scan")
     # Full scan fallback
     out = []
     for p in SESSION_DIR.glob('*.json'):
@@ -154,11 +573,16 @@ def all_sessions():
             s = Session.load(p.stem)
             if s: out.append(s)
         except Exception:
-            pass
+            logger.debug("Failed to load session from %s", p)
     for s in SESSIONS.values():
         if all(s.session_id != x.session_id for x in out): out.append(s)
-    out.sort(key=lambda s: (getattr(s, 'pinned', False), s.updated_at), reverse=True)
-    result = [s.compact() for s in out if not (s.title=='Untitled' and len(s.messages)==0)]
+    out.sort(key=lambda s: (getattr(s, 'pinned', False), _session_sort_timestamp(s)), reverse=True)
+    _now = time.time()
+    result = [s.compact(include_runtime=True, active_stream_ids=active_stream_ids) for s in out if not (
+        s.title == 'Untitled'
+        and len(s.messages) == 0
+        and (_now - s.updated_at) > 60
+    )]
     for s in result:
         if not s.get('profile'):
             s['profile'] = 'default'
@@ -194,7 +618,15 @@ def save_projects(projects) -> None:
     PROJECTS_FILE.write_text(json.dumps(projects, ensure_ascii=False, indent=2), encoding='utf-8')
 
 
-def import_cli_session(session_id: str, title: str, messages, model: str='unknown', profile=None):
+def import_cli_session(
+    session_id: str,
+    title: str,
+    messages,
+    model: str='unknown',
+    profile=None,
+    created_at=None,
+    updated_at=None,
+):
     """Create a new WebUI session populated with CLI messages.
     Returns the Session object.
     """
@@ -205,8 +637,10 @@ def import_cli_session(session_id: str, title: str, messages, model: str='unknow
         model=model,
         messages=messages,
         profile=profile,
+        created_at=created_at,
+        updated_at=updated_at,
     )
-    s.save()
+    s.save(touch_updated_at=False)
     return s
 
 
@@ -216,16 +650,11 @@ def get_cli_sessions() -> list:
     """Read CLI sessions from the agent's SQLite store and return them as
     dicts in a format the WebUI sidebar can render alongside local sessions.
 
-    Returns empty list if the SQLite DB is missing, the sqlite3 module is
-    unavailable, or any error occurs -- the bridge is purely additive and never
-    crashes the WebUI.
+    Returns empty list if the SQLite DB is missing or any error occurs -- the
+    bridge is purely additive and never crashes the WebUI.
     """
     import os
     cli_sessions = []
-    try:
-        import sqlite3
-    except ImportError:
-        return cli_sessions
 
     # Use the active WebUI profile's HERMES_HOME to find state.db.
     # The active profile is determined by what the user has selected in the UI
@@ -254,43 +683,56 @@ def get_cli_sessions() -> list:
         _cli_profile = None  # older agent -- fall back to no profile
 
     try:
-        with sqlite3.connect(str(db_path)) as conn:
-            conn.row_factory = sqlite3.Row
-            cur = conn.cursor()
-            cur.execute("""
-                SELECT s.id, s.title, s.model, s.message_count,
-                       s.started_at, s.source,
-                       MAX(m.timestamp) AS last_activity
-                FROM sessions s
-                LEFT JOIN messages m ON m.session_id = s.id
-                GROUP BY s.id
-                ORDER BY COALESCE(MAX(m.timestamp), s.started_at) DESC
-                LIMIT 200
-            """)
-            for row in cur.fetchall():
-                sid = row['id']
-                raw_ts = row['last_activity'] or row['started_at']
-                # Prefer the CLI session's own profile from the DB; fall back to
-                # the active CLI profile so sidebar filtering works either way.
-                profile = _cli_profile  # CLI DB has no profile column; use active profile
+        for row in read_importable_agent_session_rows(db_path, limit=200, log=logger):
+            sid = row['id']
+            raw_ts = row['last_activity'] or row['started_at']
+            # Prefer the CLI session's own profile from the DB; fall back to
+            # the active CLI profile so sidebar filtering works either way.
+            profile = _cli_profile  # CLI DB has no profile column; use active profile
 
-                cli_sessions.append({
-                    'session_id': sid,
-                    'title': row['title'] or 'CLI Session',
-                    'workspace': str(get_last_workspace()),
-                    'model': row['model'] or 'unknown',
-                    'message_count': row['message_count'] or 0,
-                    'created_at': row['started_at'],
-                    'updated_at': raw_ts,
-                    'pinned': False,
-                    'archived': False,
-                    'project_id': None,
-                    'profile': profile,
-                    'source_tag': 'cli',
-                    'is_cli_session': True,
-                })
-    except Exception:
-        # DB schema changed, locked, or corrupted -- silently degrade
+            _source = row['source'] or 'cli'
+            _title = row['title']
+            if not _title and _source == 'cron' and sid.startswith('cron_'):
+                # Extract job_id from session ID (cron_{job_id}_{timestamp})
+                # and look up the human-friendly job name from jobs.json
+                parts = sid.split('_')
+                if len(parts) >= 3:
+                    _job_id = parts[1]
+                    try:
+                        _jobs_path = hermes_home / 'cron' / 'jobs.json'
+                        if _jobs_path.exists():
+                            import json as _json
+                            _jobs_data = _json.loads(_jobs_path.read_text())
+                            for _j in _jobs_data.get('jobs', []):
+                                if _j.get('id') == _job_id:
+                                    _title = _j.get('name') or _title
+                                    break
+                    except Exception:
+                        pass  # degrade gracefully
+            _display_title = _title or f'{_source.title()} Session'
+            cli_sessions.append({
+                'session_id': sid,
+                'title': _display_title,
+                'workspace': str(get_last_workspace()),
+                'model': row['model'] or None,
+                'message_count': row['message_count'] or row['actual_message_count'] or 0,
+                'created_at': row['started_at'],
+                'updated_at': raw_ts,
+                'pinned': False,
+                'archived': False,
+                'project_id': None,
+                'profile': profile,
+                'source_tag': _source,
+                'is_cli_session': True,
+            })
+    except Exception as _cli_err:
+        # DB schema changed, locked, or corrupted -- log warning so admins can diagnose.
+        # Still degrade gracefully (don't crash the WebUI).
+        import logging as _logging
+        _logging.getLogger(__name__).warning(
+            "get_cli_sessions() failed — check state.db schema or path (%s): %s",
+            db_path, _cli_err,
+        )
         return []
 
     return cli_sessions

api/onboarding.py

@@ -0,0 +1,696 @@
+"""Hermes Web UI -- first-run onboarding helpers."""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+from urllib.parse import urlparse
+
+from api.auth import is_auth_enabled
+from api.config import (
+    DEFAULT_MODEL,
+    DEFAULT_WORKSPACE,
+    _FALLBACK_MODELS,
+    _HERMES_FOUND,
+    _PROVIDER_DISPLAY,
+    _PROVIDER_MODELS,
+    _get_config_path,
+    get_available_models,
+    get_config,
+    load_settings,
+    reload_config,
+    save_settings,
+    verify_hermes_imports,
+)
+from api.workspace import get_last_workspace, load_workspaces
+
+logger = logging.getLogger(__name__)
+
+
+_SUPPORTED_PROVIDER_SETUPS = {
+    # ── Easy start ──────────────────────────────────────────────────────
+    "openrouter": {
+        "label": "OpenRouter",
+        "env_var": "OPENROUTER_API_KEY",
+        "default_model": "anthropic/claude-sonnet-4.6",
+        "requires_base_url": False,
+        "models": [
+            {"id": model["id"], "label": model["label"]} for model in _FALLBACK_MODELS
+        ],
+        "category": "easy_start",
+        "quick": True,
+    },
+    "anthropic": {
+        "label": "Anthropic",
+        "env_var": "ANTHROPIC_API_KEY",
+        "default_model": "claude-sonnet-4.6",
+        "requires_base_url": False,
+        "models": list(_PROVIDER_MODELS.get("anthropic", [])),
+        "category": "easy_start",
+    },
+    "openai": {
+        "label": "OpenAI",
+        "env_var": "OPENAI_API_KEY",
+        "default_model": "gpt-4o",
+        "default_base_url": "https://api.openai.com/v1",
+        "requires_base_url": False,
+        "models": list(_PROVIDER_MODELS.get("openai", [])),
+        "category": "easy_start",
+    },
+    # ── Open / self-hosted ─────────────────────────────────────────────
+    "ollama": {
+        "label": "Ollama",
+        "env_var": "OLLAMA_API_KEY",
+        "default_model": "qwen3:32b",
+        "default_base_url": "http://localhost:11434/v1",
+        "requires_base_url": True,
+        "models": [],
+        "category": "self_hosted",
+    },
+    "lmstudio": {
+        "label": "LM Studio",
+        "env_var": "LMSTUDIO_API_KEY",
+        "default_model": "gpt-4o-mini",
+        "default_base_url": "http://localhost:1234/v1",
+        "requires_base_url": True,
+        "models": [],
+        "category": "self_hosted",
+    },
+    "custom": {
+        "label": "Custom OpenAI-compatible",
+        "env_var": "OPENAI_API_KEY",
+        "default_model": "gpt-4o-mini",
+        "requires_base_url": True,
+        "models": [],
+        "category": "self_hosted",
+    },
+    # ── Specialized / extended ──────────────────────────────────────────
+    "gemini": {
+        "label": "Google Gemini",
+        "env_var": "GOOGLE_API_KEY",
+        "default_model": "gemini-3.1-pro-preview",
+        "default_base_url": "https://generativelanguage.googleapis.com/v1beta/openai",
+        "requires_base_url": False,
+        # _PROVIDER_MODELS in api/config.py is keyed under "google" even though
+        # the agent's alias map normalizes "google" → "gemini".  Use the catalog
+        # key here so the wizard surfaces the actual model list.
+        "models": list(_PROVIDER_MODELS.get("google", [])),
+        "category": "specialized",
+    },
+    "deepseek": {
+        "label": "DeepSeek",
+        "env_var": "DEEPSEEK_API_KEY",
+        "default_model": "deepseek-chat-v3-0324",
+        "default_base_url": "https://api.deepseek.com/v1",
+        "requires_base_url": False,
+        "models": list(_PROVIDER_MODELS.get("deepseek", [])),
+        "category": "specialized",
+    },
+    "mistralai": {
+        "label": "Mistral",
+        "env_var": "MISTRAL_API_KEY",
+        "default_model": "mistral-large-latest",
+        "default_base_url": "https://api.mistral.ai/v1",
+        "requires_base_url": False,
+        # No catalog entry for mistralai today — wizard shows a free-text input.
+        "models": list(_PROVIDER_MODELS.get("mistralai", [])),
+        "category": "specialized",
+    },
+    "x-ai": {
+        "label": "xAI (Grok)",
+        "env_var": "XAI_API_KEY",
+        "default_model": "grok-4.20",
+        "default_base_url": "https://api.x.ai/v1",
+        "requires_base_url": False,
+        # Agent normalizes "x-ai" → "xai"; _PROVIDER_MODELS is also keyed "xai"
+        # when populated, so check both keys for forward-compatibility.
+        "models": list(_PROVIDER_MODELS.get("xai", []) or _PROVIDER_MODELS.get("x-ai", [])),
+        "category": "specialized",
+    },
+}
+
+_PROVIDER_CATEGORIES = [
+    {"id": "easy_start", "label": "Easy start", "order": 0},
+    {"id": "self_hosted", "label": "Open / self-hosted", "order": 1},
+    {"id": "specialized", "label": "Specialized", "order": 2},
+]
+
+_UNSUPPORTED_PROVIDER_NOTE = (
+    "OAuth and advanced provider flows such as Nous Portal, OpenAI Codex, and GitHub "
+    "Copilot are still terminal-first. Use `hermes model` for those flows."
+)
+
+
+def _get_active_hermes_home() -> Path:
+    try:
+        from api.profiles import get_active_hermes_home
+
+        return get_active_hermes_home()
+    except ImportError:
+        return Path.home() / ".hermes"
+
+
+def _load_env_file(env_path: Path) -> dict[str, str]:
+    values: dict[str, str] = {}
+    if not env_path.exists():
+        return values
+    try:
+        for raw in env_path.read_text(encoding="utf-8").splitlines():
+            line = raw.strip()
+            if not line or line.startswith("#") or "=" not in line:
+                continue
+            key, value = line.split("=", 1)
+            values[key.strip()] = value.strip().strip('"').strip("'")
+    except Exception:
+        return {}
+    return values
+
+
+def _write_env_file(env_path: Path, updates: dict[str, str]) -> None:
+    current = _load_env_file(env_path)
+    for key, value in updates.items():
+        if value is None:
+            current.pop(key, None)
+            os.environ.pop(key, None)
+            continue
+        clean = str(value).strip()
+        if not clean:
+            continue
+        # Reject embedded newlines/carriage returns to prevent .env injection
+        if "\n" in clean or "\r" in clean:
+            raise ValueError("API key must not contain newline characters.")
+        current[key] = clean
+        os.environ[key] = clean
+
+    env_path.parent.mkdir(parents=True, exist_ok=True)
+    lines = [f"{key}={current[key]}" for key in sorted(current)]
+    env_path.write_text("\n".join(lines) + ("\n" if lines else ""), encoding="utf-8")
+
+
+def _load_yaml_config(config_path: Path) -> dict:
+    try:
+        import yaml as _yaml
+    except ImportError:
+        return {}
+
+    if not config_path.exists():
+        return {}
+    try:
+        loaded = _yaml.safe_load(config_path.read_text(encoding="utf-8"))
+        return loaded if isinstance(loaded, dict) else {}
+    except Exception:
+        return {}
+
+
+def _save_yaml_config(config_path: Path, config: dict) -> None:
+    try:
+        import yaml as _yaml
+    except ImportError as exc:
+        raise RuntimeError("PyYAML is required to write Hermes config.yaml") from exc
+
+    config_path.parent.mkdir(parents=True, exist_ok=True)
+    config_path.write_text(
+        _yaml.safe_dump(config, sort_keys=False, allow_unicode=True),
+        encoding="utf-8",
+    )
+
+
+def _normalize_model_for_provider(provider: str, model: str) -> str:
+    clean = (model or "").strip()
+    if not clean:
+        return ""
+    if provider in {"anthropic", "openai"} and clean.startswith(provider + "/"):
+        return clean.split("/", 1)[1]
+    return clean
+
+
+def _normalize_base_url(base_url: str) -> str:
+    return (base_url or "").strip().rstrip("/")
+
+
+def _extract_current_provider(cfg: dict) -> str:
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, dict):
+        provider = str(model_cfg.get("provider") or "").strip().lower()
+        if provider:
+            return provider
+    return ""
+
+
+def _extract_current_model(cfg: dict) -> str:
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, str):
+        return model_cfg.strip()
+    if isinstance(model_cfg, dict):
+        return str(model_cfg.get("default") or "").strip()
+    return ""
+
+
+def _extract_current_base_url(cfg: dict) -> str:
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, dict):
+        return _normalize_base_url(str(model_cfg.get("base_url") or ""))
+    return ""
+
+
+def _provider_api_key_present(
+    provider: str, cfg: dict, env_values: dict[str, str]
+) -> bool:
+    provider = (provider or "").strip().lower()
+    if not provider:
+        return False
+
+    env_var = _SUPPORTED_PROVIDER_SETUPS.get(provider, {}).get("env_var")
+    if env_var and env_values.get(env_var):
+        return True
+
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, dict) and str(model_cfg.get("api_key") or "").strip():
+        return True
+
+    providers_cfg = cfg.get("providers", {})
+    if isinstance(providers_cfg, dict):
+        provider_cfg = providers_cfg.get(provider, {})
+        if (
+            isinstance(provider_cfg, dict)
+            and str(provider_cfg.get("api_key") or "").strip()
+        ):
+            return True
+        if provider == "custom":
+            custom_cfg = providers_cfg.get("custom", {})
+            if (
+                isinstance(custom_cfg, dict)
+                and str(custom_cfg.get("api_key") or "").strip()
+            ):
+                return True
+
+    # For providers not in _SUPPORTED_PROVIDER_SETUPS (e.g. minimax-cn, deepseek,
+    # xai, etc.), ask the hermes_cli auth registry — it knows every provider's env
+    # var names and can check os.environ for a valid key.
+    # Exclude known OAuth/token-flow providers — those are handled separately by
+    # _provider_oauth_authenticated() and should not be short-circuited here.
+    _known_oauth = {"openai-codex", "copilot", "copilot-acp", "qwen-oauth", "nous"}
+    if provider not in _SUPPORTED_PROVIDER_SETUPS and provider not in _known_oauth:
+        try:
+            from hermes_cli.auth import get_auth_status as _gas
+            status = _gas(provider)
+            if isinstance(status, dict) and status.get("logged_in"):
+                return True
+        except Exception:
+            pass
+
+    return False
+
+
+
+def _oauth_payload_has_token(payload: dict) -> bool:
+    """Return True if an auth payload contains usable token material."""
+    if not isinstance(payload, dict):
+        return False
+
+    token_fields = (
+        payload,
+        payload.get("tokens") if isinstance(payload.get("tokens"), dict) else {},
+    )
+    for candidate in token_fields:
+        if not isinstance(candidate, dict):
+            continue
+        if any(
+            str(candidate.get(key) or "").strip()
+            for key in ("access_token", "refresh_token", "api_key")
+        ):
+            return True
+    return False
+
+
+
+def _provider_oauth_authenticated(provider: str, hermes_home: "Path") -> bool:
+    """Return True if the provider has valid OAuth credentials.
+
+    Reads the profile-scoped auth.json directly so onboarding respects the
+    requested Hermes home. Known OAuth providers may store auth either in the
+    legacy providers[provider_id] singleton state or in credential_pool entries
+    used by current Hermes runtime auth resolution.
+    """
+    provider = (provider or "").strip().lower()
+    if not provider:
+        return False
+
+    _known_oauth_providers = {"openai-codex", "copilot", "copilot-acp", "qwen-oauth", "nous"}
+    if provider not in _known_oauth_providers:
+        return False
+
+    try:
+        import json as _j
+
+        auth_path = hermes_home / "auth.json"
+        if not auth_path.exists():
+            return False
+        store = _j.loads(auth_path.read_text(encoding="utf-8"))
+
+        providers_store = store.get("providers")
+        if isinstance(providers_store, dict):
+            state = providers_store.get(provider)
+            if _oauth_payload_has_token(state):
+                return True
+
+        pool_store = store.get("credential_pool")
+        if isinstance(pool_store, dict):
+            entries = pool_store.get(provider)
+            if isinstance(entries, list):
+                return any(_oauth_payload_has_token(entry) for entry in entries)
+
+        return False
+    except Exception:
+        return False
+
+
+def _status_from_runtime(cfg: dict, imports_ok: bool) -> dict:
+    provider = _extract_current_provider(cfg)
+    model = _extract_current_model(cfg)
+    base_url = _extract_current_base_url(cfg)
+    env_values = _load_env_file(_get_active_hermes_home() / ".env")
+
+    provider_configured = bool(provider and model)
+    provider_ready = False
+
+    if provider_configured:
+        if provider == "custom":
+            provider_ready = bool(
+                base_url and _provider_api_key_present(provider, cfg, env_values)
+            )
+        elif provider in _SUPPORTED_PROVIDER_SETUPS:
+            provider_ready = _provider_api_key_present(provider, cfg, env_values)
+        else:
+            # Unknown provider — may be an OAuth flow (openai-codex, copilot, etc.)
+            # OR an API-key provider not in the quick-setup list (minimax-cn, deepseek,
+            # xai, etc.).  Check both: api key presence first (covers the majority of
+            # third-party providers), then OAuth auth.json.
+            provider_ready = (
+                _provider_api_key_present(provider, cfg, env_values)
+                or _provider_oauth_authenticated(provider, _get_active_hermes_home())
+            )
+
+    chat_ready = bool(_HERMES_FOUND and imports_ok and provider_ready)
+
+    if not _HERMES_FOUND or not imports_ok:
+        state = "agent_unavailable"
+        note = (
+            "Hermes is not fully importable from the Web UI yet. Finish bootstrap or fix the "
+            "agent install before provider setup will work."
+        )
+    elif chat_ready:
+        state = "ready"
+        provider_name = _PROVIDER_DISPLAY.get(
+            provider, provider.title() if provider else "Hermes"
+        )
+        note = f"Hermes is minimally configured and ready to chat via {provider_name}."
+    elif provider_configured:
+        state = "provider_incomplete"
+        if provider == "custom" and not base_url:
+            note = (
+                "Hermes has a saved provider/model selection but still needs the "
+                "base URL and API key required to chat."
+            )
+        elif provider not in _SUPPORTED_PROVIDER_SETUPS:
+            # OAuth / unsupported provider: avoid misleading "API key" wording.
+            note = (
+                f"Provider '{provider}' is configured but not yet authenticated. "
+                "Run 'hermes auth' or 'hermes model' in a terminal to complete "
+                "setup, then reload the Web UI."
+            )
+        else:
+            note = (
+                "Hermes has a saved provider/model selection but still needs the "
+                "API key required to chat."
+            )
+    else:
+        state = "needs_provider"
+        note = "Hermes is installed, but you still need to choose a provider and save working credentials."
+
+    return {
+        "provider_configured": provider_configured,
+        "provider_ready": provider_ready,
+        "chat_ready": chat_ready,
+        "setup_state": state,
+        "provider_note": note,
+        "current_provider": provider or None,
+        "current_model": model or None,
+        "current_base_url": base_url or None,
+        "env_path": str(_get_active_hermes_home() / ".env"),
+    }
+
+
+def _build_setup_catalog(cfg: dict) -> dict:
+    current_provider = _extract_current_provider(cfg) or "openrouter"
+    current_model = _extract_current_model(cfg)
+    current_base_url = _extract_current_base_url(cfg)
+
+    providers = []
+    for provider_id, meta in _SUPPORTED_PROVIDER_SETUPS.items():
+        providers.append(
+            {
+                "id": provider_id,
+                "label": meta["label"],
+                "env_var": meta["env_var"],
+                "default_model": meta["default_model"],
+                "default_base_url": meta.get("default_base_url") or "",
+                "requires_base_url": bool(meta.get("requires_base_url")),
+                "models": list(meta.get("models", [])),
+                "category": meta.get("category", "easy_start"),
+                "quick": meta.get("quick", False),
+            }
+        )
+
+    # Sort providers by category order, then alphabetically within each category.
+    cat_order = {c["id"]: c["order"] for c in _PROVIDER_CATEGORIES}
+    providers.sort(key=lambda p: (cat_order.get(p["category"], 99), p["label"]))
+
+    # Group providers by category for the frontend.
+    categories = []
+    for cat in sorted(_PROVIDER_CATEGORIES, key=lambda c: c["order"]):
+        categories.append({
+            "id": cat["id"],
+            "label": cat["label"],
+            "providers": [p["id"] for p in providers if p["category"] == cat["id"]],
+        })
+
+    # Flag whether the currently-configured provider is OAuth-based (not in the
+    # API-key flow).  The frontend uses this to show a confirmation card instead
+    # of a key input when the user has already authenticated via 'hermes auth'.
+    current_is_oauth = current_provider not in _SUPPORTED_PROVIDER_SETUPS and bool(
+        current_provider
+    )
+
+    return {
+        "providers": providers,
+        "categories": categories,
+        "unsupported_note": _UNSUPPORTED_PROVIDER_NOTE,
+        "current_is_oauth": current_is_oauth,
+        "current": {
+            "provider": current_provider,
+            "model": current_model
+            or _SUPPORTED_PROVIDER_SETUPS.get(current_provider, {}).get(
+                "default_model", ""
+            ),
+            "base_url": current_base_url,
+        },
+    }
+
+
+def get_onboarding_status() -> dict:
+    settings = load_settings()
+    cfg = get_config()
+    imports_ok, missing, errors = verify_hermes_imports()
+    runtime = _status_from_runtime(cfg, imports_ok)
+    workspaces = load_workspaces()
+    last_workspace = get_last_workspace()
+    available_models = get_available_models()
+
+    # HERMES_WEBUI_SKIP_ONBOARDING=1 lets hosting providers (e.g. Agent37) ship
+    # a pre-configured instance without the wizard blocking the first load.
+    # This is an operator-level override and is honoured unconditionally —
+    # the operator knows their deployment is configured; we must not second-guess
+    # it by requiring chat_ready to also be true.
+    skip_env = os.environ.get("HERMES_WEBUI_SKIP_ONBOARDING", "").strip()
+    skip_requested = skip_env in {"1", "true", "yes"}
+    auto_completed = skip_requested  # unconditional: operator says skip, we skip
+
+    # Auto-complete for existing Hermes users: if config.yaml already exists
+    # AND the provider is configured (or the system is chat_ready), treat onboarding
+    # as done.  These users configured Hermes via the CLI before the Web UI existed;
+    # they must never be shown the first-run wizard — it would silently overwrite their
+    # config.  We use provider_configured (not chat_ready) so that users with
+    # non-wizard providers (ollama-cloud, deepseek, xai, kimi, etc.) are not forced
+    # through the wizard just because their provider doesn't have a detectable API key
+    # — the wizard cannot represent their provider and would overwrite their config
+    # with whichever wizard-supported provider they accidentally select.
+    config_exists = Path(_get_config_path()).exists()
+
+    # For providers not in the wizard's quick-setup list (e.g. ollama-cloud, deepseek,
+    # xai, kimi-k2.6), the wizard can never help — it only knows how to configure
+    # openrouter/anthropic/openai/google/custom.  If such a user has a configured
+    # provider + model in config.yaml, showing the wizard would only confuse them
+    # (or worse, let them accidentally overwrite their config with gpt-5.4-mini).
+    _current_provider = str(
+        (cfg.get("model", {}) or {}).get("provider", "") if isinstance(cfg.get("model"), dict)
+        else ""
+    ).strip().lower()
+    _is_non_wizard_provider = bool(
+        _current_provider and _current_provider not in _SUPPORTED_PROVIDER_SETUPS
+    )
+
+    config_auto_completed = config_exists and (
+        bool(runtime.get("chat_ready"))
+        or (_is_non_wizard_provider and bool(runtime.get("provider_configured")))
+    )
+
+    # Persist the flag so it survives future transient import failures (e.g. after
+    # a git branch switch in the hermes-agent repo).  Without this, a CLI-configured
+    # user who never ran the wizard has no onboarding_completed flag — any momentary
+    # imports_ok=False during restart makes chat_ready=False, config_auto_completed=False,
+    # and the wizard reappears with a broken dropdown that clobbers their config.
+    #
+    # Best-effort: if save_settings raises (read-only FS, disk full, permission error),
+    # log and continue.  The `config_auto_completed` branch of `completed=` below still
+    # returns True for this request, so the user sees the correct state — only the
+    # persistence-across-restart guarantee is degraded.  Raising here would turn every
+    # /api/onboarding/status call into a 500 until disk was writable, which is worse UX
+    # than losing the next-restart protection.
+    if config_auto_completed and not settings.get("onboarding_completed"):
+        try:
+            save_settings({"onboarding_completed": True})
+            settings["onboarding_completed"] = True
+        except Exception:
+            logger.debug("Failed to persist onboarding_completed", exc_info=True)
+
+    return {
+        "completed": bool(settings.get("onboarding_completed")) or auto_completed or config_auto_completed,
+        "settings": {
+            "default_model": settings.get("default_model") or DEFAULT_MODEL,
+            "default_workspace": settings.get("default_workspace")
+            or str(DEFAULT_WORKSPACE),
+            "password_enabled": is_auth_enabled(),
+            "bot_name": settings.get("bot_name") or "Hermes",
+        },
+        "system": {
+            "hermes_found": bool(_HERMES_FOUND),
+            "imports_ok": bool(imports_ok),
+            "missing_modules": missing,
+            "import_errors": errors,
+            "config_path": str(_get_config_path()),
+            "config_exists": Path(_get_config_path()).exists(),
+            **runtime,
+        },
+        "setup": _build_setup_catalog(cfg),
+        "workspaces": {
+            "items": workspaces,
+            "last": last_workspace,
+        },
+        "models": available_models,
+    }
+
+
+def apply_onboarding_setup(body: dict) -> dict:
+    # Hard guard: if the operator set SKIP_ONBOARDING, the wizard should never
+    # have appeared.  Even if the frontend somehow calls this endpoint anyway
+    # (e.g. a stale JS bundle or a curious user), we must not overwrite the
+    # operator's config.yaml or .env files.  Just mark onboarding complete and
+    # return the current status — no file writes.
+    skip_env = os.environ.get("HERMES_WEBUI_SKIP_ONBOARDING", "").strip()
+    if skip_env in {"1", "true", "yes"}:
+        save_settings({"onboarding_completed": True})
+        return get_onboarding_status()
+
+    provider = str(body.get("provider") or "").strip().lower()
+    model = str(body.get("model") or "").strip()
+    api_key = str(body.get("api_key") or "").strip()
+    base_url = _normalize_base_url(str(body.get("base_url") or ""))
+
+    if provider not in _SUPPORTED_PROVIDER_SETUPS:
+        # Unsupported providers (openai-codex, copilot, nous, etc.) are already
+        # configured via the CLI. Just mark onboarding as complete and let the
+        # user through — the agent is already set up, no further setup needed.
+        save_settings({"onboarding_completed": True})
+        return get_onboarding_status()
+    if not model:
+        raise ValueError("model is required")
+
+    provider_meta = _SUPPORTED_PROVIDER_SETUPS[provider]
+    if provider_meta.get("requires_base_url"):
+        if not base_url:
+            raise ValueError("base_url is required for custom endpoints")
+        parsed = urlparse(base_url)
+        if parsed.scheme not in {"http", "https"}:
+            raise ValueError("base_url must start with http:// or https://")
+
+    config_path = _get_config_path()
+    # Guard: if config.yaml already exists and the caller did not explicitly
+    # acknowledge the overwrite, refuse to proceed.  The frontend must pass
+    # confirm_overwrite=True after showing the user a confirmation step.
+    if Path(config_path).exists() and not body.get("confirm_overwrite"):
+        return {
+            "error": "config_exists",
+            "message": (
+                "Hermes is already configured (config.yaml exists). "
+                "Pass confirm_overwrite=true to overwrite it."
+            ),
+            "requires_confirm": True,
+        }
+
+    cfg = _load_yaml_config(config_path)
+    env_path = _get_active_hermes_home() / ".env"
+    env_values = _load_env_file(env_path)
+
+    if not api_key and not _provider_api_key_present(provider, cfg, env_values):
+        raise ValueError(f"{provider_meta['env_var']} is required")
+
+    model_cfg = cfg.get("model", {})
+    if not isinstance(model_cfg, dict):
+        model_cfg = {}
+
+    model_cfg["provider"] = provider
+    model_cfg["default"] = _normalize_model_for_provider(provider, model)
+
+    if provider_meta.get("requires_base_url"):
+        model_cfg["base_url"] = base_url
+    elif provider_meta.get("default_base_url"):
+        model_cfg["base_url"] = provider_meta["default_base_url"]
+    else:
+        model_cfg.pop("base_url", None)
+
+    cfg["model"] = model_cfg
+    _save_yaml_config(config_path, cfg)
+
+    if api_key:
+        _write_env_file(env_path, {provider_meta["env_var"]: api_key})
+
+    # Reload the hermes_cli provider/config cache so the next streaming call
+    # picks up the new key without requiring a server restart.
+    try:
+        from api.profiles import _reload_dotenv
+        _reload_dotenv(_get_active_hermes_home())
+    except Exception:
+        logger.debug("Failed to reload dotenv")
+
+    # Belt-and-braces: set directly on os.environ AFTER _reload_dotenv so the
+    # value survives even if _reload_dotenv cleared it (e.g. when _write_env_file
+    # wrote to disk but the profile isolation tracking hasn't seen it yet).
+    if api_key:
+        os.environ[provider_meta["env_var"]] = api_key
+
+    try:
+        # hermes_cli may cache config at import time; ask it to reload if possible.
+        from hermes_cli.config import reload as _cli_reload
+        _cli_reload()
+    except Exception:
+        logger.debug("Failed to reload hermes_cli config")
+
+    reload_config()
+    return get_onboarding_status()
+
+
+def complete_onboarding() -> dict:
+    save_settings({"onboarding_completed": True})
+    return get_onboarding_status()

api/profiles.py

@@ -9,12 +9,15 @@ cached paths in hermes-agent modules (skills_tool, cron/jobs) that snapshot
 HERMES_HOME at import time.
 """
 import json
+import logging
 import os
 import re
 import shutil
 import threading
 from pathlib import Path
 
+logger = logging.getLogger(__name__)
+
 # ── Constants (match hermes_cli.profiles upstream) ─────────────────────────
 _PROFILE_ID_RE = re.compile(r'^[a-z0-9][a-z0-9_-]{0,63}$')
 _PROFILE_DIRS = [
@@ -26,6 +29,13 @@ _CLONE_CONFIG_FILES = ['config.yaml', '.env', 'SOUL.md']
 # ── Module state ────────────────────────────────────────────────────────────
 _active_profile = 'default'
 _profile_lock = threading.Lock()
+_loaded_profile_env_keys: set[str] = set()
+
+# Thread-local profile context: set per-request by server.py, cleared after.
+# Enables per-client profile isolation (issue #798) — each HTTP request thread
+# reads its own profile from the hermes_profile cookie instead of the
+# process-global _active_profile.
+_tls = threading.local()
 
 def _resolve_base_hermes_home() -> Path:
     """Return the BASE ~/.hermes directory — the root that contains profiles/.
@@ -71,26 +81,78 @@ def _read_active_profile_file() -> str:
     ap_file = _DEFAULT_HERMES_HOME / 'active_profile'
     if ap_file.exists():
         try:
-            name = ap_file.read_text().strip()
+            name = ap_file.read_text(encoding="utf-8").strip()
             if name:
                 return name
         except Exception:
-            pass
+            logger.debug("Failed to read active profile file")
     return 'default'
 
 
 # ── Public API ──────────────────────────────────────────────────────────────
 
 def get_active_profile_name() -> str:
-    """Return the currently active profile name."""
+    """Return the currently active profile name.
+
+    Priority:
+      1. Thread-local (set per-request from hermes_profile cookie) — issue #798
+      2. Process-level default (_active_profile)
+    """
+    tls_name = getattr(_tls, 'profile', None)
+    if tls_name is not None:
+        return tls_name
     return _active_profile
 
 
+def set_request_profile(name: str) -> None:
+    """Set the per-request profile context for this thread.
+
+    Called by server.py at the start of each request when a hermes_profile
+    cookie is present.  Always paired with clear_request_profile() in a
+    finally block so the thread-local is released after the request.
+    """
+    _tls.profile = name
+
+
+def clear_request_profile() -> None:
+    """Clear the per-request profile context for this thread.
+
+    Called by server.py in the finally block of do_GET / do_POST.
+    Safe to call even if set_request_profile() was never called.
+    """
+    _tls.profile = None
+
+
 def get_active_hermes_home() -> Path:
-    """Return the HERMES_HOME path for the currently active profile."""
-    if _active_profile == 'default':
+    """Return the HERMES_HOME path for the currently active profile.
+
+    Uses get_active_profile_name() so per-request TLS context (issue #798)
+    is respected, not just the process-level global.
+    """
+    name = get_active_profile_name()
+    if name == 'default':
         return _DEFAULT_HERMES_HOME
-    profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / _active_profile
+    profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / name
+    if profile_dir.is_dir():
+        return profile_dir
+    return _DEFAULT_HERMES_HOME
+
+
+
+def get_hermes_home_for_profile(name: str) -> Path:
+    """Return the HERMES_HOME Path for *name* without mutating any process state.
+
+    Safe to call from per-request context (streaming, session creation) because
+    it reads only the filesystem — it never touches os.environ, module-level
+    cached paths, or the process-level _active_profile global.
+
+    Falls back to _DEFAULT_HERMES_HOME (same as 'default') when *name* is None,
+    empty, 'default', or does not match the profile-name format (rejects path
+    traversal such as '../../etc').
+    """
+    if not name or name == 'default' or not _PROFILE_ID_RE.match(name):
+        return _DEFAULT_HERMES_HOME
+    profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / name
     if profile_dir.is_dir():
         return profile_dir
     return _DEFAULT_HERMES_HOME
@@ -106,7 +168,7 @@ def _set_hermes_home(home: Path):
         _sk.HERMES_HOME = home
         _sk.SKILLS_DIR = home / 'skills'
     except (ImportError, AttributeError):
-        pass
+        logger.debug("Failed to patch skills_tool module")
 
     # Patch cron/jobs module-level cache
     try:
@@ -116,16 +178,29 @@ def _set_hermes_home(home: Path):
         _cj.JOBS_FILE = _cj.CRON_DIR / 'jobs.json'
         _cj.OUTPUT_DIR = _cj.CRON_DIR / 'output'
     except (ImportError, AttributeError):
-        pass
+        logger.debug("Failed to patch cron.jobs module")
 
 
 def _reload_dotenv(home: Path):
-    """Load .env from the profile dir into os.environ (additive)."""
+    """Load .env from the profile dir into os.environ with profile isolation.
+
+    Clears env vars that were loaded from the previously active profile before
+    applying the current profile's .env. This prevents API keys and other
+    profile-scoped secrets from leaking across profile switches.
+    """
+    global _loaded_profile_env_keys
+
+    # Remove keys loaded from the previous profile first.
+    for key in list(_loaded_profile_env_keys):
+        os.environ.pop(key, None)
+    _loaded_profile_env_keys = set()
+
     env_path = home / '.env'
     if not env_path.exists():
         return
     try:
-        for line in env_path.read_text().splitlines():
+        loaded_keys: set[str] = set()
+        for line in env_path.read_text(encoding="utf-8").splitlines():
             line = line.strip()
             if line and not line.startswith('#') and '=' in line:
                 k, v = line.split('=', 1)
@@ -133,8 +208,11 @@ def _reload_dotenv(home: Path):
                 v = v.strip().strip('"').strip("'")
                 if k and v:
                     os.environ[k] = v
+                    loaded_keys.add(k)
+        _loaded_profile_env_keys = loaded_keys
     except Exception:
-        pass
+        _loaded_profile_env_keys = set()
+        logger.debug("Failed to reload dotenv from %s", env_path)
 
 
 def init_profile_state() -> None:
@@ -150,12 +228,18 @@ def init_profile_state() -> None:
     _reload_dotenv(home)
 
 
-def switch_profile(name: str) -> dict:
+def switch_profile(name: str, *, process_wide: bool = True) -> dict:
     """Switch the active profile.
 
     Validates the profile exists, updates process state, patches module caches,
     reloads .env, and reloads config.yaml.
 
+    Args:
+        name: Profile name to switch to.
+        process_wide: If True (default), updates the process-global
+            _active_profile.  Set to False for per-client switches from the
+            WebUI where the profile is managed via cookie + thread-local (#798).
+
     Returns: {'profiles': [...], 'active': name}
     Raises ValueError if profile doesn't exist or agent is busy.
     """
@@ -176,29 +260,46 @@ def switch_profile(name: str) -> dict:
     if name == 'default':
         home = _DEFAULT_HERMES_HOME
     else:
-        home = _DEFAULT_HERMES_HOME / 'profiles' / name
+        home = _resolve_named_profile_home(name)
         if not home.is_dir():
             raise ValueError(f"Profile '{name}' does not exist.")
 
     with _profile_lock:
-        _active_profile = name
-        _set_hermes_home(home)
-        _reload_dotenv(home)
+        if process_wide:
+            global _active_profile
+            _active_profile = name
+            _set_hermes_home(home)
+            _reload_dotenv(home)
 
-    # Write sticky default for CLI consistency
-    try:
-        ap_file = _DEFAULT_HERMES_HOME / 'active_profile'
-        ap_file.write_text(name if name != 'default' else '')
-    except Exception:
-        pass
+    if process_wide:
+        # Write sticky default for CLI consistency
+        try:
+            ap_file = _DEFAULT_HERMES_HOME / 'active_profile'
+            ap_file.write_text(name if name != 'default' else '', encoding='utf-8')
+        except Exception:
+            logger.debug("Failed to write active profile file")
 
-    # Reload config.yaml from the new profile
-    reload_config()
+        # Reload config.yaml from the new profile
+        reload_config()
 
-    # Return profile-specific defaults so frontend can apply them
+    # Return profile-specific defaults so frontend can apply them.
+    # For process_wide=False (per-client switch), read the target profile's
+    # config.yaml directly from disk rather than from _cfg_cache (process-global),
+    # since reload_config() was intentionally skipped.
     from api.workspace import get_last_workspace
-    from api.config import get_config
-    cfg = get_config()
+    if process_wide:
+        from api.config import get_config
+        cfg = get_config()
+    else:
+        # Direct disk read — does not touch _cfg_cache
+        try:
+            import yaml as _yaml
+            cfg_path = home / 'config.yaml'
+            cfg = _yaml.safe_load(cfg_path.read_text(encoding='utf-8')) if cfg_path.exists() else {}
+            if not isinstance(cfg, dict):
+                cfg = {}
+        except Exception:
+            cfg = {}
     model_cfg = cfg.get('model', {})
     default_model = None
     if isinstance(model_cfg, str):
@@ -223,7 +324,7 @@ def list_profiles_api() -> list:
         # hermes_cli not available -- return just the default
         return [_default_profile_dict()]
 
-    active = _active_profile
+    active = get_active_profile_name()
     result = []
     for p in infos:
         result.append({
@@ -267,6 +368,24 @@ def _validate_profile_name(name: str):
         )
 
 
+def _profiles_root() -> Path:
+    """Return the canonical root that contains named profiles."""
+    return (_DEFAULT_HERMES_HOME / 'profiles').resolve()
+
+
+def _resolve_named_profile_home(name: str) -> Path:
+    """Resolve a named profile to a directory under the profiles root.
+
+    Validates *name* as a logical profile identifier first, then resolves the
+    final filesystem path and enforces containment under ~/.hermes/profiles.
+    """
+    _validate_profile_name(name)
+    profiles_root = _profiles_root()
+    candidate = (profiles_root / name).resolve()
+    candidate.relative_to(profiles_root)
+    return candidate
+
+
 def _create_profile_fallback(name: str, clone_from: str = None,
                               clone_config: bool = False) -> Path:
     """Create a profile directory without hermes_cli (Docker/standalone fallback)."""
@@ -294,8 +413,38 @@ def _create_profile_fallback(name: str, clone_from: str = None,
     return profile_dir
 
 
+def _write_endpoint_to_config(profile_dir: Path, base_url: str = None, api_key: str = None) -> None:
+    """Write custom endpoint fields into config.yaml for a profile."""
+    if not base_url and not api_key:
+        return
+    config_path = profile_dir / 'config.yaml'
+    try:
+        import yaml as _yaml
+    except ImportError:
+        return
+    cfg = {}
+    if config_path.exists():
+        try:
+            loaded = _yaml.safe_load(config_path.read_text(encoding="utf-8"))
+            if isinstance(loaded, dict):
+                cfg = loaded
+        except Exception:
+            logger.debug("Failed to load config from %s", config_path)
+    model_section = cfg.get('model', {})
+    if not isinstance(model_section, dict):
+        model_section = {}
+    if base_url:
+        model_section['base_url'] = base_url
+    if api_key:
+        model_section['api_key'] = api_key
+    cfg['model'] = model_section
+    config_path.write_text(_yaml.dump(cfg, default_flow_style=False, allow_unicode=True), encoding='utf-8')
+
+
 def create_profile_api(name: str, clone_from: str = None,
-                       clone_config: bool = False) -> dict:
+                       clone_config: bool = False,
+                       base_url: str = None,
+                       api_key: str = None) -> dict:
     """Create a new profile. Returns the new profile info dict."""
     _validate_profile_name(name)
     # Defense-in-depth: validate clone_from here too, even though routes.py
@@ -315,11 +464,26 @@ def create_profile_api(name: str, clone_from: str = None,
     except ImportError:
         _create_profile_fallback(name, clone_from, clone_config)
 
+    # Resolve the profile directory from the profile list when possible.
+    # hermes_cli and the webui runtime do not always agree on the exact root,
+    # so we prefer the path returned by list_profiles_api() and fall back to the
+    # standard profile location only if the profile cannot be found there yet.
+    profile_path = _DEFAULT_HERMES_HOME / 'profiles' / name
+    for p in list_profiles_api():
+        if p['name'] == name:
+            try:
+                profile_path = Path(p.get('path') or profile_path)
+            except Exception:
+                logger.debug("Failed to parse profile path")
+            break
+
+    profile_path.mkdir(parents=True, exist_ok=True)
+    _write_endpoint_to_config(profile_path, base_url=base_url, api_key=api_key)
+
     # Find and return the newly created profile info.
     # When hermes_cli is not importable, list_profiles_api() also falls back
     # to the stub default-only list and won't find the new profile by name.
     # In that case, return a complete profile dict directly.
-    profile_path = _DEFAULT_HERMES_HOME / 'profiles' / name
     for p in list_profiles_api():
         if p['name'] == name:
             return p
@@ -340,6 +504,7 @@ def delete_profile_api(name: str) -> dict:
     """Delete a profile. Switches to default first if it's the active one."""
     if name == 'default':
         raise ValueError("Cannot delete the default profile.")
+    _validate_profile_name(name)
 
     # If deleting the active profile, switch to default first
     if _active_profile == name:
@@ -357,7 +522,7 @@ def delete_profile_api(name: str) -> dict:
     except ImportError:
         # Manual fallback: just remove the directory
         import shutil
-        profile_dir = _DEFAULT_HERMES_HOME / 'profiles' / name
+        profile_dir = _resolve_named_profile_home(name)
         if profile_dir.is_dir():
             shutil.rmtree(str(profile_dir))
         else:

api/providers.py

@@ -0,0 +1,331 @@
+"""Hermes Web UI -- provider management endpoints.
+
+Provides CRUD operations for configuring provider API keys post-onboarding.
+Closes #586 (allow provider key update) and part of #604 (model picker
+multi-provider support).
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+from pathlib import Path
+from typing import Any
+
+from api.config import (
+    _PROVIDER_DISPLAY,
+    _PROVIDER_MODELS,
+    get_config,
+    invalidate_models_cache,
+)
+
+logger = logging.getLogger(__name__)
+
+# SECTION: Provider ↔ env var mapping
+
+# Maps canonical provider slug → env var name for API key.
+# Providers not listed here (OAuth/token-flow providers like copilot, nous,
+# openai-codex) cannot have their keys managed from the WebUI.
+_PROVIDER_ENV_VAR: dict[str, str] = {
+    "openrouter": "OPENROUTER_API_KEY",
+    "anthropic": "ANTHROPIC_API_KEY",
+    "openai": "OPENAI_API_KEY",
+    "google": "GOOGLE_API_KEY",
+    "gemini": "GEMINI_API_KEY",
+    "zai": "GLM_API_KEY",
+    "kimi-coding": "KIMI_API_KEY",
+    "deepseek": "DEEPSEEK_API_KEY",
+    "minimax": "MINIMAX_API_KEY",
+    "mistralai": "MISTRAL_API_KEY",
+    "x-ai": "XAI_API_KEY",
+    "opencode-zen": "OPENCODE_ZEN_API_KEY",
+    "opencode-go": "OPENCODE_GO_API_KEY",
+    "ollama": "OLLAMA_API_KEY",
+    "ollama-cloud": "OLLAMA_API_KEY",
+}
+
+# Providers that use OAuth or token flows — their credentials are managed
+# through the Hermes CLI, not via API keys.  The WebUI cannot set these.
+_OAUTH_PROVIDERS = frozenset({
+    "copilot",
+    "openai-codex",
+    "nous",
+})
+
+# SECTION: Helper functions
+
+
+def _get_hermes_home() -> Path:
+    """Return the active Hermes home directory."""
+    try:
+        from api.profiles import get_active_hermes_home
+        return get_active_hermes_home()
+    except ImportError:
+        return Path.home() / ".hermes"
+
+
+def _load_env_file(env_path: Path) -> dict[str, str]:
+    """Read key=value pairs from a .env file."""
+    values: dict[str, str] = {}
+    if not env_path.exists():
+        return values
+    try:
+        for raw in env_path.read_text(encoding="utf-8").splitlines():
+            line = raw.strip()
+            if not line or line.startswith("#") or "=" not in line:
+                continue
+            key, value = line.split("=", 1)
+            values[key.strip()] = value.strip().strip('"').strip("'")
+    except Exception:
+        return {}
+    return values
+
+
+def _write_env_file(env_path: Path, updates: dict[str, str | None]) -> None:
+    """Write key=value pairs to the .env file.
+
+    Values of ``None`` cause the key to be removed.
+    Holds ``_ENV_LOCK`` from ``api.streaming`` for the entire load → modify →
+    write cycle to prevent TOCTOU races between concurrent POST /api/providers
+    calls (each reading the same file baseline and overwriting the other's key).
+    Also serialises os.environ mutations with streaming sessions.
+    """
+    from api.streaming import _ENV_LOCK
+    import stat as _stat
+
+    with _ENV_LOCK:
+        current = _load_env_file(env_path)
+        for key, value in updates.items():
+            if value is None:
+                current.pop(key, None)
+                os.environ.pop(key, None)
+                continue
+            clean = str(value).strip()
+            if not clean:
+                continue
+            # Reject embedded newlines/carriage returns to prevent .env injection
+            if "\n" in clean or "\r" in clean:
+                raise ValueError("API key must not contain newline characters.")
+            current[key] = clean
+            os.environ[key] = clean
+
+        env_path.parent.mkdir(parents=True, exist_ok=True)
+        lines = [f"{key}={current[key]}" for key in sorted(current)]
+        # Create at owner-only mode from the first byte (O_CREAT honours the mode
+        # argument subject to umask). A trailing chmod guards pre-existing files.
+        _mode = _stat.S_IRUSR | _stat.S_IWUSR  # 0o600
+        _fd = os.open(str(env_path), os.O_WRONLY | os.O_CREAT | os.O_TRUNC, _mode)
+        with os.fdopen(_fd, "w", encoding="utf-8") as _f:
+            _f.write("\n".join(lines) + ("\n" if lines else ""))
+        try:
+            env_path.chmod(_mode)
+        except OSError:
+            pass
+
+
+def _provider_has_key(provider_id: str) -> bool:
+    """Check whether a provider has a configured API key.
+
+    Checks (in order):
+    1. ``~/.hermes/.env`` for the known env var
+    2. ``os.environ`` for the known env var
+    3. ``config.yaml → model.api_key``
+    4. ``config.yaml → providers.<id>.api_key``
+    5. ``config.yaml → custom_providers[].api_key`` (for custom providers)
+    """
+    env_var = _PROVIDER_ENV_VAR.get(provider_id)
+    if env_var:
+        env_path = _get_hermes_home() / ".env"
+        env_values = _load_env_file(env_path)
+        if env_values.get(env_var):
+            return True
+        if os.getenv(env_var):
+            return True
+
+    cfg = get_config()
+    # Check model.api_key
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, dict) and str(model_cfg.get("api_key") or "").strip():
+        return True
+    # Check providers.<id>.api_key
+    providers_cfg = cfg.get("providers", {})
+    if isinstance(providers_cfg, dict):
+        provider_cfg = providers_cfg.get(provider_id, {})
+        if isinstance(provider_cfg, dict) and str(provider_cfg.get("api_key") or "").strip():
+            return True
+    # Check custom_providers
+    custom_providers = cfg.get("custom_providers", [])
+    if isinstance(custom_providers, list):
+        for cp in custom_providers:
+            if isinstance(cp, dict):
+                cp_name = (cp.get("name") or "").strip().lower().replace(" ", "-")
+                if f"custom:{cp_name}" == provider_id or cp.get("name", "").strip().lower() == provider_id:
+                    if str(cp.get("api_key") or "").strip():
+                        return True
+    return False
+
+
+def _provider_is_oauth(provider_id: str) -> bool:
+    """Check whether a provider uses OAuth/token flows (managed by CLI)."""
+    return provider_id in _OAUTH_PROVIDERS
+
+
+# SECTION: Public API
+
+
+def get_providers() -> dict[str, Any]:
+    """Return a list of all known providers with their configuration status.
+
+    Each entry contains:
+    - ``id``: canonical provider slug
+    - ``display_name``: human-readable name
+    - ``has_key``: whether an API key is configured
+    - ``configurable``: whether the key can be set from the WebUI
+    - ``key_source``: where the key was found (``env_file``, ``env_var``,
+      ``config_yaml``, ``oauth``, ``none``)
+    - ``models``: list of known model IDs for this provider
+    """
+    providers = []
+
+    # Collect all known provider IDs from multiple sources
+    known_ids = set(_PROVIDER_DISPLAY.keys()) | set(_PROVIDER_MODELS.keys())
+
+    # Also detect providers from config.yaml providers section
+    cfg = get_config()
+    providers_cfg = cfg.get("providers", {})
+    if isinstance(providers_cfg, dict):
+        known_ids.update(providers_cfg.keys())
+
+    # Add OAuth providers even if not in _PROVIDER_DISPLAY
+    known_ids.update(_OAUTH_PROVIDERS)
+
+    for pid in sorted(known_ids):
+        display_name = _PROVIDER_DISPLAY.get(pid, pid.replace("-", " ").title())
+        is_oauth = _provider_is_oauth(pid)
+        has_key = _provider_has_key(pid)
+
+        # Determine key source
+        key_source = "none"
+        if is_oauth:
+            key_source = "oauth"
+            # Check if actually authenticated via hermes_cli
+            try:
+                from hermes_cli.auth import get_auth_status as _gas
+                status = _gas(pid)
+                if isinstance(status, dict) and status.get("logged_in"):
+                    has_key = True
+                    key_source = status.get("key_source", "oauth")
+                else:
+                    has_key = False
+            except Exception:
+                has_key = False
+        elif has_key:
+            env_var = _PROVIDER_ENV_VAR.get(pid)
+            if env_var:
+                env_path = _get_hermes_home() / ".env"
+                env_values = _load_env_file(env_path)
+                if env_values.get(env_var):
+                    key_source = "env_file"
+                elif os.getenv(env_var):
+                    key_source = "env_var"
+                else:
+                    key_source = "config_yaml"
+            else:
+                key_source = "config_yaml"
+
+        models = _PROVIDER_MODELS.get(pid, [])
+        # Also include models from config.yaml providers section
+        if isinstance(providers_cfg, dict):
+            provider_cfg = providers_cfg.get(pid, {})
+            if isinstance(provider_cfg, dict) and "models" in provider_cfg:
+                cfg_models = provider_cfg["models"]
+                if isinstance(cfg_models, dict):
+                    models = models + [{"id": k, "label": k} for k in cfg_models.keys()]
+                elif isinstance(cfg_models, list):
+                    models = models + [{"id": k, "label": k} for k in cfg_models]
+
+        providers.append({
+            "id": pid,
+            "display_name": display_name,
+            "has_key": has_key,
+            "configurable": not is_oauth and pid in _PROVIDER_ENV_VAR,
+            "key_source": key_source,
+            "models": models,
+        })
+
+    # Determine active provider
+    active_provider = None
+    model_cfg = cfg.get("model", {})
+    if isinstance(model_cfg, dict):
+        active_provider = model_cfg.get("provider")
+
+    return {
+        "providers": providers,
+        "active_provider": active_provider,
+    }
+
+
+def set_provider_key(provider_id: str, api_key: str | None) -> dict[str, Any]:
+    """Set or update the API key for a provider.
+
+    Writes the key to ``~/.hermes/.env`` using the standard env var name.
+    If ``api_key`` is None or empty, the key is removed.
+
+    Returns a status dict with the operation result.
+    """
+    provider_id = provider_id.strip().lower()
+
+    if not provider_id:
+        return {"ok": False, "error": "Provider ID is required."}
+
+    if _provider_is_oauth(provider_id):
+        return {
+            "ok": False,
+            "error": f"'{_PROVIDER_DISPLAY.get(provider_id, provider_id)}' uses OAuth authentication. "
+                     f"Use `hermes model` in the terminal to configure it.",
+        }
+
+    env_var = _PROVIDER_ENV_VAR.get(provider_id)
+    if not env_var:
+        return {
+            "ok": False,
+            "error": f"Cannot configure API key for '{_PROVIDER_DISPLAY.get(provider_id, provider_id)}'. "
+                     f"This provider does not have a known env var mapping.",
+        }
+
+    # Validate API key format (basic sanity check)
+    if api_key:
+        api_key = api_key.strip()
+        if "\n" in api_key or "\r" in api_key:
+            return {"ok": False, "error": "API key must not contain newline characters."}
+        if len(api_key) < 8:
+            return {"ok": False, "error": "API key appears too short."}
+
+    env_path = _get_hermes_home() / ".env"
+    try:
+        _write_env_file(env_path, {env_var: api_key})
+    except ValueError as exc:
+        return {"ok": False, "error": str(exc)}
+    except Exception as exc:
+        logger.exception("Failed to write env file for provider %s", provider_id)
+        return {"ok": False, "error": f"Failed to save API key: {exc}"}
+
+    # Invalidate the model cache so the dropdown refreshes on next request.
+    # Using invalidate_models_cache() instead of reload_config() to avoid
+    # disrupting active streaming sessions that may be reading config.cfg.
+    invalidate_models_cache()
+
+    return {
+        "ok": True,
+        "provider": provider_id,
+        "display_name": _PROVIDER_DISPLAY.get(provider_id, provider_id),
+        "action": "updated" if api_key else "removed",
+    }
+
+
+def remove_provider_key(provider_id: str) -> dict[str, Any]:
+    """Remove the API key for a provider.
+
+    Convenience wrapper around ``set_provider_key(id, None)``.
+    """
+    return set_provider_key(provider_id, None)

api/session_ops.py

@@ -0,0 +1,161 @@
+"""Session-mutation operations for slash commands (/retry, /undo) and
+read-only aggregators (/status, /usage). Operates on the webui's own
+JSON Session store (api/models.py), not on hermes-agent's SQLite.
+
+Behavior parity reference: gateway/run.py:_handle_*_command in
+the hermes-agent repo.
+"""
+from __future__ import annotations
+import logging
+from typing import Any
+
+from api.config import LOCK, _get_session_agent_lock
+from api.models import get_session, SESSIONS
+
+logger = logging.getLogger(__name__)
+
+
+def retry_last(session_id: str) -> dict[str, Any]:
+    """Truncate the session to before the last user message, return its text.
+
+    Mirrors gateway/run.py:_handle_retry_command. Caller (webui frontend)
+    is expected to put the returned text back in the composer and call
+    send() to resume the conversation -- the agent's gateway calls its own
+    _handle_message; the webui has no equivalent in-process pipeline.
+
+    Raises:
+        KeyError: session not found
+        ValueError: no user message in transcript
+    """
+    # Acquire the per-session agent lock as the outermost lock so that the
+    # read-modify-write of s.messages is serialised with the periodic
+    # checkpoint thread, cancel_stream, and all other session writers.
+    # Lock ordering: _agent_lock → LOCK → _write_session_index (LOCK).
+    with _get_session_agent_lock(session_id):
+        # get_session() and Session.save() both acquire the module-level LOCK
+        # internally (the latter via _write_session_index()), and LOCK is a
+        # non-reentrant threading.Lock — so they MUST be called outside our
+        # own `with LOCK:` block to avoid self-deadlocking.
+        #
+        # The race we close is the read-modify-write of s.messages: two
+        # concurrent /api/session/retry calls could otherwise both compute the
+        # same last_user_idx from the same history and double-truncate. We
+        # serialize just the in-memory mutation; persistence happens inside
+        # the per-session lock so the checkpoint thread cannot race us.
+        #
+        # Stale-object guard: on a cache miss, two concurrent get_session()
+        # calls can each load and cache a *different* Session instance for the
+        # same session_id (the second store clobbers the first). Re-bind to
+        # the canonical cached instance inside the lock so the mutation lands
+        # on the object the next reader will see, not a stale parallel copy.
+        s = get_session(session_id)  # raises KeyError if missing
+        with LOCK:
+            s = SESSIONS.get(session_id, s)
+            history = s.messages or []
+            last_user_idx = None
+            for i in range(len(history) - 1, -1, -1):
+                if history[i].get('role') == 'user':
+                    last_user_idx = i
+                    break
+            if last_user_idx is None:
+                raise ValueError('No previous message to retry.')
+
+            last_user_text = _extract_text(history[last_user_idx].get('content', ''))
+            removed_count = len(history) - last_user_idx
+            s.messages = history[:last_user_idx]
+        s.save()
+    return {'last_user_text': last_user_text, 'removed_count': removed_count}
+
+
+def undo_last(session_id: str) -> dict[str, Any]:
+    """Remove the most recent user message and everything after it.
+
+    Mirrors gateway/run.py:_handle_undo_command. Returns a preview of the
+    removed text so the UI can confirm to the user.
+
+    Raises:
+        KeyError: session not found
+        ValueError: no user message in transcript
+    """
+    # Acquire the per-session agent lock as the outermost lock so that the
+    # read-modify-write of s.messages is serialised with the periodic
+    # checkpoint thread, cancel_stream, and all other session writers.
+    # Lock ordering: _agent_lock → LOCK → _write_session_index (LOCK).
+    with _get_session_agent_lock(session_id):
+        s = get_session(session_id)  # acquires LOCK transiently
+        with LOCK:
+            # Stale-object guard — see retry_last for the rationale.
+            s = SESSIONS.get(session_id, s)
+            history = s.messages or []
+            last_user_idx = None
+            for i in range(len(history) - 1, -1, -1):
+                if history[i].get('role') == 'user':
+                    last_user_idx = i
+                    break
+            if last_user_idx is None:
+                raise ValueError('Nothing to undo.')
+
+            removed_text = _extract_text(history[last_user_idx].get('content', ''))
+            removed_count = len(history) - last_user_idx
+            s.messages = history[:last_user_idx]
+        s.save()  # outside LOCK -- save() re-acquires LOCK via _write_session_index()
+    preview = (removed_text[:40] + '...') if len(removed_text) > 40 else removed_text
+    return {
+        'removed_count': removed_count,
+        'removed_preview': preview,
+    }
+
+
+def session_status(session_id: str) -> dict[str, Any]:
+    """Return a snapshot of session state for /status.
+
+    Webui equivalent of gateway/run.py:_handle_status_command. The agent's
+    "agent_running" comes from `session_key in self._running_agents`; the
+    webui equivalent is whether the session has an active stream
+    (active_stream_id is set).
+    """
+    s = get_session(session_id)
+    return {
+        'session_id': s.session_id,
+        'title': s.title,
+        'model': s.model,
+        'workspace': s.workspace,
+        'personality': s.personality,
+        'message_count': len(s.messages or []),
+        'created_at': s.created_at,
+        'updated_at': s.updated_at,
+        'agent_running': bool(getattr(s, 'active_stream_id', None)),
+    }
+
+
+def session_usage(session_id: str) -> dict[str, Any]:
+    """Return token usage and cost for /usage.
+
+    Mirrors gateway/run.py:_handle_usage_command's basic counters. The
+    agent shows additional fields (rate-limit headroom etc.) that depend
+    on provider API responses we don't have in webui -- those are deferred.
+    """
+    s = get_session(session_id)
+    inp = int(s.input_tokens or 0)
+    out = int(s.output_tokens or 0)
+    return {
+        'input_tokens': inp,
+        'output_tokens': out,
+        'total_tokens': inp + out,
+        'estimated_cost': s.estimated_cost,
+        'model': s.model,
+    }
+
+
+def _extract_text(content: Any) -> str:
+    """Flatten message content to plain text. Agent stores either a string
+    or a list of {type, text|...} parts; webui needs the user-typed text."""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = []
+        for p in content:
+            if isinstance(p, dict) and p.get('type') == 'text':
+                parts.append(p.get('text', ''))
+        return ' '.join(parts)
+    return str(content)

api/startup.py

@@ -0,0 +1,104 @@
+"""Hermes Web UI -- startup helpers."""
+from __future__ import annotations
+import os, stat, subprocess, sys
+from pathlib import Path
+
+# Credential files that should never be world-readable
+_SENSITIVE_FILES = (
+    '.env',
+    'google_token.json',
+    'google_client_secret.json',
+    '.signing_key',
+    'auth.json',
+)
+
+
+def fix_credential_permissions() -> None:
+    """Ensure sensitive files in HERMES_HOME are chmod 600 (owner-only)."""
+    hermes_home = Path(os.environ.get('HERMES_HOME', str(Path.home() / '.hermes')))
+    if not hermes_home.is_dir():
+        return
+    for name in _SENSITIVE_FILES:
+        fpath = hermes_home / name
+        if not fpath.exists():
+            continue
+        try:
+            current = stat.S_IMODE(fpath.stat().st_mode)
+            if current & 0o077:  # group or other bits set
+                fpath.chmod(0o600)
+                print(f'  [security] fixed permissions on {fpath.name} ({oct(current)} -> 0600)', flush=True)
+        except OSError:
+            pass  # best-effort; don't abort startup
+
+
+def _agent_dir() -> Path | None:
+    hermes_home = Path(os.environ.get('HERMES_HOME', str(Path.home() / '.hermes')))
+    for raw in [os.environ.get('HERMES_WEBUI_AGENT_DIR', '').strip(), str(hermes_home / 'hermes-agent')]:
+        if not raw:
+            continue
+        p = Path(raw).expanduser()
+        if p.is_dir():
+            return p.resolve()
+    return None
+
+def _trusted_agent_dir(agent_dir: Path) -> bool:
+    """Return True if agent_dir passes ownership and permission checks.
+
+    Validates that the directory is not world- or group-writable and,
+    on POSIX systems, is owned by the current process user.
+
+    Intentionally does NOT enforce a canonical path (i.e. does not require
+    the dir to be ~/.hermes/hermes-agent), so custom HERMES_WEBUI_AGENT_DIR
+    paths work correctly when HERMES_WEBUI_AUTO_INSTALL=1 is set.
+    """
+    try:
+        st = agent_dir.stat()
+        if stat.S_IMODE(st.st_mode) & 0o022:
+            # World- or group-writable — untrusted
+            return False
+        if hasattr(os, 'getuid') and st.st_uid != os.getuid():
+            # Not owned by current user (POSIX only; Windows fallback skips)
+            return False
+        return True
+    except OSError:
+        return False
+
+
+def auto_install_agent_deps() -> bool:
+    enabled = os.environ.get('HERMES_WEBUI_AUTO_INSTALL', '').strip().lower() in ('1', 'true', 'yes')
+    if not enabled:
+        print('[!!] Auto-install disabled. Set HERMES_WEBUI_AUTO_INSTALL=1 to enable.', flush=True)
+        return False
+    agent_dir = _agent_dir()
+    if agent_dir is None:
+        print('[!!] Auto-install skipped: agent directory not found.', flush=True)
+        return False
+    if not _trusted_agent_dir(agent_dir):
+        print('[!!] Auto-install skipped: agent directory failed trust check (check ownership/permissions).', flush=True)
+        return False
+    req_file = agent_dir / 'requirements.txt'
+    pyproject = agent_dir / 'pyproject.toml'
+    if req_file.exists():
+        install_args = [sys.executable, '-m', 'pip', 'install', '--quiet', '-r', str(req_file)]
+        print(f'     Installing from {req_file} ...', flush=True)
+    elif pyproject.exists():
+        install_args = [sys.executable, '-m', 'pip', 'install', '--quiet', str(agent_dir)]
+        print(f'     Installing from {agent_dir} (pyproject.toml) ...', flush=True)
+    else:
+        print('[!!] Auto-install skipped: no requirements.txt or pyproject.toml in agent dir.', flush=True)
+        return False
+    try:
+        result = subprocess.run(install_args, capture_output=True, text=True, timeout=120)
+        if result.returncode != 0:
+            print(f'[!!] pip install failed (exit {result.returncode}):', flush=True)
+            for line in (result.stderr or '').splitlines()[-10:]:
+                print(f'     {line}', flush=True)
+            return False
+        print('[ok] pip install completed.', flush=True)
+        return True
+    except subprocess.TimeoutExpired:
+        print('[!!] Auto-install timed out after 120s.', flush=True)
+        return False
+    except Exception as e:
+        print(f'[!!] Auto-install error: {e}', flush=True)
+        return False

api/state_sync.py

@@ -13,9 +13,12 @@ The bridge uses absolute token counts (not deltas) because the WebUI
 Session object already accumulates totals across turns. This avoids
 any double-counting risk.
 """
+import logging
 import os
 from pathlib import Path
 
+logger = logging.getLogger(__name__)
+
 
 def _get_state_db():
     """Get a SessionDB instance for the active profile's state.db.
@@ -31,6 +34,7 @@ def _get_state_db():
         from api.profiles import get_active_hermes_home
         hermes_home = Path(get_active_hermes_home()).expanduser().resolve()
     except Exception:
+        logger.debug("Failed to resolve hermes home, using default")
         hermes_home = Path(os.getenv('HERMES_HOME', str(Path.home() / '.hermes')))
 
     db_path = hermes_home / 'state.db'
@@ -40,6 +44,7 @@ def _get_state_db():
     try:
         return SessionDB(db_path)
     except Exception:
+        logger.debug("Failed to open state.db")
         return None
 
 
@@ -57,16 +62,17 @@ def sync_session_start(session_id: str, model=None) -> None:
             model=model,
         )
     except Exception:
-        pass  # never crash the WebUI for sync failures
+        logger.debug("Failed to sync session start to state.db")
     finally:
         try:
             db.close()
         except Exception:
-            pass
+            logger.debug("Failed to close state.db")
 
 
 def sync_session_usage(session_id: str, input_tokens: int=0, output_tokens: int=0,
-                       estimated_cost=None, model=None, title: str=None) -> None:
+                       estimated_cost=None, model=None, title: str=None,
+                       message_count: int=None) -> None:
     """Update token usage and title for a WebUI session in state.db.
     Called after each turn completes. Uses absolute=True to set totals
     (the WebUI Session already accumulates across turns).
@@ -91,11 +97,22 @@ def sync_session_usage(session_id: str, input_tokens: int=0, output_tokens: int=
             try:
                 db.set_session_title(session_id, title)
             except Exception:
-                pass
+                logger.debug("Failed to sync session title to state.db")
+        # Update message count
+        if message_count is not None:
+            try:
+                def _set_msg_count(conn):
+                    conn.execute(
+                        "UPDATE sessions SET message_count = ? WHERE id = ?",
+                        (message_count, session_id),
+                    )
+                db._execute_write(_set_msg_count)
+            except Exception:
+                logger.debug("Failed to sync message count to state.db")
     except Exception:
-        pass  # never crash the WebUI for sync failures
+        logger.debug("Failed to sync session usage to state.db")
     finally:
         try:
             db.close()
         except Exception:
-            pass
+            logger.debug("Failed to close state.db")

api/updates.py

@@ -29,15 +29,81 @@ CACHE_TTL = 1800  # 30 minutes
 
 
 def _run_git(args, cwd, timeout=10):
-    """Run a git command and return (stdout, ok)."""
+    """Run a git command and return (useful output, ok).
+
+    On failure, returns stderr (or stdout as fallback) so callers can
+    surface actionable git error messages instead of empty strings.
+    """
     try:
         r = subprocess.run(
             ['git'] + args, cwd=str(cwd), capture_output=True,
             text=True, timeout=timeout,
         )
-        return r.stdout.strip(), r.returncode == 0
-    except (subprocess.TimeoutExpired, FileNotFoundError, OSError):
-        return '', False
+        stdout = r.stdout.strip()
+        stderr = r.stderr.strip()
+        if r.returncode == 0:
+            return stdout, True
+        return stderr or stdout or f"git exited with status {r.returncode}", False
+    except subprocess.TimeoutExpired as exc:
+        detail = (getattr(exc, 'stderr', None) or getattr(exc, 'stdout', None) or '').strip()
+        return detail or f"git {' '.join(args)} timed out after {timeout}s", False
+    except FileNotFoundError:
+        return 'git executable not found', False
+    except OSError as exc:
+        return f'git failed to start: {exc}', False
+
+
+def _detect_webui_version() -> str:
+    """Detect the running WebUI version from git or a baked-in fallback file.
+
+    Resolution order:
+      1. ``git describe --tags --always --dirty`` — works in any git checkout.
+         Returns the exact tag on tagged commits (e.g. ``v0.50.124``), a
+         post-tag descriptor between releases (e.g. ``v0.50.124-1-ge91325d``),
+         or a bare SHA when no tags exist (shallow clones, fresh forks).
+      2. ``api/_version.py`` — a fallback written by the Docker / CI release
+         workflow when ``.git`` is not present in the image.  Expected to define
+         ``__version__ = 'vX.Y.Z'``.
+      3. ``'unknown'`` — last resort; displayed as-is in the settings badge.
+    """
+    # Timeout capped at 3s: git describe on a healthy local repo is <50ms;
+    # a 10s stall on import (NFS-mounted .git, broken git binary) is unacceptable.
+    out, ok = _run_git(['describe', '--tags', '--always', '--dirty'], REPO_ROOT, timeout=3)
+    if ok and out:
+        return out
+
+    # Docker / baked-image fallback: api/_version.py written by CI at build time.
+    # Parse with regex rather than exec() — the file holds exactly one assignment
+    # and regex is sufficient; exec() on a build artifact is an unnecessary surface.
+    version_file = REPO_ROOT / 'api' / '_version.py'
+    if version_file.exists():
+        try:
+            import re as _re
+            m = _re.search(
+                r"""__version__\s*=\s*['"]([^'"]+)['"]""",
+                version_file.read_text(encoding='utf-8'),
+            )
+            if m:
+                return m.group(1)
+        except Exception:
+            pass
+
+    return 'unknown'
+
+
+# Resolved once at import time — tags cannot change without a process restart.
+WEBUI_VERSION: str = _detect_webui_version()
+
+
+def _split_remote_ref(ref):
+    """Split 'origin/branch-name' into ('origin', 'branch-name').
+
+    Returns (None, ref) if ref contains no slash.
+    """
+    if '/' not in ref:
+        return None, ref
+    remote, branch = ref.split('/', 1)
+    return remote, branch
 
 
 def _detect_default_branch(path):
@@ -64,22 +130,32 @@ def _check_repo(path, name):
     if not fetch_ok:
         return {'name': name, 'behind': 0, 'error': 'fetch failed'}
 
-    branch = _detect_default_branch(path)
+    # Use the current branch's upstream tracking branch, not the repo default.
+    # This avoids false "N updates behind" alerts when the user is on a feature
+    # branch and master/main has moved forward with unrelated commits.
+    # If no upstream is set (brand-new local branch), fall back to the default branch.
+    upstream, ok = _run_git(['rev-parse', '--abbrev-ref', '@{upstream}'], path)
+    if ok and upstream:
+        # upstream is like "origin/feat/foo" — use it directly in rev-list
+        compare_ref = upstream
+    else:
+        branch = _detect_default_branch(path)
+        compare_ref = f'origin/{branch}'
 
     # Count commits behind
-    out, ok = _run_git(['rev-list', '--count', f'HEAD..origin/{branch}'], path)
+    out, ok = _run_git(['rev-list', '--count', f'HEAD..{compare_ref}'], path)
     behind = int(out) if ok and out.isdigit() else 0
 
     # Get short SHAs for display
     current, _ = _run_git(['rev-parse', '--short', 'HEAD'], path)
-    latest, _ = _run_git(['rev-parse', '--short', f'origin/{branch}'], path)
+    latest, _ = _run_git(['rev-parse', '--short', compare_ref], path)
 
     return {
         'name': name,
         'behind': behind,
         'current_sha': current,
         'latest_sha': latest,
-        'branch': branch,
+        'branch': compare_ref,
     }
 
 
@@ -107,6 +183,111 @@ def check_for_updates(force=False):
         _check_in_progress = False
 
 
+def _schedule_restart(delay: float = 2.0) -> None:
+    """Re-exec this process after *delay* seconds.
+
+    Called after a successful update so that the freshly-pulled code is
+    loaded on the next request, rather than running with a mix of old and
+    new Python modules in sys.modules.
+
+    os.execv() replaces the current process image with a fresh interpreter
+    running the same argv — sessions are preserved on disk, the HTTP port
+    is reclaimed within the delay window, and the client's own
+    ``setTimeout(() => location.reload(), 2500)`` lands after the restart.
+
+    Coordinates with ``_apply_lock``: when the user updates both webui
+    and agent, the client POSTs them sequentially.  Without coordination
+    the restart timer scheduled by the first update's success would fire
+    while the second update's git-pull is still running, killing it mid-
+    stream and leaving the second repo in an unknown partial state.
+    Blocking on ``_apply_lock`` before ``os.execv`` means a pending
+    second update always completes before the restart happens.
+    """
+    import os
+    import sys
+
+    def _do():
+        import time
+        time.sleep(delay)
+        # Hold _apply_lock through os.execv so no new update can start between
+        # the lock-release and the process replacement.  Any in-flight update
+        # finishes first (since it holds the lock), and then the process is
+        # replaced while still holding the lock — meaning no new update can
+        # sneak in during the brief TOCTOU window that existed with the
+        # original acquire-release-execv sequence.
+        # Threads die when execv replaces the process image, so the lock is
+        # released atomically by the kernel.
+        with _apply_lock:
+            try:
+                os.execv(sys.executable, [sys.executable] + sys.argv)
+            except Exception:
+                # Last-resort: if execv fails (e.g. frozen binary), just exit
+                # so the process supervisor (start.sh / Docker) restarts us.
+                os._exit(0)
+
+    threading.Thread(target=_do, daemon=True).start()
+
+
+def apply_force_update(target: str) -> dict:
+    """Force-reset the target repo to the latest remote HEAD.
+
+    Unlike apply_update() which requires a clean working tree and refuses
+    merge conflicts, this discards all local modifications (checkout .) and
+    resets to origin/<branch> — equivalent to what the diverged/conflict
+    error messages ask the user to run manually.
+
+    Should only be called when apply_update() has already returned a
+    response with ``conflict: True`` or ``diverged: True`` and the user
+    has confirmed they want to discard local changes.
+    """
+    if not _apply_lock.acquire(blocking=False):
+        return {'ok': False, 'message': 'Update already in progress'}
+    try:
+        if target == 'webui':
+            path = REPO_ROOT
+        elif target == 'agent':
+            path = _AGENT_DIR
+        else:
+            return {'ok': False, 'message': f'Unknown target: {target}'}
+
+        if path is None or not (path / '.git').exists():
+            return {'ok': False, 'message': 'Not a git repository'}
+
+        _, fetch_ok = _run_git(['fetch', 'origin', '--quiet'], path, timeout=15)
+        if not fetch_ok:
+            return {
+                'ok': False,
+                'message': 'Could not reach the remote repository. Check your connection.',
+            }
+
+        upstream, ok = _run_git(['rev-parse', '--abbrev-ref', '@{upstream}'], path)
+        if ok and upstream:
+            compare_ref = upstream
+        else:
+            branch = _detect_default_branch(path)
+            compare_ref = f'origin/{branch}'
+
+        # Discard local modifications then reset to remote HEAD
+        _run_git(['checkout', '.'], path)
+        _, ok = _run_git(['reset', '--hard', compare_ref], path)
+        if not ok:
+            return {'ok': False, 'message': f'Force reset to {compare_ref} failed'}
+
+        with _cache_lock:
+            _update_cache['checked_at'] = 0
+
+        _schedule_restart()
+
+        return {
+            'ok': True,
+            'message': f'{target} force-updated to {compare_ref}',
+            'target': target,
+            'restart_scheduled': True,
+        }
+    finally:
+        _apply_lock.release()
+
+
 def apply_update(target):
     """Stash, pull --ff-only, pop for the given target repo."""
     if not _apply_lock.acquire(blocking=False):
@@ -129,10 +310,46 @@ def _apply_update_inner(target):
     if path is None or not (path / '.git').exists():
         return {'ok': False, 'message': 'Not a git repository'}
 
-    branch = _detect_default_branch(path)
+    # Use the current branch's upstream for pull, matching the behaviour
+    # of _check_repo. Falls back to default branch if no upstream is set.
+    upstream, ok = _run_git(['rev-parse', '--abbrev-ref', '@{upstream}'], path)
+    if ok and upstream:
+        compare_ref = upstream
+    else:
+        branch = _detect_default_branch(path)
+        compare_ref = f'origin/{branch}'
 
-    # Check for dirty working tree
-    status_out, _ = _run_git(['status', '--porcelain'], path)
+    # Fetch before attempting pull, so the remote ref is current.
+    _, fetch_ok = _run_git(['fetch', 'origin', '--quiet'], path, timeout=15)
+    if not fetch_ok:
+        return {
+            'ok': False,
+            'message': (
+                'Could not reach the remote repository. '
+                'Check your internet connection and try again.'
+            ),
+        }
+
+    # Check for dirty working tree (ignore untracked files — git stash
+    # doesn't include them, so stashing on '??' alone leaves nothing to pop)
+    status_out, status_ok = _run_git(
+        ['status', '--porcelain', '--untracked-files=no'], path
+    )
+    if not status_ok:
+        return {'ok': False, 'message': f'Failed to inspect repo status: {status_out[:200]}'}
+    # Fail early on unresolved merge conflicts
+    if any(line[:2] in {'DD', 'AU', 'UD', 'UA', 'DU', 'AA', 'UU'}
+           for line in status_out.splitlines()):
+        return {
+            'ok': False,
+            'message': (
+                f'The local {target} repo has unresolved merge conflicts. '
+                'To reset to the latest remote version run: '
+                'git -C ' + str(path) + ' checkout . && '
+                'git -C ' + str(path) + ' pull --ff-only'
+            ),
+            'conflict': True,
+        }
     stashed = False
     if status_out:
         _, ok = _run_git(['stash'], path)
@@ -140,12 +357,44 @@ def _apply_update_inner(target):
             return {'ok': False, 'message': 'Failed to stash local changes'}
         stashed = True
 
-    # Pull with ff-only (no merge commits)
-    pull_out, pull_ok = _run_git(['pull', '--ff-only', 'origin', branch], path, timeout=30)
+    # Pull with ff-only (no merge commits).
+    # Split tracking refs like 'origin/main' into separate remote + branch
+    # arguments — git treats 'origin/main' as a repository name otherwise.
+    remote, branch = _split_remote_ref(compare_ref)
+    pull_args = ['pull', '--ff-only']
+    if remote:
+        pull_args.extend([remote, branch])
+    else:
+        pull_args.append(compare_ref)
+    pull_out, pull_ok = _run_git(pull_args, path, timeout=30)
     if not pull_ok:
         if stashed:
             _run_git(['stash', 'pop'], path)
-        return {'ok': False, 'message': f'Pull failed: {pull_out[:200]}'}
+
+        # Diagnose the most common failure modes and surface actionable messages.
+        pull_lower = pull_out.lower()
+        if 'not possible to fast-forward' in pull_lower or 'diverged' in pull_lower:
+            return {
+                'ok': False,
+                'message': (
+                    f'The local {target} repo has commits that are not on the remote '
+                    'branch, so a fast-forward update is not possible. '
+                    'Run: git -C ' + str(path) + ' fetch origin && '
+                    'git -C ' + str(path) + ' reset --hard ' + compare_ref
+                ),
+                'diverged': True,
+            }
+        if 'does not track' in pull_lower or 'no tracking information' in pull_lower:
+            return {
+                'ok': False,
+                'message': (
+                    f'The local {target} branch has no upstream tracking branch configured. '
+                    'Run: git -C ' + str(path) + ' branch --set-upstream-to=' + compare_ref
+                ),
+            }
+        # Generic fallback — include the raw git output for debugging.
+        detail = pull_out.strip()[:300] if pull_out.strip() else '(no output from git)'
+        return {'ok': False, 'message': f'Pull failed: {detail}'}
 
     # Pop stash if we stashed
     if stashed:
@@ -161,4 +410,22 @@ def _apply_update_inner(target):
     with _cache_lock:
         _update_cache['checked_at'] = 0
 
-    return {'ok': True, 'message': f'{target} updated successfully', 'target': target}
+    # Schedule a self-restart so the updated code is loaded fresh.  A plain
+    # git pull leaves stale Python modules in sys.modules — agent imports that
+    # reference new symbols (functions, classes) added in the update will fail
+    # on the next request with AttributeError / ImportError.  os.execv() re-
+    # execs the same interpreter with the same argv, picking up the new code
+    # cleanly without requiring the user to restart manually.
+    #
+    # The 2 s delay gives the HTTP response time to flush to the client before
+    # the process replaces itself.  The client already does
+    # setTimeout(() => location.reload(), 1500) on success, so the page reload
+    # and the restart land at roughly the same time.
+    _schedule_restart()
+
+    return {
+        'ok': True,
+        'message': f'{target} updated successfully',
+        'target': target,
+        'restart_scheduled': True,
+    }

api/upload.py

@@ -3,6 +3,7 @@ Hermes Web UI -- File upload: multipart parser and upload handler.
 """
 import re as _re
 import email.parser
+import tempfile
 from pathlib import Path
 
 from api.config import MAX_UPLOAD_BYTES
@@ -50,8 +51,15 @@ def parse_multipart(rfile, content_type, content_length) -> tuple:
     return fields, files
 
 
+def _sanitize_upload_name(filename: str) -> str:
+    safe_name = _re.sub(r'[^\w.\-]', '_', Path(filename).name)[:200]
+    if not safe_name or safe_name.strip('.') == '':
+        raise ValueError('Invalid filename')
+    return safe_name
+
+
 def handle_upload(handler):
-    import re as _re, traceback as _tb
+    import traceback as _tb
     try:
         content_type = handler.headers.get('Content-Type', '')
         content_length = int(handler.headers.get('Content-Length', 0) or 0)
@@ -69,14 +77,55 @@ def handle_upload(handler):
         except KeyError:
             return j(handler, {'error': 'Session not found'}, status=404)
         workspace = Path(s.workspace)
-        safe_name = _re.sub(r'[^\w.\-]', '_', Path(filename).name)[:200]
-        # Reject names that are purely dots (path traversal: ".." survives regex)
-        if not safe_name or safe_name.strip('.') == '':
-            return j(handler, {'error': 'Invalid filename'}, status=400)
-        # Verify the resolved path stays within the workspace
+        safe_name = _sanitize_upload_name(filename)
         dest = safe_resolve_ws(workspace, safe_name)
         dest.write_bytes(file_bytes)
         return j(handler, {'filename': safe_name, 'path': str(dest), 'size': dest.stat().st_size})
-    except Exception as e:
+    except ValueError as e:
+        return j(handler, {'error': str(e)}, status=400)
+    except Exception:
         print('[webui] upload error: ' + _tb.format_exc(), flush=True)
         return j(handler, {'error': 'Upload failed'}, status=500)
+
+
+def handle_transcribe(handler):
+    import traceback as _tb
+    temp_path = None
+    try:
+        content_type = handler.headers.get('Content-Type', '')
+        content_length = int(handler.headers.get('Content-Length', 0) or 0)
+        if content_length > MAX_UPLOAD_BYTES:
+            return j(handler, {'error': f'File too large (max {MAX_UPLOAD_BYTES//1024//1024}MB)'}, status=413)
+        fields, files = parse_multipart(handler.rfile, content_type, content_length)
+        if 'file' not in files:
+            return j(handler, {'error': 'No file field in request'}, status=400)
+        filename, file_bytes = files['file']
+        if not filename:
+            return j(handler, {'error': 'No filename in upload'}, status=400)
+        safe_name = _sanitize_upload_name(filename)
+        suffix = Path(safe_name).suffix or '.webm'
+        with tempfile.NamedTemporaryFile(prefix='webui-stt-', suffix=suffix, delete=False) as tmp:
+            temp_path = tmp.name
+            tmp.write(file_bytes)
+        try:
+            from tools.transcription_tools import transcribe_audio
+        except ImportError:
+            return j(handler, {'error': 'Speech-to-text is unavailable on this server'}, status=503)
+        result = transcribe_audio(temp_path)
+        if not result.get('success'):
+            msg = str(result.get('error') or 'Transcription failed')
+            status = 503 if 'unavailable' in msg.lower() or 'not configured' in msg.lower() else 400
+            return j(handler, {'error': msg}, status=status)
+        transcript = str(result.get('transcript') or '').strip()
+        return j(handler, {'ok': True, 'transcript': transcript})
+    except ValueError as e:
+        return j(handler, {'error': str(e)}, status=400)
+    except Exception:
+        print('[webui] transcribe error: ' + _tb.format_exc(), flush=True)
+        return j(handler, {'error': 'Transcription failed'}, status=500)
+    finally:
+        if temp_path:
+            try:
+                Path(temp_path).unlink(missing_ok=True)
+            except Exception:
+                pass

api/workspace.py

@@ -8,10 +8,13 @@ profile has its own workspace configuration.  State files live at
 paths are used as fallback when no profile module is available.
 """
 import json
+import logging
 import os
 import subprocess
 from pathlib import Path
 
+logger = logging.getLogger(__name__)
+
 from api.config import (
     WORKSPACES_FILE as _GLOBAL_WS_FILE,
     LAST_WORKSPACE_FILE as _GLOBAL_LW_FILE,
@@ -37,7 +40,7 @@ def _profile_state_dir() -> Path:
             d.mkdir(parents=True, exist_ok=True)
             return d
     except ImportError:
-        pass
+        logger.debug("Failed to import profiles module, using global state dir")
     return _GLOBAL_WS_FILE.parent
 
 
@@ -80,7 +83,7 @@ def _profile_default_workspace() -> str:
                 if p.is_dir():
                     return str(p)
     except (ImportError, Exception):
-        pass
+        logger.debug("Failed to load profile default workspace config")
     return str(_BOOT_DEFAULT_WORKSPACE)
 
 
@@ -89,7 +92,6 @@ def _profile_default_workspace() -> str:
 def _clean_workspace_list(workspaces: list) -> list:
     """Sanitize a workspace list:
     - Remove entries whose paths no longer exist on disk.
-    - Remove entries that look like test artifacts (webui-mvp-test, test-workspace).
     - Remove entries whose paths live inside another profile's directory
       (e.g. ~/.hermes/profiles/X/... should not appear on a different profile).
     - Rename any entry whose name is literally 'default' to 'Home' (avoids
@@ -102,18 +104,24 @@ def _clean_workspace_list(workspaces: list) -> list:
         path = w.get('path', '')
         name = w.get('name', '')
         p = Path(path).resolve() if path else Path('/')
-        # Skip test artifacts
-        if 'test-workspace' in path or 'webui-mvp-test' in path:
-            continue
         # Skip paths that no longer exist
         if not p.is_dir():
             continue
-        # Skip paths inside a named profile's directory (cross-profile leak)
+        # Skip paths inside a DIFFERENT profile's directory (cross-profile leak).
+        # Allow paths inside the CURRENT profile's own directory (e.g. test workspaces
+        # created under ~/.hermes/profiles/webui/webui-mvp-test/).
         try:
             p.relative_to(hermes_profiles)
-            continue  # it IS under profiles/ — remove it
+            # p is under ~/.hermes/profiles/ — only skip if it's under a DIFFERENT profile
+            try:
+                from api.profiles import get_active_hermes_home
+                own_profile_dir = get_active_hermes_home().resolve()
+                p.relative_to(own_profile_dir)
+                # p is under our own profile dir — keep it
+            except (ValueError, Exception):
+                continue  # under profiles/ but not our own — cross-profile leak, skip
         except ValueError:
-            pass
+            pass  # not under profiles/ at all — keep it
         # Rename confusing 'default' label to 'Home'
         if name.lower() == 'default':
             name = 'Home'
@@ -156,10 +164,10 @@ def load_workspaces() -> list:
                         json.dumps(cleaned, ensure_ascii=False, indent=2), encoding='utf-8'
                     )
                 except Exception:
-                    pass
+                    logger.debug("Failed to persist cleaned workspace list")
             return cleaned or [{'path': _profile_default_workspace(), 'name': 'Home'}]
         except Exception:
-            pass
+            logger.debug("Failed to load workspaces from %s", ws_file)
     # No profile-local file yet.
     # For the DEFAULT profile: migrate from the legacy global file (one-time cleanup).
     # For NAMED profiles: always start clean with just their own workspace.
@@ -190,7 +198,7 @@ def get_last_workspace() -> str:
             if p and Path(p).is_dir():
                 return p
         except Exception:
-            pass
+            logger.debug("Failed to read last workspace from %s", lw_file)
     # Fallback: try global file
     if _GLOBAL_LW_FILE.exists():
         try:
@@ -198,7 +206,7 @@ def get_last_workspace() -> str:
             if p and Path(p).is_dir():
                 return p
         except Exception:
-            pass
+            logger.debug("Failed to read global last workspace")
     return _profile_default_workspace()
 
 
@@ -208,8 +216,249 @@ def set_last_workspace(path: str) -> None:
         lw_file.parent.mkdir(parents=True, exist_ok=True)
         lw_file.write_text(str(path), encoding='utf-8')
     except Exception:
+        logger.debug("Failed to set last workspace")
+
+
+def _workspace_blocked_roots() -> tuple[Path, ...]:
+    return (
+        # Linux / macOS
+        Path('/etc'),
+        Path('/usr'),
+        Path('/var'),
+        Path('/bin'),
+        Path('/sbin'),
+        Path('/boot'),
+        Path('/proc'),
+        Path('/sys'),
+        Path('/dev'),
+        Path('/lib'),
+        Path('/lib64'),
+        Path('/opt/homebrew'),
+    )
+
+
+def _is_within(path: Path, root: Path) -> bool:
+    try:
+        path.relative_to(root)
+        return True
+    except ValueError:
+        return False
+
+
+def _trusted_workspace_roots() -> list[Path]:
+    roots: list[Path] = []
+
+    def add(candidate: str | Path | None) -> None:
+        if candidate in (None, ""):
+            return
+        try:
+            p = Path(candidate).expanduser().resolve()
+        except Exception:
+            return
+        if not p.exists() or not p.is_dir():
+            return
+        if any(_is_within(p, blocked) for blocked in _workspace_blocked_roots()):
+            return
+        if p not in roots:
+            roots.append(p)
+
+    add(Path.home())
+    add(_BOOT_DEFAULT_WORKSPACE)
+    for w in load_workspaces():
+        add(w.get("path"))
+    roots.sort(key=lambda p: len(str(p)))
+    return roots
+
+
+def list_workspace_suggestions(prefix: str = "", limit: int = 12) -> list[str]:
+    """Return workspace path suggestions under trusted roots only.
+
+    Suggestions are limited to directories under one of:
+      - Path.home()
+      - the boot default workspace
+      - already-saved workspace roots
+
+    Arbitrary system prefixes return an empty list rather than an error so the
+    UI can safely autocomplete while the user types.
+    """
+    roots = _trusted_workspace_roots()
+    if not roots:
+        return []
+
+    raw = (prefix or "").strip()
+    if not raw:
+        return [str(p) for p in roots[:limit]]
+
+    if raw.startswith("~"):
+        target = Path(raw).expanduser()
+    elif Path(raw).is_absolute():
+        target = Path(raw)
+    else:
+        target = Path.home() / raw
+
+    normalized = str(target)
+    normalized_lower = normalized.lower()
+    suggestions: list[str] = []
+
+    def add(path: Path) -> None:
+        value = str(path)
+        if value not in suggestions:
+            suggestions.append(value)
+
+    # If the user is typing a partial trusted root like /Users/xuef..., suggest
+    # the matching trusted roots without scanning arbitrary system parents.
+    for root in roots:
+        if str(root).lower().startswith(normalized_lower):
+            add(root)
+
+    in_root = [
+        root
+        for root in roots
+        if normalized == str(root) or normalized.startswith(str(root) + os.sep)
+    ]
+    if not in_root:
+        return suggestions[:limit]
+
+    anchor_root = max(in_root, key=lambda p: len(str(p)))
+    ends_with_sep = raw.endswith(os.sep) or raw.endswith('/')
+    parent = target if ends_with_sep else target.parent
+    leaf = '' if ends_with_sep else target.name
+    show_hidden = leaf.startswith('.')
+
+    try:
+        parent_resolved = parent.expanduser().resolve()
+    except Exception:
+        return suggestions[:limit]
+
+    if not parent_resolved.exists() or not parent_resolved.is_dir():
+        return suggestions[:limit]
+    if not _is_within(parent_resolved, anchor_root):
+        return suggestions[:limit]
+
+    leaf_lower = leaf.lower()
+    try:
+        children = sorted(parent_resolved.iterdir(), key=lambda p: p.name.lower())
+    except OSError:
+        return suggestions[:limit]
+
+    for child in children:
+        if not child.is_dir():
+            continue
+        if child.name.startswith('.') and not show_hidden:
+            continue
+        if leaf_lower and not child.name.lower().startswith(leaf_lower):
+            continue
+        add(child.resolve())
+        if len(suggestions) >= limit:
+            break
+    return suggestions[:limit]
+
+
+def resolve_trusted_workspace(path: str | Path | None = None) -> Path:
+    """Resolve and validate a workspace path.
+
+    A path is trusted if it satisfies at least one of:
+      (A) It is under the user's home directory (Path.home()).
+          Works cross-platform: ~/... on Linux/macOS, C:\\Users\\... on Windows.
+      (B) It is already in the profile's saved workspace list.
+          This covers self-hosted deployments where workspaces live outside home
+          (e.g. /data/projects, /opt/workspace) — once a workspace is saved by
+          an admin, it can be reused without re-validation.
+
+    Additionally enforced regardless of (A)/(B):
+      1. The path must exist.
+      2. The path must be a directory.
+      3. The path must not be a known system root (/etc, /usr, /var, /bin, /sbin,
+         /boot, /proc, /sys, /dev, /root on Linux/macOS; Windows system dirs).
+         This prevents even admin-saved workspaces from pointing at OS internals.
+
+    None/empty path falls back to the boot-time DEFAULT_WORKSPACE, which is always
+    trusted (it was validated at server startup).
+    """
+    if path in (None, ""):
+        return Path(_BOOT_DEFAULT_WORKSPACE).expanduser().resolve()
+
+    candidate = Path(path).expanduser().resolve()
+
+    if not candidate.exists():
+        raise ValueError(f"Path does not exist: {candidate}")
+    if not candidate.is_dir():
+        raise ValueError(f"Path is not a directory: {candidate}")
+
+    # Block known system roots and their children
+    for blocked in _workspace_blocked_roots():
+        try:
+            candidate.relative_to(blocked)
+            raise ValueError(f"Path points to a system directory: {candidate}")
+        except ValueError as e:
+            if "system directory" in str(e):
+                raise
+            # relative_to raised ValueError = candidate is NOT under blocked = safe
+
+    # (A) Trusted if under the user's home directory — cross-platform via Path.home()
+    try:
+        candidate.relative_to(Path.home().resolve())
+        return candidate
+    except ValueError:
         pass
 
+    # (B) Trusted if already in the saved workspace list — covers non-home installs
+    try:
+        saved = load_workspaces()
+        saved_paths = {Path(w["path"]).resolve() for w in saved if w.get("path")}
+        if candidate in saved_paths:
+            return candidate
+    except Exception:
+        pass
+
+    # (C) Trusted if it is equal to or under the boot-time DEFAULT_WORKSPACE.
+    #     In Docker deployments HERMES_WEBUI_DEFAULT_WORKSPACE is often set to a
+    #     volume mount outside the user's home (e.g. /data/workspace).  That path
+    #     was already validated at server startup, so any sub-path of it is safe
+    #     without requiring the user to add it to the workspace list manually.
+    try:
+        boot_default = Path(_BOOT_DEFAULT_WORKSPACE).expanduser().resolve()
+        candidate.relative_to(boot_default)
+        return candidate
+    except ValueError:
+        pass
+
+    raise ValueError(
+        f"Path is outside the user home directory, not in the saved workspace "
+        f"list, and not under the default workspace: {candidate}. "
+        f"Add it via Settings → Workspaces first."
+    )
+
+
+
+
+def validate_workspace_to_add(path: str) -> Path:
+    """Validate a path for *adding* to the workspace list (less restrictive than resolve_trusted_workspace).
+
+    When a user explicitly adds a new workspace path, we trust their intent — they
+    have console or filesystem access to that path and are consciously registering it.
+    We only block: non-existent paths, non-directories, and known system roots.
+
+    The stricter ``resolve_trusted_workspace`` is used when *using* an existing workspace
+    (file reads/writes) to prevent path traversal after the list is built.
+    """
+    candidate = Path(path).expanduser().resolve()
+
+    if not candidate.exists():
+        raise ValueError(f"Path does not exist: {candidate}")
+    if not candidate.is_dir():
+        raise ValueError(f"Path is not a directory: {candidate}")
+
+    # Block known system roots and their immediate children
+    for blocked in _workspace_blocked_roots():
+        try:
+            candidate.relative_to(blocked)
+            raise ValueError(f"Path points to a system directory: {candidate}")
+        except ValueError as e:
+            if "system directory" in str(e):
+                raise
+
+    return candidate
 
 def safe_resolve_ws(root: Path, requested: str) -> Path:
     """Resolve a relative path inside a workspace root, raising ValueError on traversal."""

bootstrap.py

@@ -0,0 +1,276 @@
+#!/usr/bin/env python3
+"""One-shot bootstrap launcher for Hermes Web UI."""
+
+from __future__ import annotations
+
+import argparse
+import os
+import platform
+import shutil
+import subprocess
+import sys
+import time
+import urllib.error
+import urllib.request
+import venv
+import webbrowser
+from pathlib import Path
+
+
+INSTALLER_URL = "https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh"
+REPO_ROOT = Path(__file__).resolve().parent
+
+
+def _load_repo_dotenv() -> None:
+    """Load REPO_ROOT/.env into os.environ.
+
+    Mirrors what start.sh does via ``set -a; source .env`` so that running
+    ``python3 bootstrap.py`` directly behaves identically to ``./start.sh``.
+    Variables are set unconditionally (matching shell source semantics), so a
+    value in .env overrides one already present in the shell environment.
+    To keep a CLI-supplied value, unset it from .env or launch via start.sh
+    and override there.
+
+    Only loads the webui repo .env — not ~/.hermes/.env, which the server
+    loads independently at startup for provider credentials.
+
+    Note: does not handle the ``export FOO=bar`` prefix — strip ``export``
+    from .env values if copy-pasting from a shell rc file.
+    """
+    env_path = REPO_ROOT / ".env"
+    if not env_path.exists():
+        return
+    try:
+        for raw_line in env_path.read_text(encoding="utf-8").splitlines():
+            line = raw_line.strip()
+            if not line or line.startswith("#") or "=" not in line:
+                continue
+            k, v = line.split("=", 1)
+            k = k.strip()
+            # Strip optional 'export' prefix (common in copy-pasted shell snippets)
+            if k.startswith("export "):
+                k = k[7:].strip()
+            v = v.strip().strip('"').strip("'")
+            if k:
+                os.environ[k] = v
+    except Exception as exc:
+        import sys as _sys
+        print(f"[bootstrap] Warning: could not load .env — {exc}", file=_sys.stderr)
+
+
+# Side effect: loads REPO_ROOT/.env into os.environ on import.
+# Must run before DEFAULT_HOST / DEFAULT_PORT so os.getenv() picks up
+# values from .env even when bootstrap.py is invoked directly (not via start.sh).
+_load_repo_dotenv()
+
+DEFAULT_HOST = os.getenv("HERMES_WEBUI_HOST", "127.0.0.1")
+DEFAULT_PORT = int(os.getenv("HERMES_WEBUI_PORT", "8787"))
+# Set HERMES_WEBUI_SKIP_ONBOARDING=1 to bypass the first-run wizard when
+# the environment is already fully configured (e.g. managed hosting).
+
+
+def info(msg: str) -> None:
+    print(f"[bootstrap] {msg}", flush=True)
+
+
+def is_wsl() -> bool:
+    if platform.system() != "Linux":
+        return False
+    release = platform.release().lower()
+    return (
+        "microsoft" in release or "wsl" in release or bool(os.getenv("WSL_DISTRO_NAME"))
+    )
+
+
+def ensure_supported_platform() -> None:
+    if platform.system() == "Windows" and not is_wsl():
+        raise RuntimeError(
+            "Native Windows is not supported for this bootstrap yet. "
+            "Please run it from Linux, macOS, or inside WSL2."
+        )
+
+
+def discover_agent_dir() -> Path | None:
+    home = Path(os.getenv("HERMES_HOME", str(Path.home() / ".hermes"))).expanduser()
+    candidates = [
+        os.getenv("HERMES_WEBUI_AGENT_DIR", ""),
+        str(home / "hermes-agent"),
+        str(REPO_ROOT.parent / "hermes-agent"),
+        str(Path.home() / ".hermes" / "hermes-agent"),
+        str(Path.home() / "hermes-agent"),
+    ]
+    for raw in candidates:
+        if not raw:
+            continue
+        candidate = Path(raw).expanduser().resolve()
+        if candidate.exists() and (candidate / "run_agent.py").exists():
+            return candidate
+    return None
+
+
+def discover_launcher_python(agent_dir: Path | None) -> str:
+    env_python = os.getenv("HERMES_WEBUI_PYTHON")
+    if env_python:
+        return env_python
+    if agent_dir:
+        for rel in ("venv/bin/python", "venv/Scripts/python.exe", ".venv/bin/python", ".venv/Scripts/python.exe"):
+            candidate = agent_dir / rel
+            if candidate.exists():
+                return str(candidate)
+    for rel in (".venv/bin/python", ".venv/Scripts/python.exe"):
+        candidate = REPO_ROOT / rel
+        if candidate.exists():
+            return str(candidate)
+    return shutil.which("python3") or shutil.which("python") or sys.executable
+
+
+def ensure_python_has_webui_deps(python_exe: str) -> str:
+    check = subprocess.run(
+        [python_exe, "-c", "import yaml"],
+        capture_output=True,
+        text=True,
+    )
+    if check.returncode == 0:
+        return python_exe
+
+    venv_dir = REPO_ROOT / ".venv"
+    venv_python = venv_dir / (
+        "Scripts/python.exe" if platform.system() == "Windows" else "bin/python"
+    )
+    if not venv_python.exists():
+        info(f"Creating local virtualenv at {venv_dir}")
+        venv.EnvBuilder(with_pip=True).create(venv_dir)
+
+    info("Installing WebUI dependencies into local virtualenv")
+    subprocess.run(
+        [str(venv_python), "-m", "pip", "install", "--quiet", "--upgrade", "pip"],
+        check=True,
+    )
+    subprocess.run(
+        [
+            str(venv_python),
+            "-m",
+            "pip",
+            "install",
+            "--quiet",
+            "-r",
+            str(REPO_ROOT / "requirements.txt"),
+        ],
+        check=True,
+    )
+    return str(venv_python)
+
+
+def hermes_command_exists() -> bool:
+    return shutil.which("hermes") is not None
+
+
+def install_hermes_agent() -> None:
+    info(f"Hermes Agent not found. Attempting install via {INSTALLER_URL}")
+    subprocess.run(
+        ["/bin/bash", "-lc", f"curl -fsSL {INSTALLER_URL} | bash"], check=True
+    )
+
+
+def wait_for_health(url: str, timeout: float = 25.0) -> bool:
+    deadline = time.time() + timeout
+    # Validate URL scheme to prevent file:// and other dangerous schemes
+    if not url.startswith(("http://", "https://")):
+        raise ValueError(f"Invalid health check URL: {url}")
+    while time.time() < deadline:
+        try:
+            with urllib.request.urlopen(url, timeout=2) as response:  # nosec B310
+                if b'"status": "ok"' in response.read():
+                    return True
+        except Exception:
+            time.sleep(0.4)
+    return False
+
+
+def open_browser(url: str) -> None:
+    try:
+        webbrowser.open(url)
+    except Exception as exc:
+        info(f"Could not open browser automatically: {exc}")
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Bootstrap Hermes Web UI onboarding.")
+    parser.add_argument("port", nargs="?", type=int, default=DEFAULT_PORT)
+    parser.add_argument("--host", default=DEFAULT_HOST)
+    parser.add_argument(
+        "--no-browser",
+        action="store_true",
+        help="Do not open a browser tab automatically.",
+    )
+    parser.add_argument(
+        "--skip-agent-install",
+        action="store_true",
+        help="Fail instead of attempting the official Hermes installer.",
+    )
+    return parser.parse_args()
+
+
+def main() -> int:
+    args = parse_args()
+    ensure_supported_platform()
+
+    agent_dir = discover_agent_dir()
+    if not agent_dir and not hermes_command_exists():
+        if args.skip_agent_install:
+            raise RuntimeError(
+                "Hermes Agent was not found and auto-install was disabled."
+            )
+        install_hermes_agent()
+        agent_dir = discover_agent_dir()
+
+    python_exe = ensure_python_has_webui_deps(discover_launcher_python(agent_dir))
+    state_dir = Path(
+        os.getenv("HERMES_WEBUI_STATE_DIR", str(Path.home() / ".hermes" / "webui"))
+    ).expanduser()
+    state_dir.mkdir(parents=True, exist_ok=True)
+    log_path = state_dir / f"bootstrap-{args.port}.log"
+
+    env = os.environ.copy()
+    env["HERMES_WEBUI_HOST"] = args.host
+    env["HERMES_WEBUI_PORT"] = str(args.port)
+    env.setdefault("HERMES_WEBUI_STATE_DIR", str(state_dir))
+    if agent_dir:
+        env["HERMES_WEBUI_AGENT_DIR"] = str(agent_dir)
+
+    info(f"Starting Hermes Web UI on http://{args.host}:{args.port}")
+    with log_path.open("ab") as log_file:
+        proc = subprocess.Popen(
+            [python_exe, str(REPO_ROOT / "server.py")],
+            cwd=str(agent_dir or REPO_ROOT),
+            env=env,
+            stdout=log_file,
+            stderr=subprocess.STDOUT,
+            start_new_session=True,
+        )
+
+    health_url = f"http://{args.host}:{args.port}/health"
+    if not wait_for_health(health_url):
+        raise RuntimeError(
+            f"Web UI did not become healthy at {health_url}. "
+            f"Check the log at {log_path}. Server PID: {proc.pid}"
+        )
+
+    app_url = (
+        f"http://localhost:{args.port}"
+        if args.host in ("127.0.0.1", "localhost")
+        else f"http://{args.host}:{args.port}"
+    )
+    info(f"Web UI is ready: {app_url}")
+    info(f"Log file: {log_path}")
+    if not args.no_browser:
+        open_browser(app_url)
+    return 0
+
+
+if __name__ == "__main__":
+    try:
+        raise SystemExit(main())
+    except Exception as exc:
+        print(f"[bootstrap] ERROR: {exc}", file=sys.stderr)
+        raise SystemExit(1)

docker-compose.three-container.yml

@@ -0,0 +1,121 @@
+# Three-container Docker Compose: Hermes Agent + Dashboard + WebUI
+#
+# This extends the two-container setup with the Hermes Dashboard for
+# monitoring agent activity, sessions, and resource usage.
+#
+# Usage:
+#   docker compose -f docker-compose.three-container.yml up -d
+#
+# Services:
+#   hermes-agent   — gateway API on port 8642 (CLI, Telegram, cron, tools)
+#   hermes-dashboard — monitoring dashboard on port 9119
+#   hermes-webui   — browser chat interface on port 8787
+#
+# All three share the same hermes-home volume so config, sessions,
+# skills, and memory are consistent across all surfaces.
+#
+# NOTE ON VOLUMES:
+#   This file uses named Docker volumes (hermes-home, hermes-agent-src) which
+#   work out of the box. If you prefer bind mounts (e.g. to an existing directory),
+#   see the two-container compose file for a bind-mount example.
+#   When using bind mounts, ALL containers must mount the same host path.
+
+services:
+  hermes-agent:
+    image: nousresearch/hermes-agent:latest
+    container_name: hermes-agent
+    command: gateway run
+    ports:
+      - "127.0.0.1:8642:8642"
+    volumes:
+      # Persist config, state, sessions, skills, memory across restarts
+      - hermes-home:/home/hermes/.hermes
+      # Expose agent source so the WebUI can install dependencies from it
+      - hermes-agent-src:/opt/hermes
+    environment:
+      - HERMES_HOME=/home/hermes/.hermes
+      - HERMES_UID=${HERMES_UID:-10000}
+      - HERMES_GID=${HERMES_GID:-10000}
+    restart: unless-stopped
+    deploy:
+      resources:
+        limits:
+          memory: 4G
+          cpus: "2.0"
+    networks:
+      - hermes-net
+
+  hermes-dashboard:
+    image: nousresearch/hermes-agent:latest
+    container_name: hermes-dashboard
+    command: dashboard --host 0.0.0.0 --insecure
+    ports:
+      - "127.0.0.1:9119:9119"
+    volumes:
+      - hermes-home:/home/hermes/.hermes
+    environment:
+      - HERMES_HOME=/home/hermes/.hermes
+      - HERMES_UID=${HERMES_UID:-10000}
+      - HERMES_GID=${HERMES_GID:-10000}
+      # Dashboard connects to the gateway for health/session data
+      - GATEWAY_HEALTH_URL=http://hermes-agent:8642
+    depends_on:
+      - hermes-agent
+    restart: unless-stopped
+    deploy:
+      resources:
+        limits:
+          memory: 512M
+          cpus: "0.5"
+    networks:
+      - hermes-net
+
+  hermes-webui:
+    image: ghcr.io/nesquena/hermes-webui:latest
+    container_name: hermes-webui
+    depends_on:
+      - hermes-agent
+    ports:
+      # Expose on localhost only. Remove 127.0.0.1: to expose on all interfaces
+      # (set HERMES_WEBUI_PASSWORD if doing so).
+      - "127.0.0.1:8787:8787"
+    volumes:
+      # Same hermes home as the agent — shares config, sessions, state
+      - hermes-home:/home/hermeswebui/.hermes
+      # Agent source mounted where docker_init.bash expects it.
+      # At startup the init script runs:
+      #   uv pip install /home/hermeswebui/.hermes/hermes-agent
+      # which installs the agent and all its Python dependencies.
+      - hermes-agent-src:/home/hermeswebui/.hermes/hermes-agent
+      # Workspace directory — browse and edit files from the WebUI.
+      # Adapt the host path to your project directory.
+      - ${HERMES_WORKSPACE:-~/workspace}:/workspace
+    environment:
+      - HERMES_WEBUI_HOST=0.0.0.0
+      - HERMES_WEBUI_PORT=8787
+      - HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui
+      # Match your host user's UID/GID for correct file permissions.
+      # Run `id -u` and `id -g` to find your values.
+      # On macOS, UIDs start at 501 (not 1000) — set these in a .env file:
+      #   echo "UID=$(id -u)" >> .env && echo "GID=$(id -g)" >> .env
+      - WANTED_UID=${UID:-1000}
+      - WANTED_GID=${GID:-1000}
+      # NOTE: When using bind-mount volumes shared across containers, ALL containers
+      # that write to the same host directory must run as the same UID/GID.
+      # If hermes-agent initialises the state dir as root (UID 0), hermes-webui
+      # will get a PermissionError accessing those paths — including a crash on every
+      # HTTP request if the auth signing-key file is unreadable. Either set WANTED_UID
+      # to match the agent container's UID, or use a named Docker volume (preferred).
+      # Optional: set a password for remote access
+      # - HERMES_WEBUI_PASSWORD=your-secret-password
+    restart: unless-stopped
+    networks:
+      - hermes-net
+
+networks:
+  hermes-net:
+    driver: bridge
+
+volumes:
+  hermes-home:
+  hermes-agent-src:

docker-compose.two-container.yml

@@ -0,0 +1,95 @@
+# Two-container Docker Compose: Hermes Agent + Hermes WebUI
+#
+# This runs the agent and web UI in separate containers connected via
+# shared volumes. The WebUI installs the agent's Python dependencies
+# at startup from the shared agent source volume.
+#
+# Usage:
+#   docker compose -f docker-compose.two-container.yml up -d
+#
+# The agent container runs the gateway (CLI, Telegram, cron, etc.).
+# The WebUI container serves the browser interface on port 8787.
+# Both share ~/.hermes for config, sessions, and state.
+#
+# NOTE ON VOLUMES:
+#   This file uses named Docker volumes (hermes-home, hermes-agent-src) which
+#   work out of the box. If you prefer bind mounts (e.g. to an existing directory),
+#   replace the named volumes at the bottom. Example for hermes-agent-src:
+#
+#     hermes-agent-src:
+#       driver: local
+#       driver_opts:
+#         type: none
+#         o: bind
+#         device: /opt/hermes-agent
+#
+#   When using bind mounts, BOTH containers must mount the same host path.
+#   The agent exposes source at /opt/hermes, the WebUI reads it from
+#   /home/hermeswebui/.hermes/hermes-agent — as long as both point to the
+#   same host directory, the paths align correctly.
+
+services:
+  hermes-agent:
+    image: nousresearch/hermes-agent:latest
+    container_name: hermes-agent
+    command: gateway run
+    ports:
+      # Gateway API — exposed on localhost only.
+      # Other containers on hermes-net reach it via http://hermes-agent:8642.
+      # Remove 127.0.0.1: to expose on the host network (e.g. for remote clients).
+      - "127.0.0.1:8642:8642"
+    volumes:
+      # Persist config, state, sessions, skills, memory across restarts
+      - hermes-home:/home/hermes/.hermes
+      # Expose agent source so the WebUI can install dependencies from it
+      - hermes-agent-src:/opt/hermes
+    environment:
+      - HERMES_HOME=/home/hermes/.hermes
+    restart: unless-stopped
+    networks:
+      - hermes-net
+
+  hermes-webui:
+    image: ghcr.io/nesquena/hermes-webui:latest
+    container_name: hermes-webui
+    depends_on:
+      - hermes-agent
+    ports:
+      - "127.0.0.1:8787:8787"
+    volumes:
+      # Same hermes home as the agent — shares config, sessions, state
+      - hermes-home:/home/hermeswebui/.hermes
+      # Agent source mounted where docker_init.bash expects it.
+      # At startup the init script runs:
+      #   uv pip install /home/hermeswebui/.hermes/hermes-agent
+      # which installs the agent and all its Python dependencies.
+      - hermes-agent-src:/home/hermeswebui/.hermes/hermes-agent
+      # Workspace directory — browse and edit files from the WebUI.
+      # Adapt the host path to your project directory.
+      # Override with: HERMES_WORKSPACE=/your/path docker compose up
+      - ${HERMES_WORKSPACE:-~/workspace}:/workspace
+    environment:
+      - HERMES_WEBUI_HOST=0.0.0.0
+      - HERMES_WEBUI_PORT=8787
+      - HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui
+      # Match your host user's UID/GID for correct file permissions.
+      # In two-container setups the WebUI auto-detects UID/GID from the shared
+      # hermes-home volume, but you can override explicitly if needed (#668):
+      # Run `id -u` and `id -g` to find your values.
+      # On macOS, UIDs start at 501 — set these in a .env file:
+      #   echo "UID=$(id -u)" >> .env && echo "GID=$(id -g)" >> .env
+      - WANTED_UID=${UID:-1000}
+      - WANTED_GID=${GID:-1000}
+      # Optional: set a password for remote access
+      # - HERMES_WEBUI_PASSWORD=***
+    restart: unless-stopped
+    networks:
+      - hermes-net
+
+networks:
+  hermes-net:
+    driver: bridge
+
+volumes:
+  hermes-home:
+  hermes-agent-src:

docker-compose.yml

@@ -4,19 +4,34 @@ services:
   hermes-webui:
     build: .
     ports:
+    # select only one; use 127.0.0.1 version to expose to localhost only
       - "127.0.0.1:8787:8787"
+#      - "8787:8787"
     volumes:
-      # Persist session data, settings, and projects across restarts
-      - hermes-data:/data
-      # Mount hermes home for agent features and profile management
-      - ${HERMES_HOME:-${HOME}/.hermes}:/root/.hermes
+      # Mount your Hermes home directory into the container.
+      # The default (${HOME}/.hermes) works on both macOS (/Users/<you>/.hermes)
+      # and Linux (/home/<you>/.hermes) — no change needed for standard installs.
+      # Only set HERMES_HOME explicitly if your .hermes lives somewhere non-standard.
+      # macOS note: set UID and GID below to match your user ID (run `id -u` and `id -g`).
+      - ${HERMES_HOME:-${HOME}/.hermes}:/home/hermeswebui/.hermes
+      # Your workspace directory shown on first launch (adapt if yours is different, the container will use the mounted /workspace)
+      - ${HERMES_WORKSPACE:-${HOME}/workspace}:/workspace
     environment:
+      # Set to your host user ID: run `id -u` and `id -g` to find them.
+      # On macOS, UIDs start at 501 (not 1000), so set UID and GID in a .env file:
+      #   echo "UID=$(id -u)" >> .env
+      #   echo "GID=$(id -g)" >> .env
+      # Without this, the container may not be able to read your mounted files.
+      - WANTED_UID=${UID:-1000}
+      - WANTED_GID=${GID:-1000}
+      # Required: bind address and port
       - HERMES_WEBUI_HOST=0.0.0.0
       - HERMES_WEBUI_PORT=8787
-      - HERMES_WEBUI_STATE_DIR=/data
+      # Where to store sessions, workspaces, and other state (default: ~/.hermes/webui)
+      - HERMES_WEBUI_STATE_DIR=/home/hermeswebui/.hermes/webui
+      # Default workspace directory shown on first launch
+#      - HERMES_WEBUI_DEFAULT_WORKSPACE=/workspace
       # Optional: set a password for remote access
       # - HERMES_WEBUI_PASSWORD=your-secret-password
     restart: unless-stopped
 
-volumes:
-  hermes-data:

docker_init.bash

@@ -0,0 +1,330 @@
+#!/bin/bash
+
+set -e
+
+error_exit() {
+  echo -n "!! ERROR: "
+  echo $*
+  echo "!! Exiting script (ID: $$)"
+  exit 1
+}
+
+ok_exit() {
+  echo $*
+  echo "++ Exiting script (ID: $$)"
+  exit 0
+}
+
+## Environment variables loaded when passing environment variables from user to user
+# Ignore list: variables to ignore when loading environment variables from user to user
+export ENV_IGNORELIST="HOME PWD USER SHLVL TERM OLDPWD SHELL _ SUDO_COMMAND HOSTNAME LOGNAME MAIL SUDO_GID SUDO_UID SUDO_USER CHECK_NV_CUDNN_VERSION VIRTUAL_ENV VIRTUAL_ENV_PROMPT ENV_IGNORELIST ENV_OBFUSCATE_PART"
+# Obfuscate part: part of the key to obfuscate when loading environment variables from user to user, ex: HF_TOKEN, ...
+export ENV_OBFUSCATE_PART="TOKEN API KEY"
+
+# Check for ENV_IGNORELIST and ENV_OBFUSCATE_PART
+if [ -z "${ENV_IGNORELIST+x}" ]; then error_exit "ENV_IGNORELIST not set"; fi
+if [ -z "${ENV_OBFUSCATE_PART+x}" ]; then error_exit "ENV_OBFUSCATE_PART not set"; fi
+
+whoami=`whoami`
+script_dir=$(dirname $0)
+script_name=$(basename $0)
+echo ""; echo ""
+echo "======================================"
+echo "=================== Starting script (ID: $$)"
+echo "== Running ${script_name} in ${script_dir} as ${whoami}"
+script_fullname=$0
+echo "  - script_fullname: ${script_fullname}"
+ignore_value="VALUE_TO_IGNORE"
+
+# everyone can read our files by default
+umask 0022
+
+# Write a world-writeable file (preferably inside /tmp -- ie within the container)
+write_worldtmpfile() {
+  tmpfile=$1
+  if [ -z "${tmpfile}" ]; then error_exit "write_worldfile: missing argument"; fi
+  if [ -f $tmpfile ]; then rm -f $tmpfile; fi
+  echo -n $2 > ${tmpfile}
+  chmod 777 ${tmpfile}
+}
+
+itdir=/tmp/hermeswebui_init
+if [ ! -d $itdir ]; then mkdir $itdir; chmod 777 $itdir; fi
+if [ ! -d $itdir ]; then error_exit "Failed to create $itdir"; fi
+
+# Set user and group id
+# logic: if not set and file exists, use file value, else use default. Create file for persistence when the container is re-run
+# reasoning: needed when using docker compose as the file will exist in the stopped container, and changing the value from environment variables or configuration file must be propagated from hermeswebuitoo to hermeswebuitoo transition (those values are the only ones loaded before the environment variables dump file are loaded)
+it=$itdir/hermeswebui_user_uid
+if [ -z "${WANTED_UID+x}" ]; then
+  if [ -f $it ]; then WANTED_UID=$(cat $it); fi
+fi
+# Auto-detect from mounted volumes if still unset (#569, #668).
+# On macOS, host UIDs start at 501. Using the wrong UID means the container
+# user cannot read the bind-mounted files, making the workspace appear empty.
+# In two-container setups (hermes-agent + hermes-webui), the shared hermes-home
+# volume may be owned by the agent container's UID — detect from there first.
+if [ -z "${WANTED_UID+x}" ] || [ "${WANTED_UID}" = "1024" ]; then
+  # Priority 1: hermes-home shared volume — covers two-container Zeabur/Compose setups (#668)
+  for _probe_dir in "/home/hermeswebui/.hermes" "$HERMES_HOME" "/opt/data"; do
+    if [ -d "$_probe_dir" ]; then
+      _detected_uid=$(stat -c '%u' "$_probe_dir" 2>/dev/null || echo "")
+      if [ -n "$_detected_uid" ] && [ "$_detected_uid" != "0" ]; then
+        echo "-- Auto-detected UID: $_detected_uid (from $_probe_dir)"
+        WANTED_UID=$_detected_uid
+        break
+      fi
+    fi
+  done
+fi
+if [ -z "${WANTED_UID+x}" ] || [ "${WANTED_UID}" = "1024" ]; then
+  # Priority 2: /workspace bind-mount — the standard single-container mount point
+  if [ -d "/workspace" ]; then
+    _detected_uid=$(stat -c '%u' "/workspace" 2>/dev/null || echo "")
+    if [ -n "$_detected_uid" ] && [ "$_detected_uid" != "0" ]; then
+      echo "-- Auto-detected workspace UID: $_detected_uid (from /workspace)"
+      WANTED_UID=$_detected_uid
+    fi
+  fi
+fi
+WANTED_UID=${WANTED_UID:-1024}
+write_worldtmpfile $it "$WANTED_UID"
+echo "-- WANTED_UID: \"${WANTED_UID}\""
+
+it=$itdir/hermeswebui_user_gid
+if [ -z "${WANTED_GID+x}" ]; then
+  if [ -f $it ]; then WANTED_GID=$(cat $it); fi
+fi
+# Auto-detect GID from mounted volumes to match (#569, #668)
+if [ -z "${WANTED_GID+x}" ] || [ "${WANTED_GID}" = "1024" ]; then
+  # Priority 1: hermes-home shared volume
+  for _probe_dir in "/home/hermeswebui/.hermes" "$HERMES_HOME" "/opt/data"; do
+    if [ -d "$_probe_dir" ]; then
+      _detected_gid=$(stat -c '%g' "$_probe_dir" 2>/dev/null || echo "")
+      if [ -n "$_detected_gid" ] && [ "$_detected_gid" != "0" ]; then
+        echo "-- Auto-detected GID: $_detected_gid (from $_probe_dir)"
+        WANTED_GID=$_detected_gid
+        break
+      fi
+    fi
+  done
+fi
+if [ -z "${WANTED_GID+x}" ] || [ "${WANTED_GID}" = "1024" ]; then
+  # Priority 2: /workspace bind-mount
+  if [ -d "/workspace" ]; then
+    _detected_gid=$(stat -c '%g' "/workspace" 2>/dev/null || echo "")
+    if [ -n "$_detected_gid" ] && [ "$_detected_gid" != "0" ]; then
+      echo "-- Auto-detected workspace GID: $_detected_gid (from /workspace)"
+      WANTED_GID=$_detected_gid
+    fi
+  fi
+fi
+WANTED_GID=${WANTED_GID:-1024}
+write_worldtmpfile $it "$WANTED_GID"
+echo "-- WANTED_GID: \"${WANTED_GID}\""
+
+echo "== Most Environment variables set"
+
+# Check user id and group id
+new_gid=`id -g`
+new_uid=`id -u`
+echo "== user ($whoami)"
+echo "  uid: $new_uid / WANTED_UID: $WANTED_UID"
+echo "  gid: $new_gid / WANTED_GID: $WANTED_GID"
+
+save_env() {
+  tosave=$1
+  echo "-- Saving environment variables to $tosave"
+  env | sort > "$tosave"
+}
+
+load_env() {
+  tocheck=$1
+  overwrite_if_different=$2
+  ignore_list="${ENV_IGNORELIST}"
+  obfuscate_part="${ENV_OBFUSCATE_PART}"
+  if [ -f "$tocheck" ]; then
+    echo "-- Loading environment variables from $tocheck (overwrite existing: $overwrite_if_different) (ignorelist: $ignore_list) (obfuscate: $obfuscate_part)"
+    while IFS='=' read -r key value; do
+      doit=false
+      # checking if the key is in the ignorelist
+      for i in $ignore_list; do
+        if [[ "A$key" ==  "A$i" ]]; then doit=ignore; break; fi
+      done
+      if [[ "A$doit" == "Aignore" ]]; then continue; fi
+      rvalue=$value
+      # checking if part of the key is in the obfuscate list
+      doobs=false
+      for i in $obfuscate_part; do
+        if [[ "A$key" == *"$i"* ]]; then doobs=obfuscate; break; fi
+      done
+      if [[ "A$doobs" == "Aobfuscate" ]]; then rvalue="**OBFUSCATED**"; fi
+
+      if [ -z "${!key}" ]; then
+        echo "  ++ Setting environment variable $key [$rvalue]"
+        doit=true
+      elif [ "A$overwrite_if_different" == "Atrue" ]; then
+        cvalue="${!key}"
+        if [[ "A${doobs}" == "Aobfuscate" ]]; then cvalue="**OBFUSCATED**"; fi
+        if [[ "A${!key}" != "A${value}" ]]; then
+          echo "  @@ Overwriting environment variable $key [$cvalue] -> [$rvalue]"
+          doit=true
+        else
+          echo "  == Environment variable $key [$rvalue] already set and value is unchanged"
+        fi
+      fi
+      if [[ "A$doit" == "Atrue" ]]; then
+        export "$key=$value"
+      fi
+    done < "$tocheck"
+  fi
+}
+
+# hermeswebuitoo is a specfiic user not existing by default on ubuntu, we can check its whomai
+if [ "A${whoami}" == "Ahermeswebuitoo" ]; then 
+  echo "-- Running as hermeswebuitoo, will switch hermeswebui to the desired UID/GID"
+  # The script is started as hermeswebuitoo -- UID/GID 1025/1025
+
+  # We are altering the UID/GID of the hermeswebui user to the desired ones and restarting as that user
+  # using usermod for the already create hermeswebui user, knowing it is not already in use
+  # per usermod manual: "You must make certain that the named user is not executing any processes when this command is being executed"
+  sudo groupmod -o -g ${WANTED_GID} hermeswebui || error_exit "Failed to set GID of hermeswebui user"
+  sudo usermod -o -u ${WANTED_UID} hermeswebui || error_exit "Failed to set UID of hermeswebui user"
+  sudo chown -R ${WANTED_UID}:${WANTED_GID} /home/hermeswebui || error_exit "Failed to set owner of /home/hermeswebui"
+  save_env /tmp/hermeswebuitoo_env.txt  
+  # restart the script as hermeswebui set with the correct UID/GID this time
+  echo "-- Restarting as hermeswebui user with UID ${WANTED_UID} GID ${WANTED_GID}"
+  sudo su hermeswebui $script_fullname || error_exit "subscript failed"
+  ok_exit "Clean exit"
+fi
+
+# If we are here, the script is started as another user than hermeswebuitoo
+# because the whoami value for the hermeswebui user can be any existing user, we can not check against it
+# instead we check if the UID/GID are the expected ones
+if [ "$WANTED_GID" != "$new_gid" ]; then error_exit "hermeswebui MUST be running as UID ${WANTED_UID} GID ${WANTED_GID}, current UID ${new_uid} GID ${new_gid}"; fi
+if [ "$WANTED_UID" != "$new_uid" ]; then error_exit "hermeswebui MUST be running as UID ${WANTED_UID} GID ${WANTED_GID}, current UID ${new_uid} GID ${new_gid}"; fi
+
+########## 'hermeswebui' specific section below
+
+# We are therefore running as hermeswebui
+echo ""; echo "== Running as hermeswebui"
+
+# Load environment variables one by one if they do not exist from /tmp/hermeswebuitoo_env.txt
+it=/tmp/hermeswebuitoo_env.txt
+if [ -f $it ]; then
+  echo "-- Loading not already set environment variables from $it"
+  load_env $it true
+fi
+
+##
+echo ""; echo "-- Making sure /app is owned by the hermeswebui user to avoid permission issues when running the server "
+sudo mkdir -p /app || error_exit "Failed to create /app directory"
+sudo chown hermeswebui:hermeswebui /app || error_exit "Failed to set owner of /app to hermeswebui user"
+sudo rsync -av --chown=hermeswebui:hermeswebui /apptoo/ /app/ || error_exit "Failed to sync /apptoo to /app with correct ownership"
+it=/app/.testfile; touch $it || error_exit "Failed to verify /app directory"
+rm -f $it || error_exit "Failed to delete test file in /app"
+
+######## Environment variables (consume AFTER the load_env)
+
+echo ""; echo "== Checking required environment variables for hermes-webui"
+
+echo ""; echo "-- HERMES_WEBUI_VERSION: Where to store sessions, workspaces, and other state (default: ~/.hermes/webui-mvp)"
+if [ -z "${HERMES_WEBUI_STATE_DIR+x}" ]; then error_exit "HERMES_WEBUI_STATE_DIR not set"; fi; 
+echo "-- HERMES_WEBUI_STATE_DIR: $HERMES_WEBUI_STATE_DIR"
+if [ ! -d "$HERMES_WEBUI_STATE_DIR" ]; then mkdir -p $HERMES_WEBUI_STATE_DIR || error_exit "Failed to create state directory at $HERMES_WEBUI_STATE_DIR"; fi
+if [ ! -d "$HERMES_WEBUI_STATE_DIR" ]; then error_exit "HERMES_WEBUI_STATE_DIR directory does not exist at $HERMES_WEBUI_STATE_DIR"; fi
+it="$HERMES_WEBUI_STATE_DIR/.testfile"; touch $it || error_exit "Failed to verify state directory at $HERMES_WEBUI_STATE_DIR"
+rm -f $it || error_exit "Failed to delete test file in $HERMES_WEBUI_STATE_DIR"
+
+echo ""; echo "-- HERMES_WEBUI_DEFAULT_WORKSPACE: Default workspace directory shown on first launch"
+if [ -z "${HERMES_WEBUI_DEFAULT_WORKSPACE+x}" ]; then echo "HERMES_WEBUI_DEFAULT_WORKSPACE not set, setting to /workspace"; export HERMES_WEBUI_DEFAULT_WORKSPACE="/workspace"; fi;
+echo "-- HERMES_WEBUI_DEFAULT_WORKSPACE: $HERMES_WEBUI_DEFAULT_WORKSPACE"
+# Use sudo for mkdir — Docker may auto-create bind-mount directories as root (#357).
+# Skip mkdir if the directory already exists (e.g. a read-only mount — #670).
+if [ ! -d "$HERMES_WEBUI_DEFAULT_WORKSPACE" ]; then
+  sudo mkdir -p "$HERMES_WEBUI_DEFAULT_WORKSPACE" || error_exit "Failed to create default workspace at $HERMES_WEBUI_DEFAULT_WORKSPACE"
+fi
+if [ ! -d "$HERMES_WEBUI_DEFAULT_WORKSPACE" ]; then error_exit "HERMES_WEBUI_DEFAULT_WORKSPACE directory does not exist at $HERMES_WEBUI_DEFAULT_WORKSPACE"; fi
+# Only chown and write-test if the workspace is writable. Read-only bind-mounts
+# (:ro) are valid — the workspace is used for browsing, not writing by the server.
+if [ -w "$HERMES_WEBUI_DEFAULT_WORKSPACE" ]; then
+  sudo chown hermeswebui:hermeswebui "$HERMES_WEBUI_DEFAULT_WORKSPACE" || echo "!! WARNING: Could not chown $HERMES_WEBUI_DEFAULT_WORKSPACE (continuing)"
+  it="$HERMES_WEBUI_DEFAULT_WORKSPACE/.testfile"; touch $it && rm -f $it || echo "!! WARNING: Could not write to $HERMES_WEBUI_DEFAULT_WORKSPACE (continuing)"
+else
+  echo "-- HERMES_WEBUI_DEFAULT_WORKSPACE is read-only — skipping chown/write check (read-only workspace is supported)"
+fi
+
+echo ""; echo "==================="
+echo ""; echo "== Installing uv and creating a new virtual environment for hermes-webui"
+
+export PATH="/home/hermeswebui/.local/bin/:$PATH"
+if command -v uv &>/dev/null; then
+  echo "-- uv already installed ($(uv --version)), skipping download"
+else
+  echo "-- uv not found, downloading..."
+  curl -LsSf https://astral.sh/uv/install.sh | sh || error_exit "Failed to install uv — check network connectivity"
+fi
+export UV_PROJECT_ENVIRONMENT=venv
+
+export UV_CACHE_DIR=/uv_cache
+sudo mkdir -p ${UV_CACHE_DIR} || error_exit "Failed to create /uv_cache directory"
+sudo chown hermeswebui:hermeswebui ${UV_CACHE_DIR} || error_exit "Failed to set owner of ${UV_CACHE_DIR} to hermeswebui user"
+
+cd /app
+if [ -f /app/venv/bin/python3 ]; then
+  echo ""; echo "== Existing virtual environment found — reusing (fast restart)"
+else
+  echo ""; echo "== Creating new virtual environment"
+  uv venv venv
+fi
+export VIRTUAL_ENV=/app/venv
+test -d /app/venv
+test -f /app/venv/bin/activate
+
+echo "";echo "== Activating hermes webui's virtual environment"
+source /app/venv/bin/activate || error_exit "Failed to activate hermeswebui virtual environment"
+test -x /app/venv/bin/python3
+
+if [ -f /app/venv/.deps_installed ]; then
+  echo ""; echo "== Dependencies already installed — skipping (fast restart)"
+else
+  echo ""; echo "== Installing hermes-webui dependencies"
+  uv pip install -r requirements.txt --trusted-host pypi.org --trusted-host files.pythonhosted.org
+  uv pip install -U pip setuptools --trusted-host pypi.org --trusted-host files.pythonhosted.org
+  test -x /app/venv/bin/pip
+
+  echo ""; echo "== Adding hermes-agent's pyproject.toml base dependencies to the virtual environment"
+  _agent_paths=(
+    "/home/hermeswebui/.hermes/hermes-agent"
+    "/opt/hermes"
+  )
+  _agent_src=""
+  for _p in "${_agent_paths[@]}"; do
+    if [ -d "$_p" ] && [ -f "$_p/pyproject.toml" ]; then
+      _agent_src="$_p"
+      break
+    fi
+  done
+  if [ -n "$_agent_src" ]; then
+    uv pip install "$_agent_src[all]" --trusted-host pypi.org --trusted-host files.pythonhosted.org || error_exit "Failed to install hermes-agent's requirements"
+  else
+    echo ""
+    echo "!! WARNING: hermes-agent source not found."
+    echo "!!   Looked in: ${_agent_paths[0]}"
+    echo "!!              ${_agent_paths[1]}"
+    echo "!! The WebUI will start with reduced functionality (no model auto-detection,"
+    echo "!! no personality routing, no CLI session imports)."
+    echo "!! To fix: mount the agent source volume into the container:"
+    echo "!!   -v /path/to/hermes-agent:/home/hermeswebui/.hermes/hermes-agent"
+    echo "!! Or see the two-container compose example:"
+    echo "!!   https://github.com/nesquena/hermes-webui/blob/master/docker-compose.two-container.yml"
+    echo ""
+  fi
+  touch /app/venv/.deps_installed
+fi
+
+echo ""; echo "== Running hermes-webui"
+cd /app; python server.py || error_exit "hermes-webui failed or exited with an error"
+
+# we should never be here because the server should be running indefinitely, but if we are, we exit safely
+ok_exit "Clean exit"

docs/ui-ux/index.html

@@ -0,0 +1,838 @@
+<!doctype html>
+<html lang="en" data-theme="slate">
+<head>
+  <meta charset="utf-8">
+  <title>Hermes WebUI — Messages UI Inventory</title>
+  <meta name="viewport" content="width=device-width,initial-scale=1">
+  <!-- Real app stylesheet -->
+  <link rel="stylesheet" href="../../static/style.css">
+  <!-- Prism (same theme the app pulls at runtime) -->
+  <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/prism/1.29.0/themes/prism-tomorrow.min.css">
+  <!-- KaTeX -->
+  <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.16.9/dist/katex.min.css">
+  <style>
+    /* Showcase scaffold — styles only for the doc chrome. Everything inside
+       .messages uses the real app CSS unchanged. */
+    /* Real app CSS makes <body> a fixed-height flex shell. Undo that so this
+       doc page can scroll normally with a stacked header + main. */
+    body{display:block !important;height:auto !important;min-height:100vh;overflow:auto !important;}
+    .doc-main{display:block;}
+    .doc-header{position:sticky;top:0;z-index:50;background:var(--topbar-bg);backdrop-filter:blur(12px);border-bottom:1px solid var(--border);padding:14px 24px;display:flex;flex-wrap:wrap;align-items:center;gap:14px;}
+    .doc-title{font-size:16px;font-weight:700;letter-spacing:-.01em;color:var(--text);}
+    .doc-title small{display:block;font-size:11px;font-weight:500;color:var(--muted);margin-top:3px;}
+    .doc-toggles{display:flex;flex-wrap:wrap;gap:6px;margin-left:auto;}
+    .doc-toggles button{font:inherit;font-size:11px;padding:5px 10px;border-radius:7px;border:1px solid var(--border2);background:var(--input-bg);color:var(--muted);cursor:pointer;}
+    .doc-toggles button.on{background:rgba(124,185,255,.12);border-color:rgba(124,185,255,.4);color:var(--blue);}
+    .doc-main{max-width:1100px;margin:0 auto;padding:24px 24px 120px;}
+    .doc-section{margin:40px 0 8px;padding-top:20px;border-top:1px dashed var(--border);}
+    .doc-section:first-of-type{border-top:none;padding-top:0;margin-top:0;}
+    .doc-kicker{font-size:10px;font-weight:700;letter-spacing:.14em;text-transform:uppercase;color:var(--blue);}
+    .doc-h{font-size:18px;font-weight:700;color:var(--text);margin:4px 0 4px;}
+    .doc-note{font-size:12px;color:var(--muted);line-height:1.55;max-width:760px;margin-bottom:10px;}
+    .doc-card{position:relative;background:var(--main-bg);border:1px solid var(--border);border-radius:12px;padding:4px 6px;margin:12px 0;}
+    .doc-label{position:absolute;top:-9px;left:12px;font-size:10px;font-weight:700;text-transform:uppercase;letter-spacing:.08em;padding:2px 8px;background:var(--bg);color:var(--muted);border:1px solid var(--border);border-radius:999px;}
+    /* Force-show hover-only affordances inside explicitly flagged demos */
+    .force-show .msg-actions,
+    .force-show .msg-time,
+    .force-show .msg-foot{opacity:1 !important;}
+    /* Chat demo container mimics the app's .messages scroll wrapper but not fullscreen */
+    .messages.doc-messages{overflow:visible;display:block;}
+    .messages-inner.doc-inner{padding:14px 16px;}
+    /* Make the in-page demos of approval/clarify cards visible without JS */
+    .approval-card.doc-visible,
+    .clarify-card.doc-visible{display:block;}
+    .reconnect-banner.doc-visible{display:flex;align-items:center;justify-content:space-between;gap:12px;background:rgba(201,168,76,.12);border:1px solid rgba(201,168,76,.3);color:var(--gold);padding:8px 14px;border-radius:8px;font-size:12px;}
+    .reconnect-banner.doc-visible .reconnect-btn{background:none;border:1px solid rgba(201,168,76,.35);color:var(--gold);padding:4px 10px;border-radius:6px;font-size:11px;cursor:pointer;}
+    .bg-error-banner.doc-visible{border-radius:8px;}
+    /* Two-up grid for short comparisons */
+    .doc-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(320px,1fr));gap:12px;}
+  </style>
+</head>
+<body>
+
+<header class="doc-header">
+  <div class="doc-title">Hermes WebUI — Messages UI Inventory<small>Every message-area element &amp; combination, wired to the real <code>static/style.css</code>. &nbsp;·&nbsp; <a href="./two-stage-proposal.html" style="color:var(--blue);text-decoration:none;">Two-stage proposal (#536) →</a></small></div>
+  <div class="doc-toggles">
+    <strong style="font-size:10px;color:var(--muted);letter-spacing:.08em;text-transform:uppercase;align-self:center;margin-right:4px;">Theme</strong>
+    <button data-theme-btn="default">Default</button>
+    <button data-theme-btn="slate" class="on">Slate</button>
+    <button data-theme-btn="light">Light</button>
+    <button data-theme-btn="solarized">Solarized</button>
+    <button data-theme-btn="monokai">Monokai</button>
+    <button data-theme-btn="nord">Nord</button>
+    <button data-theme-btn="oled">OLED</button>
+    <span style="width:1px;height:18px;background:var(--border);margin:0 4px;align-self:center;"></span>
+  </div>
+</header>
+
+<main class="doc-main">
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">1 · Empty state</div>
+  <h2 class="doc-h">First load / no messages</h2>
+  <p class="doc-note">Renders inside <code>#messages</code> when <code>S.messages</code> is empty. Logo + title + subtitle + 3 suggestion buttons.</p>
+  <div class="doc-card"><span class="doc-label">.empty-state</span>
+    <div class="messages doc-messages">
+      <div class="empty-state" style="min-height:340px;flex:0 0 auto;">
+        <div class="empty-logo">H</div>
+        <h2>What can I help with?</h2>
+        <p>Ask anything, run commands, explore files, or manage your scheduled tasks.</p>
+        <div class="suggestion-grid">
+          <button class="suggestion">📁 What files are in this workspace?</button>
+          <button class="suggestion">📅 What's on my schedule today?</button>
+          <button class="suggestion">🗺️ Help me plan a small project.</button>
+        </div>
+      </div>
+    </div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">2 · User messages</div>
+  <h2 class="doc-h">Right-aligned bubble, attachments, and edit mode</h2>
+  <p class="doc-note">User rows have no avatar/label — the right-edge alignment and tinted bubble identify the sender. Timestamp + edit/copy live in a <code>.msg-foot</code> below the bubble, revealed on hover (forced visible here).</p>
+
+  <div class="doc-card"><span class="doc-label">.msg-row[data-role="user"] — plain</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row force-show" data-role="user" data-raw-text="How do I run the dev server and point it at a specific workspace path?">
+        <div class="msg-body"><p>How do I run the dev server and point it at a specific workspace path?</p></div>
+        <div class="msg-foot">
+          <span class="msg-time" title="Thu, Apr 16 2026, 10:42 AM">10:42</span>
+          <span class="msg-actions">
+            <button class="msg-action-btn" title="Edit"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 20h9"/><path d="M16.5 3.5a2.121 2.121 0 0 1 3 3L7 19l-4 1 1-4L16.5 3.5z"/></svg></button>
+            <button class="msg-copy-btn msg-action-btn" title="Copy"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="9" y="9" width="13" height="13" rx="2" ry="2"/><path d="M5 15H4a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2h9a2 2 0 0 1 2 2v1"/></svg></button>
+          </span>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.msg-files — attachments above body (right-aligned)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row" data-role="user">
+        <div class="msg-files">
+          <span class="msg-file-badge">📎 architecture-notes.pdf</span>
+          <span class="msg-file-badge">📎 Q1-forecast.xlsx</span>
+          <span class="msg-file-badge">📎 meeting.docx</span>
+          <span class="msg-file-badge">📎 screenshot.png</span>
+        </div>
+        <div class="msg-body"><p>Please review these docs and summarise the key decisions.</p></div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.msg-edit-area + .msg-edit-bar — edit mode</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row" data-role="user" data-editing="1">
+        <textarea class="msg-edit-area">How do I run the dev server and point it at a specific workspace path — and can I do it without docker?</textarea>
+        <div class="msg-edit-bar">
+          <button class="msg-edit-send">Send edit</button>
+          <button class="msg-edit-cancel">Cancel</button>
+        </div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">3 · Assistant — markdown basics</div>
+  <h2 class="doc-h">Paragraphs, emphasis, lists, blockquote, hr, links</h2>
+  <p class="doc-note">Assistant output is a single <code>.msg-row.assistant-turn</code> that holds one role header + an <code>.assistant-turn-blocks</code> column of one-or-more <code>.assistant-segment</code> children. Each segment may contain a <code>.thinking-card</code>, a <code>.msg-body</code>, and its own <code>.msg-foot</code> (copy / regen). This lets a turn stream reasoning → text → tool calls → more text without repeating the Hermes avatar each time.</p>
+
+  <div class="doc-card"><span class="doc-label">.msg-body — rich prose</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn force-show" data-role="assistant">
+        <div class="msg-role assistant" title="Thu, Apr 16 2026, 10:42 AM">
+          <span class="role-icon assistant">H</span>
+          <span>Hermes</span>
+        </div>
+        <div class="assistant-turn-blocks">
+          <div class="assistant-segment" data-raw-text="Running the dev server...">
+            <div class="msg-body">
+              <h1>Running the dev server</h1>
+              <p>You can start Hermes with the built-in launcher. The <strong>simplest path</strong> is <em>no docker, no proxy</em> — the CLI handles everything.</p>
+              <h2>Prerequisites</h2>
+              <ul>
+                <li>Node <code>&gt;= 18</code></li>
+                <li>A workspace directory you own
+                  <ul>
+                    <li>Read/write permissions</li>
+                    <li>No existing <code>.hermes</code> folder</li>
+                  </ul>
+                </li>
+                <li>An API key set via <code>HERMES_API_KEY</code></li>
+              </ul>
+              <h2>Steps</h2>
+              <ol>
+                <li>Clone the repo</li>
+                <li>Run <code>npm install</code></li>
+                <li>Start with <code>npm run dev -- --workspace ~/code</code></li>
+              </ol>
+              <blockquote>Tip: the <code>--workspace</code> flag accepts absolute or <code>~</code>-prefixed paths. Relative paths are resolved against the CWD.</blockquote>
+              <hr>
+              <p>For full setup options see the <a href="#">configuration guide</a>.</p>
+            </div>
+            <div class="msg-foot">
+              <span class="msg-actions">
+                <button class="msg-copy-btn msg-action-btn" title="Copy"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="9" y="9" width="13" height="13" rx="2" ry="2"/><path d="M5 15H4a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2h9a2 2 0 0 1 2 2v1"/></svg></button>
+                <button class="msg-action-btn" title="Regenerate"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><polyline points="1 4 1 10 7 10"/><path d="M3.51 15a9 9 0 1 0 2.13-9.36L1 10"/></svg></button>
+              </span>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.msg-body table</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks">
+          <div class="assistant-segment">
+            <div class="msg-body">
+              <p>Model comparison:</p>
+              <table>
+                <thead><tr><th>Model</th><th>Context</th><th>Good for</th><th>Cost / 1M in</th></tr></thead>
+                <tbody>
+                  <tr><td>Opus 4.6</td><td>1M</td><td>Deep reasoning, long code</td><td><code>$15.00</code></td></tr>
+                  <tr><td>Sonnet 4.6</td><td>1M</td><td>Daily driver, agents</td><td><code>$3.00</code></td></tr>
+                  <tr><td>Haiku 4.5</td><td>200k</td><td>Fast tasks, tool loops</td><td><code>$0.80</code></td></tr>
+                </tbody>
+              </table>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">4 · Code blocks</div>
+  <h2 class="doc-h">Plain, with header, with copy button, multi-language</h2>
+
+  <div class="doc-card"><span class="doc-label">pre + code (no header)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+<pre><code class="language-bash">npm install
+npm run dev -- --workspace ~/code</code></pre>
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.pre-header + pre + .code-copy-btn</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+            <div style="position:relative;">
+              <div class="pre-header">typescript <button class="code-copy-btn" style="margin-left:auto;">Copy</button></div>
+<pre><code class="language-typescript">export async function startServer(opts: ServerOptions) {
+  const port = opts.port ?? 3000;
+  const app = createApp();
+  app.listen(port, () =&gt; {
+    console.log(`Hermes listening on :${port}`);
+  });
+  return app;
+}</code></pre>
+            </div>
+            <div style="position:relative;margin-top:14px;">
+              <div class="pre-header">python <button class="code-copy-btn" style="margin-left:auto;">Copy</button></div>
+<pre><code class="language-python">from hermes import Agent
+
+def main() -&gt; None:
+    agent = Agent(model="claude-opus-4-6")
+    reply = agent.run("Summarise today's commits")
+    print(reply)
+
+if __name__ == "__main__":
+    main()</code></pre>
+            </div>
+            <div style="position:relative;margin-top:14px;">
+              <div class="pre-header">json <button class="code-copy-btn" style="margin-left:auto;">Copy</button></div>
+<pre><code class="language-json">{
+  "model": "claude-sonnet-4-6",
+  "stream": true,
+  "tools": ["bash", "edit_file", "search"]
+}</code></pre>
+            </div>
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">5 · Inline media</div>
+  <h2 class="doc-h">Images (default &amp; zoomed) and downloadable links</h2>
+
+  <div class="doc-card"><span class="doc-label">.msg-media-img (default + .msg-media-img--full)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+            <p>Here's the screenshot you asked for (click to zoom):</p>
+            <img class="msg-media-img" alt="demo" src="data:image/svg+xml;utf8,%3Csvg xmlns='http://www.w3.org/2000/svg' width='640' height='360'%3E%3Cdefs%3E%3ClinearGradient id='g' x1='0' x2='1'%3E%3Cstop offset='0' stop-color='%237cb9ff'/%3E%3Cstop offset='1' stop-color='%23c9a84c'/%3E%3C/linearGradient%3E%3C/defs%3E%3Crect fill='url(%23g)' width='640' height='360'/%3E%3Ctext x='50%25' y='50%25' font-family='system-ui' font-size='28' fill='white' text-anchor='middle' dominant-baseline='middle'%3E.msg-media-img (480×400 cap)%3C/text%3E%3C/svg%3E">
+            <p style="margin-top:10px;">And the full-width variant:</p>
+            <img class="msg-media-img msg-media-img--full" alt="demo-full" src="data:image/svg+xml;utf8,%3Csvg xmlns='http://www.w3.org/2000/svg' width='1280' height='320'%3E%3Crect fill='%231e2023' width='1280' height='320'/%3E%3Ctext x='50%25' y='50%25' font-family='system-ui' font-size='28' fill='%2382aaff' text-anchor='middle' dominant-baseline='middle'%3E.msg-media-img--full (unbounded)%3C/text%3E%3C/svg%3E">
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.msg-media-link — non-image downloads</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+            <p>I saved the generated files:</p>
+            <p><a class="msg-media-link" href="#">📎 report-2026-Q1.pdf</a> <a class="msg-media-link" href="#">📎 revenue.csv</a> <a class="msg-media-link" href="#">📎 diagram.svg</a></p>
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">6 · Math &amp; diagrams</div>
+  <h2 class="doc-h">KaTeX inline / block &amp; Mermaid block</h2>
+
+  <div class="doc-card"><span class="doc-label">.katex-inline + .katex-block</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+            <p>Inline math: <span class="katex-inline" data-math-inline>\(E = mc^2\)</span> and the quadratic formula below:</p>
+            <div class="katex-block" data-math-block>$$x = \frac{-b \pm \sqrt{b^2 - 4ac}}{2a}$$</div>
+            <p>A tidier form: <span class="katex-inline" data-math-inline>\(\sum_{i=1}^{n} i = \frac{n(n+1)}{2}\)</span>.</p>
+            <div class="katex-block" data-math-block>$$\int_{-\infty}^{\infty} e^{-x^2}\,dx = \sqrt{\pi}$$</div>
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.mermaid-block (pre-render placeholder)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body">
+            <p>The request flow:</p>
+            <div class="mermaid-block"><pre style="margin:0;background:none;border:none;padding:0;color:var(--muted);font-family:'SF Mono',ui-monospace,monospace;font-size:12px;">graph LR
+  U[User] --&gt; C[Composer]
+  C --&gt; API[/api/chat/]
+  API --&gt; M((Model))
+  M --&gt; T{tool?}
+  T -- yes --&gt; X[Tool Runner]
+  T -- no  --&gt; R[Reply]
+  X --&gt; R
+  R --&gt; U</pre></div>
+          </div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">7 · Thinking / reasoning</div>
+  <h2 class="doc-h">Bordered panel (collapsed / open, animated), live loader, streaming cursor</h2>
+  <p class="doc-note">Thinking cards are rendered at the top of an <code>.assistant-segment</code>. They're now bordered gold-tinted panels (no more left-rule-only look) and expand/collapse with a <code>max-height</code> + opacity transition. Click the header in either example below to see the animation live.</p>
+
+  <div class="doc-grid">
+    <div class="doc-card"><span class="doc-label">.thinking-card (collapsed, inside .assistant-segment)</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner" style="padding-top:8px;">
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+            <div class="thinking-card">
+              <div class="thinking-card-header">
+                <span class="thinking-card-icon">💡</span>
+                <span class="thinking-card-label">Thought for 4.3s</span>
+                <span class="thinking-card-toggle">▶</span>
+              </div>
+              <div class="thinking-card-body"><pre>The user asked about the dev server...</pre></div>
+            </div>
+            <div class="msg-body"><p>Here's the shortest path…</p></div>
+          </div></div>
+        </div>
+      </div></div>
+    </div>
+
+    <div class="doc-card"><span class="doc-label">.thinking-card.open (animated — max-height + opacity)</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner" style="padding-top:8px;">
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+            <div class="thinking-card open">
+              <div class="thinking-card-header">
+                <span class="thinking-card-icon">💡</span>
+                <span class="thinking-card-label">Thought for 4.3s</span>
+                <span class="thinking-card-toggle">▶</span>
+              </div>
+              <div class="thinking-card-body"><pre>The user is asking about launching the dev server.
+Options: npm script, docker, or the bundled CLI.
+The CLI is the simplest — no container runtime needed.
+I should show the exact commands and the --workspace flag,
+then mention the env var for the API key at the end.</pre></div>
+            </div>
+            <div class="msg-body"><p>Here's the shortest path…</p></div>
+          </div></div>
+        </div>
+      </div></div>
+    </div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.thinking — live 3-dot loader (pre-reasoning)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment" data-live-assistant="1">
+          <div class="thinking">Thinking <span class="dot"></span><span class="dot"></span><span class="dot"></span></div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">[data-live-assistant="1"] — streaming cursor at end of last child</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant" id="liveAssistantTurn">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment" data-live-assistant="1">
+          <div class="msg-body"><p>Sure — the simplest way is to run <code>npm run dev</code>. The CLI will pick up the default</p></div>
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">8 · Tool cards</div>
+  <h2 class="doc-h">Running, done, expanded, subagent, error, multi-card toggle</h2>
+  <p class="doc-note">Tool cards sit in <code>.tool-card-row</code> wrappers (no longer nested under <code>.msg-row</code>). The details panel now animates open/closed via <code>max-height</code> + opacity — click any header below to see the transition.</p>
+
+  <div class="doc-card"><span class="doc-label">.tool-card.tool-card-running (collapsed, pulsing dot)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-card-row">
+        <div class="tool-card tool-card-running">
+          <div class="tool-card-header">
+            <span class="tool-card-running-dot"></span>
+            <span class="tool-card-icon">⚡</span>
+            <span class="tool-card-name">bash</span>
+            <span class="tool-card-preview">npm run build</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.tool-card — done, collapsed</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-card-row">
+        <div class="tool-card">
+          <div class="tool-card-header">
+            <span class="tool-card-icon">📄</span>
+            <span class="tool-card-name">read_file</span>
+            <span class="tool-card-preview">static/style.css · 1155 lines</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.tool-card.open — args table + result snippet + Show more (animated detail)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-card-row">
+        <div class="tool-card open">
+          <div class="tool-card-header">
+            <span class="tool-card-icon">⚡</span>
+            <span class="tool-card-name">bash</span>
+            <span class="tool-card-preview">grep -rn "msg-role" static/ · exit 0 · 380ms</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+          <div class="tool-card-detail">
+            <div class="tool-card-args">
+              <div><span class="tool-arg-key">command:</span> <span class="tool-arg-val">grep -rn "msg-role" static/</span></div>
+              <div><span class="tool-arg-key">cwd:</span> <span class="tool-arg-val">/Users/aron/hermes-webui</span></div>
+              <div><span class="tool-arg-key">timeout:</span> <span class="tool-arg-val">30000</span></div>
+            </div>
+            <div class="tool-card-result">
+<pre>static/style.css:430:  .msg-role{font-size:12px;font-weight:500...}
+static/style.css:431:  .msg-role.user{color:rgba(124,185,255,0.65);}
+static/style.css:432:  .msg-role.assistant{color:rgba(201,168,76,0.6);}
+static/ui.js:1141:    const roleEl = el('div', 'msg-role ' + role);</pre>
+              <button class="tool-card-more">Show more (+142 lines)</button>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.tool-card.tool-card-subagent — delegated work</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-card-row">
+        <div class="tool-card tool-card-subagent">
+          <div class="tool-card-header">
+            <span class="tool-card-icon">🤖</span>
+            <span class="tool-card-name">Subagent</span>
+            <span class="tool-card-preview">Explore · Map chat messages UI elements</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+        </div>
+      </div>
+      <div class="tool-card-row">
+        <div class="tool-card tool-card-subagent">
+          <div class="tool-card-header">
+            <span class="tool-card-icon">🤖</span>
+            <span class="tool-card-name">Delegate task</span>
+            <span class="tool-card-preview">Plan · Propose redesign variants</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.tool-card (error snippet)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-card-row">
+        <div class="tool-card open">
+          <div class="tool-card-header">
+            <span class="tool-card-icon">⚡</span>
+            <span class="tool-card-name">bash</span>
+            <span class="tool-card-preview">npm run typecheck · exit 1 · 2.3s</span>
+            <span class="tool-card-toggle">▶</span>
+          </div>
+          <div class="tool-card-detail">
+            <div class="tool-card-args">
+              <div><span class="tool-arg-key">command:</span> <span class="tool-arg-val">npm run typecheck</span></div>
+            </div>
+            <div class="tool-card-result">
+<pre style="color:#fca5a5;">src/server.ts:42:7 - error TS2345: Argument of type 'string | undefined'
+  is not assignable to parameter of type 'number'.
+
+42       app.listen(opts.port, () =&gt; {
+         ~~~~~~~~~</pre>
+            </div>
+          </div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.tool-cards-toggle — Expand/Collapse All (≥2 cards)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="tool-cards-toggle">
+        <button>Expand all (3)</button>
+        <button>Collapse all</button>
+      </div>
+      <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">📄</span><span class="tool-card-name">read_file</span><span class="tool-card-preview">package.json</span><span class="tool-card-toggle">▶</span></div></div></div>
+      <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">🔎</span><span class="tool-card-name">grep</span><span class="tool-card-preview">"listen" in src/</span><span class="tool-card-toggle">▶</span></div></div></div>
+      <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">⚡</span><span class="tool-card-name">bash</span><span class="tool-card-preview">npm run typecheck · exit 0 · 4.1s</span><span class="tool-card-toggle">▶</span></div></div></div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">9 · Meta affordances</div>
+  <h2 class="doc-h">Role timestamp tooltip, footer action toolbar, token-usage badge</h2>
+  <p class="doc-note">Assistant timestamps live on the <code>.msg-role</code> <code>title</code> attribute (hover for full date). Copy/regen buttons sit in the per-segment <code>.msg-foot</code>, 45% opacity at rest, full on turn hover. The <code>.msg-usage</code> badge is always visible at the bottom of the turn.</p>
+
+  <div class="doc-card"><span class="doc-label">Full hover state — .msg-foot actions + .msg-usage</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn force-show" data-role="assistant">
+        <div class="msg-role assistant" title="Thu, Apr 16 2026, 10:42 AM">
+          <span class="role-icon assistant">H</span>
+          <span>Hermes</span>
+        </div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+          <div class="msg-body"><p>Built and type-checked successfully — server is running on <code>:3000</code>.</p></div>
+          <div class="msg-foot">
+            <span class="msg-actions">
+              <button class="msg-copy-btn msg-action-btn" title="Copy"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><rect x="9" y="9" width="13" height="13" rx="2" ry="2"/><path d="M5 15H4a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2h9a2 2 0 0 1 2 2v1"/></svg></button>
+              <button class="msg-action-btn" title="Regenerate"><svg width="13" height="13" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><polyline points="1 4 1 10 7 10"/><path d="M3.51 15a9 9 0 1 0 2.13-9.36L1 10"/></svg></button>
+            </span>
+          </div>
+        </div></div>
+        <div class="msg-usage">3.2K in · 481 out · ~$0.012</div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">10 · Full composition</div>
+  <h2 class="doc-h">User turn → assistant turn (segment 1: thinking + body + tool cards) → usage</h2>
+  <p class="doc-note">A realistic turn: one role header up top, then the segment hosting a thinking card plus the first body; tool cards follow as siblings of the turn inside <code>.messages-inner</code>; the usage badge closes the turn.</p>
+
+  <div class="doc-card"><span class="doc-label">All-in-one turn</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row force-show" data-role="user">
+        <div class="msg-files"><span class="msg-file-badge">📎 server.ts</span></div>
+        <div class="msg-body"><p>The build fails — can you type-check and explain?</p></div>
+        <div class="msg-foot">
+          <span class="msg-time">10:40</span>
+          <span class="msg-actions">
+            <button class="msg-action-btn" title="Edit">✎</button>
+            <button class="msg-copy-btn msg-action-btn" title="Copy">⎘</button>
+          </span>
+        </div>
+      </div>
+
+      <div class="msg-row assistant-turn force-show" data-role="assistant">
+        <div class="msg-role assistant" title="Thu, Apr 16 2026, 10:42 AM">
+          <span class="role-icon assistant">H</span><span>Hermes</span>
+        </div>
+        <div class="assistant-turn-blocks">
+          <div class="assistant-segment">
+            <div class="thinking-card open">
+              <div class="thinking-card-header"><span class="thinking-card-icon">💡</span><span class="thinking-card-label">Thought for 2.1s</span><span class="thinking-card-toggle">▶</span></div>
+              <div class="thinking-card-body"><pre>Attached server.ts — probably typing issue.
+Run typecheck to confirm, then patch.</pre></div>
+            </div>
+            <div class="msg-body">
+              <p>The build fails because <code>opts.port</code> can be <code>undefined</code>. Two fixes below — pick the one that matches your intent.</p>
+              <h3>Option A — require the port</h3>
+<pre><code class="language-typescript">export function startServer(opts: { port: number }) {
+  app.listen(opts.port);
+}</code></pre>
+              <h3>Option B — default to 3000</h3>
+<pre><code class="language-typescript">export function startServer(opts: { port?: number } = {}) {
+  const port = opts.port ?? 3000;
+  app.listen(port);
+}</code></pre>
+              <p>I ran the checks below to confirm.</p>
+            </div>
+            <div class="msg-foot">
+              <span class="msg-actions">
+                <button class="msg-copy-btn msg-action-btn" title="Copy">⎘</button>
+                <button class="msg-action-btn" title="Regenerate">↻</button>
+              </span>
+            </div>
+          </div>
+        </div>
+        <div class="msg-usage">11.4K in · 612 out · ~$0.049</div>
+      </div>
+
+      <div class="tool-cards-toggle">
+        <button>Expand all (3)</button><button>Collapse all</button>
+      </div>
+      <div class="tool-card-row"><div class="tool-card open">
+        <div class="tool-card-header"><span class="tool-card-icon">📄</span><span class="tool-card-name">read_file</span><span class="tool-card-preview">src/server.ts · 58 lines</span><span class="tool-card-toggle">▶</span></div>
+        <div class="tool-card-detail">
+          <div class="tool-card-args"><div><span class="tool-arg-key">path:</span> <span class="tool-arg-val">src/server.ts</span></div></div>
+          <div class="tool-card-result"><pre>export function startServer(opts: ServerOptions) {
+  app.listen(opts.port, () =&gt; { ... });
+}</pre></div>
+        </div>
+      </div></div>
+      <div class="tool-card-row"><div class="tool-card">
+        <div class="tool-card-header"><span class="tool-card-icon">⚡</span><span class="tool-card-name">bash</span><span class="tool-card-preview">npm run typecheck · exit 1 · 2.3s</span><span class="tool-card-toggle">▶</span></div>
+      </div></div>
+      <div class="tool-card-row"><div class="tool-card">
+        <div class="tool-card-header"><span class="tool-card-icon">✏️</span><span class="tool-card-name">edit_file</span><span class="tool-card-preview">src/server.ts +1 / -1</span><span class="tool-card-toggle">▶</span></div>
+      </div></div>
+    </div></div>
+  </div>
+</section>
+
+
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">12 · System / inline notes</div>
+  <h2 class="doc-h">Compression, cancellation, errors — rendered as italicised assistant messages</h2>
+
+  <div class="doc-card"><span class="doc-label">Italic system notices (still italic — info, not errors)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks">
+          <div class="assistant-segment"><div class="msg-body"><p><em>[Context was auto-compressed to continue the conversation]</em></p></div></div>
+          <div class="assistant-segment"><div class="msg-body"><p><em>Task cancelled.</em></p></div></div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.assistant-segment[data-error="1"] — real error card, red accent, no italic</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks">
+          <div class="assistant-segment" data-error="1"><div class="msg-body"><p><strong>Error:</strong> Connection lost. Your last message was saved — refresh to continue.</p></div></div>
+          <div class="assistant-segment" data-error="1"><div class="msg-body"><p><strong>Error:</strong> Upstream rate-limited (429). Retrying in 30s…</p></div></div>
+        </div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">12b · Turn boundaries &amp; date separators</div>
+  <h2 class="doc-h">Right-alignment separates user turns · day-change separator</h2>
+  <p class="doc-note">The dashed divider before each user turn was removed — the right-edge bubble alignment is its own visual break, so only a small vertical gap (10px top margin) remains between turns. Day changes still get a centred <code>.msg-date-sep</code>.</p>
+  <div class="doc-card"><span class="doc-label">.msg-date-sep — Today / Yesterday / weekday / date</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-date-sep">Yesterday</div>
+      <div class="msg-row" data-role="user"><div class="msg-body"><p>Can you summarise the PR I opened earlier?</p></div></div>
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment"><div class="msg-body"><p>Yes — three files changed, net +42 / -18. Main change is the new rail variable…</p></div></div></div>
+      </div>
+      <div class="msg-date-sep">Today</div>
+      <div class="msg-row" data-role="user"><div class="msg-body"><p>Did CI pass overnight?</p></div></div>
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment"><div class="msg-body"><p>All green — three jobs, 4m 12s total. Here's the breakdown:</p></div></div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">13 · Overlay cards (adjacent to transcript)</div>
+  <h2 class="doc-h">Approval &amp; Clarify cards + reconnect banner</h2>
+
+  <div class="doc-card"><span class="doc-label">.approval-card — 4 button variants (once / session / always / deny)</span>
+    <div class="approval-card doc-visible">
+      <div class="approval-inner">
+        <div class="approval-header">⚠ Approval required</div>
+        <div class="approval-desc" style="font-size:12px;color:var(--muted);margin-bottom:8px;">The agent wants to run a shell command in <code>/Users/aron/hermes-webui</code>.</div>
+        <div class="approval-cmd">rm -rf node_modules &amp;&amp; npm install</div>
+        <div class="approval-btns">
+          <button class="approval-btn once">✓ <span class="approval-btn-label">Allow once</span><kbd class="approval-kbd">↵</kbd></button>
+          <button class="approval-btn session">🔒 <span class="approval-btn-label">Allow session</span></button>
+          <button class="approval-btn always">★ <span class="approval-btn-label">Always allow</span></button>
+          <button class="approval-btn deny">✕ <span class="approval-btn-label">Deny</span></button>
+        </div>
+      </div>
+    </div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">.clarify-card — choice buttons + free-text fallback</span>
+    <div class="clarify-card doc-visible">
+      <div class="clarify-inner">
+        <div class="clarify-header">? Clarification needed</div>
+        <div class="clarify-question">Which environment should I deploy this to?</div>
+        <div class="clarify-choices">
+          <button class="clarify-choice"><span class="clarify-choice-badge">A</span><span class="clarify-choice-text">Staging — safe sandbox, auto-teardown nightly</span></button>
+          <button class="clarify-choice"><span class="clarify-choice-badge">B</span><span class="clarify-choice-text">Production EU — customer-facing, requires change ticket</span></button>
+          <button class="clarify-choice"><span class="clarify-choice-badge">C</span><span class="clarify-choice-text">Production US — same caveats as EU</span></button>
+          <button class="clarify-choice other"><span class="clarify-choice-badge other">✎</span><span class="clarify-choice-text">Other — I'll type it below</span></button>
+        </div>
+        <div class="clarify-response">
+          <input class="clarify-input" type="text" placeholder="Type your response…">
+          <button class="clarify-submit">Send</button>
+        </div>
+        <div class="clarify-hint">Pick a choice, or type your own answer below.</div>
+      </div>
+    </div>
+  </div>
+
+  <div class="doc-card"><span class="doc-label">Reconnect / mid-stream recovery banner</span>
+    <div class="reconnect-banner doc-visible">
+      <span>⚠ A response may have been in progress when you last left. Reload messages?</span>
+      <div style="display:flex;gap:8px;">
+        <button class="reconnect-btn">Dismiss</button>
+        <button class="reconnect-btn">↻ Reload</button>
+      </div>
+    </div>
+    <div class="bg-error-banner doc-visible" style="margin-top:8px;">
+      <span>⚠ Agent run exited with non-zero status (code 1). Check the logs.</span>
+      <button class="reconnect-btn">Dismiss</button>
+    </div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">14 · Structure &amp; data-attribute cheat sheet</div>
+  <h2 class="doc-h">Wrappers and state markers produced by <code>renderMessages()</code></h2>
+
+  <div class="doc-card" style="padding:14px 18px;">
+    <h3 style="font-size:13px;color:var(--text);margin:0 0 8px;">Wrappers</h3>
+    <ul style="color:var(--muted);font-size:12px;line-height:1.9;list-style:disc;padding-left:20px;">
+      <li><code>.msg-row[data-role="user"]</code> — one user turn (right-aligned bubble, 60% max-width)</li>
+      <li><code>.msg-row.assistant-turn[data-role="assistant"]</code> — one assistant turn; contains <strong>one</strong> <code>.msg-role</code> and <strong>one</strong> <code>.assistant-turn-blocks</code></li>
+      <li><code>.assistant-turn-blocks</code> — flex-column holder for segments</li>
+      <li><code>.assistant-segment</code> — a single logical chunk inside a turn: optional <code>.thinking-card</code> + optional <code>.msg-body</code> + optional <code>.msg-foot</code></li>
+      <li><code>.assistant-segment-anchor</code> — hidden segment kept as a DOM anchor for tool cards when the model emitted no text</li>
+      <li><code>.tool-card-row</code> — per-tool-card wrapper, sibling of the turn inside <code>.messages-inner</code></li>
+      <li><code>.msg-foot</code> — per-segment (or per-user-row) footer holding <code>.msg-time</code> + <code>.msg-actions</code></li>
+    </ul>
+    <h3 style="font-size:13px;color:var(--text);margin:14px 0 8px;">Data attributes &amp; IDs</h3>
+    <ul style="color:var(--muted);font-size:12px;line-height:1.9;list-style:disc;padding-left:20px;">
+      <li><code>data-role="user|assistant"</code> — role marker on the row</li>
+      <li><code>data-msgIdx="N"</code> — index into <code>S.messages</code>; on user rows <em>and</em> assistant segments</li>
+      <li><code>data-raw-text="…"</code> — plain-text source for copy (now lives on <code>.assistant-segment</code> for assistant output)</li>
+      <li><code>data-live-assistant="1"</code> — the segment that's currently streaming</li>
+      <li><code>data-editing="1"</code> — row is in edit mode</li>
+      <li><code>data-error="1"</code> — error state; applies to <code>.msg-row</code> (user) or <code>.assistant-segment</code></li>
+      <li><code>id="liveAssistantTurn"</code> — on the turn that contains the streaming segment</li>
+      <li><code>.tool-card-row[data-live-tid="…"]</code> — live tool-call card (removed when the turn settles)</li>
+      <li><code>data-mermaid-id</code>, <code>data-katex</code>, <code>data-rendered</code> — block rendering state</li>
+    </ul>
+  </div>
+</section>
+
+</main>
+
+<!-- ============================================================= -->
+<!-- Prism autoloader for real syntax highlighting -->
+<script src="https://cdnjs.cloudflare.com/ajax/libs/prism/1.29.0/components/prism-core.min.js"></script>
+<script src="https://cdnjs.cloudflare.com/ajax/libs/prism/1.29.0/plugins/autoloader/prism-autoloader.min.js"></script>
+<!-- KaTeX auto-render -->
+<script defer src="https://cdn.jsdelivr.net/npm/katex@0.16.9/dist/katex.min.js"></script>
+<script defer src="https://cdn.jsdelivr.net/npm/katex@0.16.9/dist/contrib/auto-render.min.js"
+  onload="renderMathInElement(document.body,{delimiters:[{left:'$$',right:'$$',display:true},{left:'\\[',right:'\\]',display:true},{left:'\\(',right:'\\)',display:false},{left:'$',right:'$',display:false}],throwOnError:false});"></script>
+
+<script>
+  // Theme picker
+  document.querySelectorAll('[data-theme-btn]').forEach(btn => {
+    btn.addEventListener('click', () => {
+      const t = btn.dataset.themeBtn;
+      if (t === 'default') document.documentElement.removeAttribute('data-theme');
+      else document.documentElement.setAttribute('data-theme', t);
+      document.querySelectorAll('[data-theme-btn]').forEach(b => b.classList.toggle('on', b === btn));
+    });
+  });
+  // Bubble-layout toggle
+
+  // Thinking / tool-card click-to-toggle (so the demo feels live)
+  document.querySelectorAll('.thinking-card-header, .tool-card-header').forEach(h => {
+    h.addEventListener('click', () => h.parentElement.classList.toggle('open'));
+  });
+</script>
+
+</body>
+</html>

docs/ui-ux/two-stage-proposal.html

@@ -0,0 +1,742 @@
+<!doctype html>
+<html lang="en" data-theme="slate">
+<head>
+  <meta charset="utf-8">
+  <title>Hermes WebUI — Two-Stage Chat Proposal (Issue #536)</title>
+  <meta name="viewport" content="width=device-width,initial-scale=1">
+  <link rel="stylesheet" href="../../static/style.css">
+  <style>
+    /* ──────────────────────────────────────────────────────────────
+       Doc-chrome scaffold (same pattern as index.html) — real app CSS
+       is used unchanged inside .messages / .msg-row. New proposed
+       elements are prefixed .p2s- so nothing collides with the app.
+       ────────────────────────────────────────────────────────────── */
+    body{display:block !important;height:auto !important;min-height:100vh;overflow:auto !important;}
+    .doc-header{position:sticky;top:0;z-index:50;background:var(--topbar-bg);backdrop-filter:blur(12px);border-bottom:1px solid var(--border);padding:14px 24px;display:flex;flex-wrap:wrap;align-items:center;gap:14px;}
+    .doc-title{font-size:16px;font-weight:700;letter-spacing:-.01em;color:var(--text);}
+    .doc-title small{display:block;font-size:11px;font-weight:500;color:var(--muted);margin-top:3px;}
+    .doc-title a{color:var(--blue);text-decoration:none;}
+    .doc-toggles{display:flex;flex-wrap:wrap;gap:6px;margin-left:auto;}
+    .doc-toggles button{font:inherit;font-size:11px;padding:5px 10px;border-radius:7px;border:1px solid var(--border2);background:var(--input-bg);color:var(--muted);cursor:pointer;}
+    .doc-toggles button.on{background:rgba(124,185,255,.12);border-color:rgba(124,185,255,.4);color:var(--blue);}
+    .doc-main{max-width:1180px;margin:0 auto;padding:24px 24px 120px;}
+    .doc-section{margin:48px 0 8px;padding-top:22px;border-top:1px dashed var(--border);}
+    .doc-section:first-of-type{border-top:none;padding-top:0;margin-top:0;}
+    .doc-kicker{font-size:10px;font-weight:700;letter-spacing:.14em;text-transform:uppercase;color:var(--blue);}
+    .doc-h{font-size:20px;font-weight:700;color:var(--text);margin:4px 0 6px;letter-spacing:-.01em;}
+    .doc-note{font-size:12.5px;color:var(--muted);line-height:1.6;max-width:780px;margin-bottom:14px;}
+    .doc-note code{color:var(--text);background:rgba(255,255,255,.05);padding:1px 5px;border-radius:4px;font-size:11.5px;}
+    .doc-card{position:relative;background:var(--main-bg);border:1px solid var(--border);border-radius:14px;padding:6px 8px;margin:14px 0;}
+    .doc-label{position:absolute;top:-9px;left:14px;font-size:10px;font-weight:700;text-transform:uppercase;letter-spacing:.08em;padding:2px 9px;background:var(--bg);color:var(--muted);border:1px solid var(--border);border-radius:999px;}
+    .doc-label.current{color:var(--muted);}
+    .doc-label.proposed{color:var(--gold);border-color:rgba(201,168,76,.35);background:var(--bg);}
+    .force-show .msg-actions,.force-show .msg-time,.force-show .msg-foot{opacity:1 !important;}
+    .messages.doc-messages{overflow:visible;display:block;}
+    .messages-inner.doc-inner{padding:14px 16px;}
+    .approval-card.doc-visible,.clarify-card.doc-visible{display:block;}
+    .doc-grid-2{display:grid;grid-template-columns:repeat(auto-fit,minmax(440px,1fr));gap:14px;}
+
+    /* ──────────────────────────────────────────────────────────────
+       Proposed two-stage elements (prefix .p2s-)
+
+       The proposal introduces one container (.p2s-stage1) that wraps
+       the execution history (thinking + tool cards) and one visual
+       treatment (.p2s-answer) for the final-answer segment. The same
+       DOM can be rendered in three modes:
+
+         .p2s-stage1.is-live      →  Working timer + expanded history
+         .p2s-stage1.is-settled   →  Collapsed to one-line summary
+         .p2s-stage1.is-settled.is-open → expanded on demand
+
+       Everything else (thinking-card, tool-card-row, msg-body) is the
+       existing app CSS unchanged.
+       ────────────────────────────────────────────────────────────── */
+
+    /* Worklog bar — the header of Stage 1.
+       Aligns with every other rail child via --msg-rail / --msg-max. */
+    .p2s-worklog{
+      display:flex;align-items:center;gap:10px;
+      margin:4px 0 6px var(--msg-rail);
+      max-width:var(--msg-max);
+      padding:8px 12px;
+      border:1px solid var(--border);
+      border-radius:10px;
+      background:rgba(255,255,255,.025);
+      font-size:12px;color:var(--muted);
+      cursor:pointer;user-select:none;
+      transition:border-color .15s,background .15s;
+    }
+    .p2s-worklog:hover{border-color:var(--border2);background:rgba(255,255,255,.04);}
+    .p2s-worklog-dot{
+      width:8px;height:8px;border-radius:50%;background:var(--gold);flex-shrink:0;
+      box-shadow:0 0 0 0 rgba(201,168,76,.4);
+    }
+    .p2s-stage1.is-live .p2s-worklog-dot{
+      animation:p2sPulse 1.4s ease-in-out infinite;
+    }
+    .p2s-stage1.is-settled .p2s-worklog-dot{
+      background:var(--muted);opacity:.6;
+    }
+    @keyframes p2sPulse{
+      0%,100%{box-shadow:0 0 0 0 rgba(201,168,76,.45);}
+      50%{box-shadow:0 0 0 6px rgba(201,168,76,0);}
+    }
+    .p2s-worklog-label{color:var(--text);font-weight:500;}
+    .p2s-worklog-stats{margin-left:auto;display:flex;gap:12px;color:var(--muted);font-size:11.5px;}
+    .p2s-worklog-stats b{color:var(--text);font-weight:600;}
+    .p2s-worklog-caret{
+      display:inline-block;width:14px;height:14px;line-height:14px;text-align:center;
+      color:var(--muted);font-size:10px;transition:transform .2s;
+      margin-left:6px;
+    }
+    .p2s-stage1.is-live .p2s-worklog-caret{display:none;}
+    .p2s-stage1.is-settled.is-open .p2s-worklog-caret{transform:rotate(90deg);}
+
+    /* Stage 1 body — holds thinking + tool cards + round separators. */
+    .p2s-stage1-body{
+      overflow:hidden;
+      transition:max-height .35s ease,opacity .25s ease;
+    }
+    .p2s-stage1.is-live .p2s-stage1-body,
+    .p2s-stage1.is-settled.is-open .p2s-stage1-body{
+      max-height:2000px;opacity:1;
+    }
+    .p2s-stage1.is-settled:not(.is-open) .p2s-stage1-body{
+      max-height:0;opacity:0;pointer-events:none;
+    }
+
+    /* Round separator — shown inside Stage 1 between execution rounds. */
+    .p2s-round-sep{
+      display:flex;align-items:center;gap:10px;
+      margin:10px 0 6px var(--msg-rail);
+      max-width:var(--msg-max);
+      color:var(--muted);
+      font-size:10.5px;font-weight:700;letter-spacing:.1em;text-transform:uppercase;
+    }
+    .p2s-round-sep::before,.p2s-round-sep::after{
+      content:"";flex:1;height:1px;background:var(--border);
+    }
+
+    /* Stage 1 → Stage 2 transition divider. */
+    .p2s-transition{
+      margin:14px 0 10px var(--msg-rail);
+      max-width:var(--msg-max);
+      height:1px;
+      background:linear-gradient(
+        to right,transparent,var(--border) 20%,var(--border) 80%,transparent
+      );
+    }
+
+    /* Stage 2 — the final answer wrapper.
+
+       Design intent: nothing loud. A small "Answer" kicker in gold,
+       slightly taller line-height, the existing .msg-body styling,
+       and a gentle top breathing-space. The user arrives at this
+       block and it *feels* like a conclusion, not another tool row.
+    */
+    .p2s-answer{margin-top:8px;}
+    .p2s-answer-kicker{
+      margin:0 0 4px var(--msg-rail);
+      max-width:var(--msg-max);
+      font-size:10px;font-weight:700;letter-spacing:.14em;text-transform:uppercase;
+      color:var(--gold);opacity:.8;
+    }
+    .p2s-answer .msg-body{
+      font-size:14.5px;line-height:1.78;
+    }
+
+    /* Clarify slot — placed at the transition rather than inline. */
+    .p2s-clarify-slot{
+      margin:12px 0 4px var(--msg-rail);
+      max-width:var(--msg-max);
+    }
+    .p2s-clarify-slot .clarify-card{margin:0;}
+
+    /* Comparison-grid accents. */
+    .doc-compare-caption{
+      font-size:11px;color:var(--muted);text-align:center;padding:6px 0;
+    }
+  </style>
+</head>
+<body>
+
+<header class="doc-header">
+  <div class="doc-title">
+    Two-Stage Chat UX — Proposal for <a href="https://github.com/nesquena/hermes-webui/issues/536" target="_blank">issue #536</a>
+    <small>Companion to <a href="./index.html">index.html</a> — shows <em>Working → Final answer</em> as a distinct two-phase interaction model.</small>
+  </div>
+  <div class="doc-toggles">
+    <strong style="font-size:10px;color:var(--muted);letter-spacing:.08em;text-transform:uppercase;align-self:center;margin-right:4px;">Theme</strong>
+    <button data-theme-btn="default">Default</button>
+    <button data-theme-btn="slate" class="on">Slate</button>
+    <button data-theme-btn="light">Light</button>
+    <button data-theme-btn="solarized">Solarized</button>
+    <button data-theme-btn="monokai">Monokai</button>
+    <button data-theme-btn="nord">Nord</button>
+    <button data-theme-btn="oled">OLED</button>
+  </div>
+</header>
+
+<main class="doc-main">
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">0 · The model</div>
+  <h2 class="doc-h">One turn, two stages</h2>
+  <p class="doc-note">
+    Today an assistant turn is a flat stream: thinking card → tool cards → answer, all stacked
+    inline with equal visual weight. The proposal wraps the execution history in a
+    <code>.p2s-stage1</code> container with a <em>worklog bar</em> as its header, and marks the
+    final answer as <code>.p2s-answer</code>. The same DOM renders three ways:
+  </p>
+  <ul class="doc-note" style="padding-left:18px;list-style:disc;">
+    <li><b>Live</b> — worklog shows <em>Working… 0:42 · 2 tools</em> with a pulsing dot; history is fully visible.</li>
+    <li><b>Settled</b> — worklog collapses to a single line (<em>Worked 1:42 · 4 tools · 2 thinking</em>); final answer sits below as the calm conclusion.</li>
+    <li><b>Settled + opened</b> — user clicks the worklog to re-expand the history for audit.</li>
+  </ul>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">1 · Current vs proposed — settled turn</div>
+  <h2 class="doc-h">Side-by-side comparison</h2>
+  <p class="doc-note">
+    Same turn, same tool calls, same answer. Left is what #587 ships today. Right is the
+    proposal: execution history collapses to a one-line summary; the final answer stands alone
+    with a small <em>Answer</em> kicker.
+  </p>
+
+  <div class="doc-grid-2">
+
+    <!-- CURRENT ──────────────────────────────────────────────── -->
+    <div class="doc-card"><span class="doc-label current">Current (PR #587)</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner">
+        <div class="msg-row" data-role="user">
+          <div class="msg-body"><p>Does our dev server pick up the workspace from an env var or a flag?</p></div>
+        </div>
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+            <div class="thinking-card open">
+              <div class="thinking-card-header">
+                <span class="thinking-card-icon">💡</span>
+                <span class="thinking-card-label">Thought for 3.1s</span>
+                <span class="thinking-card-toggle">▶</span>
+              </div>
+              <div class="thinking-card-body"><pre>Check how the CLI resolves workspace:
+grep for HERMES_WORKSPACE and --workspace
+inspect argv vs env precedence.</pre></div>
+            </div>
+          </div></div>
+        </div>
+        <div class="tool-card-row">
+          <div class="tool-card open">
+            <div class="tool-card-header">
+              <span class="tool-card-icon">⚡</span>
+              <span class="tool-card-name">bash</span>
+              <span class="tool-card-preview">grep -rn "HERMES_WORKSPACE" . · exit 0</span>
+              <span class="tool-card-toggle">▶</span>
+            </div>
+            <div class="tool-card-detail">
+              <div class="tool-card-result"><pre>cli/main.py:14:WORKSPACE_ENV = "HERMES_WORKSPACE"
+cli/main.py:92:    ws = os.getenv(WORKSPACE_ENV) or args.workspace</pre></div>
+            </div>
+          </div>
+        </div>
+        <div class="tool-card-row">
+          <div class="tool-card">
+            <div class="tool-card-header">
+              <span class="tool-card-icon">📄</span>
+              <span class="tool-card-name">read_file</span>
+              <span class="tool-card-preview">cli/main.py · 148 lines</span>
+              <span class="tool-card-toggle">▶</span>
+            </div>
+          </div>
+        </div>
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+            <div class="msg-body">
+              <p>Both work, but <strong>env wins</strong>. The CLI reads
+              <code>HERMES_WORKSPACE</code> first and only falls back to the
+              <code>--workspace</code> flag if the env var is unset.</p>
+              <p>So in practice:</p>
+              <ul>
+                <li>CI / daemons → set the env var.</li>
+                <li>Ad-hoc runs → pass <code>--workspace</code>.</li>
+              </ul>
+            </div>
+          </div></div>
+        </div>
+      </div></div>
+      <div class="doc-compare-caption">Everything stacks equally — the answer is just the next block.</div>
+    </div>
+
+    <!-- PROPOSED ─────────────────────────────────────────────── -->
+    <div class="doc-card"><span class="doc-label proposed">Proposed — two-stage, settled</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner">
+        <div class="msg-row" data-role="user">
+          <div class="msg-body"><p>Does our dev server pick up the workspace from an env var or a flag?</p></div>
+        </div>
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+
+            <!-- Stage 1 — settled, collapsed to summary (click to expand) -->
+            <div class="p2s-stage1 is-settled" data-p2s-toggle>
+              <div class="p2s-worklog">
+                <span class="p2s-worklog-dot"></span>
+                <span class="p2s-worklog-label">Worked for 0:08</span>
+                <span class="p2s-worklog-stats">
+                  <span><b>2</b> tools</span>
+                  <span><b>1</b> thinking round</span>
+                </span>
+                <span class="p2s-worklog-caret">▶</span>
+              </div>
+              <div class="p2s-stage1-body">
+                <div class="thinking-card open">
+                  <div class="thinking-card-header">
+                    <span class="thinking-card-icon">💡</span>
+                    <span class="thinking-card-label">Thought for 3.1s</span>
+                    <span class="thinking-card-toggle">▶</span>
+                  </div>
+                  <div class="thinking-card-body"><pre>Check how the CLI resolves workspace:
+grep for HERMES_WORKSPACE and --workspace
+inspect argv vs env precedence.</pre></div>
+                </div>
+                <div class="tool-card-row">
+                  <div class="tool-card">
+                    <div class="tool-card-header">
+                      <span class="tool-card-icon">⚡</span>
+                      <span class="tool-card-name">bash</span>
+                      <span class="tool-card-preview">grep -rn "HERMES_WORKSPACE" . · exit 0</span>
+                      <span class="tool-card-toggle">▶</span>
+                    </div>
+                  </div>
+                </div>
+                <div class="tool-card-row">
+                  <div class="tool-card">
+                    <div class="tool-card-header">
+                      <span class="tool-card-icon">📄</span>
+                      <span class="tool-card-name">read_file</span>
+                      <span class="tool-card-preview">cli/main.py · 148 lines</span>
+                      <span class="tool-card-toggle">▶</span>
+                    </div>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+            <!-- Stage 2 — the final answer -->
+            <div class="p2s-transition"></div>
+            <div class="p2s-answer">
+              <div class="p2s-answer-kicker">Answer</div>
+              <div class="msg-body">
+                <p>Both work, but <strong>env wins</strong>. The CLI reads
+                <code>HERMES_WORKSPACE</code> first and only falls back to the
+                <code>--workspace</code> flag if the env var is unset.</p>
+                <p>So in practice:</p>
+                <ul>
+                  <li>CI / daemons → set the env var.</li>
+                  <li>Ad-hoc runs → pass <code>--workspace</code>.</li>
+                </ul>
+              </div>
+            </div>
+
+          </div></div>
+        </div>
+      </div></div>
+      <div class="doc-compare-caption">Click the worklog bar to expand the execution history.</div>
+    </div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">2 · Stage 1 · Live run</div>
+  <h2 class="doc-h">Working timer + live execution history</h2>
+  <p class="doc-note">
+    The worklog bar at the top is the anchor for the whole active run: pulsing dot, elapsed
+    timer that ticks every second, and live counts that increment as tool cards resolve.
+    Thinking cards and tool cards render inside <code>.p2s-stage1-body</code> exactly as today.
+    A <em>Round N</em> separator is inserted when the agent starts a new reasoning/tool cycle.
+  </p>
+
+  <div class="doc-card"><span class="doc-label proposed">.p2s-stage1.is-live — Round 1 done, Round 2 running</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment" data-live-assistant="1">
+
+          <div class="p2s-stage1 is-live">
+            <div class="p2s-worklog">
+              <span class="p2s-worklog-dot"></span>
+              <span class="p2s-worklog-label">Working… <span id="p2sTimer">0:42</span></span>
+              <span class="p2s-worklog-stats">
+                <span><b>3</b> tools</span>
+                <span><b>2</b> thinking</span>
+              </span>
+            </div>
+            <div class="p2s-stage1-body">
+
+              <div class="thinking-card open">
+                <div class="thinking-card-header">
+                  <span class="thinking-card-icon">💡</span>
+                  <span class="thinking-card-label">Thought for 2.4s</span>
+                  <span class="thinking-card-toggle">▶</span>
+                </div>
+                <div class="thinking-card-body"><pre>Need to map the streaming code path first,
+then check the persistence layer.</pre></div>
+              </div>
+
+              <div class="tool-card-row">
+                <div class="tool-card">
+                  <div class="tool-card-header">
+                    <span class="tool-card-icon">📄</span>
+                    <span class="tool-card-name">read_file</span>
+                    <span class="tool-card-preview">api/streaming.py · 612 lines</span>
+                    <span class="tool-card-toggle">▶</span>
+                  </div>
+                </div>
+              </div>
+
+              <div class="tool-card-row">
+                <div class="tool-card">
+                  <div class="tool-card-header">
+                    <span class="tool-card-icon">⚡</span>
+                    <span class="tool-card-name">bash</span>
+                    <span class="tool-card-preview">grep -rn "tool_call_id" api/ · exit 0 · 88ms</span>
+                    <span class="tool-card-toggle">▶</span>
+                  </div>
+                </div>
+              </div>
+
+              <div class="p2s-round-sep">Round 2</div>
+
+              <div class="thinking-card">
+                <div class="thinking-card-header">
+                  <span class="thinking-card-icon">💡</span>
+                  <span class="thinking-card-label">Thought for 1.8s</span>
+                  <span class="thinking-card-toggle">▶</span>
+                </div>
+                <div class="thinking-card-body"><pre>Streaming looks fine — drill into how
+tool_calls get attached before save.</pre></div>
+              </div>
+
+              <div class="tool-card-row">
+                <div class="tool-card tool-card-running">
+                  <div class="tool-card-header">
+                    <span class="tool-card-running-dot"></span>
+                    <span class="tool-card-icon">⚡</span>
+                    <span class="tool-card-name">bash</span>
+                    <span class="tool-card-preview">pytest tests/test_tool_call_persistence.py -q</span>
+                    <span class="tool-card-toggle">▶</span>
+                  </div>
+                </div>
+              </div>
+
+            </div>
+          </div>
+
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">3 · Approve vs Clarify — placement</div>
+  <h2 class="doc-h">Approvals stay in Stage 1; Clarify moves to the transition</h2>
+  <p class="doc-note">
+    Per the issue: <em>approvals are part of doing the work</em> (they gate a single tool),
+    <em>clarifications stabilise the answer path</em> (they precede the conclusion). The
+    proposal keeps <code>.approval-card</code> inline among tool cards, and places
+    <code>.clarify-card</code> at the Stage 1 → Stage 2 seam, above the final answer.
+  </p>
+
+  <div class="doc-grid-2">
+
+    <!-- Approve inline in Stage 1 -->
+    <div class="doc-card"><span class="doc-label proposed">Approve card — inline in Stage 1</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner">
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment" data-live-assistant="1">
+
+            <div class="p2s-stage1 is-live">
+              <div class="p2s-worklog">
+                <span class="p2s-worklog-dot"></span>
+                <span class="p2s-worklog-label">Working… 0:18</span>
+                <span class="p2s-worklog-stats"><span><b>1</b> tool</span></span>
+              </div>
+              <div class="p2s-stage1-body">
+                <div class="tool-card-row">
+                  <div class="tool-card">
+                    <div class="tool-card-header">
+                      <span class="tool-card-icon">⚡</span>
+                      <span class="tool-card-name">bash</span>
+                      <span class="tool-card-preview">ls -la ~/.hermes/sessions · exit 0</span>
+                      <span class="tool-card-toggle">▶</span>
+                    </div>
+                  </div>
+                </div>
+                <div class="approval-card doc-visible">
+                  <div class="approval-card-header">
+                    <span class="approval-card-icon">🔐</span>
+                    <span class="approval-card-title">Approve command</span>
+                  </div>
+                  <div class="approval-card-body">
+                    <p class="approval-card-desc">Hermes wants to run a potentially destructive command:</p>
+                    <pre class="approval-card-cmd">rm -rf ~/.hermes/sessions/*.json.bak</pre>
+                  </div>
+                  <div class="approval-card-actions">
+                    <button class="approval-btn approve">Approve</button>
+                    <button class="approval-btn deny">Deny</button>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+          </div></div>
+        </div>
+      </div></div>
+      <div class="doc-compare-caption">Permission gate sits next to the tools it gates.</div>
+    </div>
+
+    <!-- Clarify at transition -->
+    <div class="doc-card"><span class="doc-label proposed">Clarify card — Stage 1 → Stage 2 transition</span>
+      <div class="messages doc-messages"><div class="messages-inner doc-inner">
+        <div class="msg-row assistant-turn" data-role="assistant">
+          <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+          <div class="assistant-turn-blocks"><div class="assistant-segment">
+
+            <div class="p2s-stage1 is-settled" data-p2s-toggle>
+              <div class="p2s-worklog">
+                <span class="p2s-worklog-dot"></span>
+                <span class="p2s-worklog-label">Worked for 0:12</span>
+                <span class="p2s-worklog-stats"><span><b>2</b> tools</span></span>
+                <span class="p2s-worklog-caret">▶</span>
+              </div>
+              <div class="p2s-stage1-body">
+                <div class="tool-card-row">
+                  <div class="tool-card">
+                    <div class="tool-card-header">
+                      <span class="tool-card-icon">📄</span>
+                      <span class="tool-card-name">read_file</span>
+                      <span class="tool-card-preview">package.json · 48 lines</span>
+                      <span class="tool-card-toggle">▶</span>
+                    </div>
+                  </div>
+                </div>
+                <div class="tool-card-row">
+                  <div class="tool-card">
+                    <div class="tool-card-header">
+                      <span class="tool-card-icon">⚡</span>
+                      <span class="tool-card-name">bash</span>
+                      <span class="tool-card-preview">ls src/ · exit 0</span>
+                      <span class="tool-card-toggle">▶</span>
+                    </div>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+            <div class="p2s-transition"></div>
+            <div class="p2s-clarify-slot">
+              <div class="clarify-card doc-visible">
+                <div class="clarify-card-header">
+                  <span class="clarify-card-icon">❓</span>
+                  <span class="clarify-card-title">One quick question before I answer</span>
+                </div>
+                <div class="clarify-card-body">
+                  <p>I can wire the dev server either as an <strong>npm script</strong> in the
+                  existing <code>package.json</code>, or as a standalone <strong>CLI
+                  entry-point</strong>. Which would you prefer?</p>
+                </div>
+                <div class="clarify-card-actions">
+                  <button class="clarify-opt">npm script</button>
+                  <button class="clarify-opt">CLI entry-point</button>
+                  <button class="clarify-opt">Let Hermes pick</button>
+                </div>
+              </div>
+            </div>
+
+          </div></div>
+        </div>
+      </div></div>
+      <div class="doc-compare-caption">Stage 1 is already settled; the answer is paused on clarification.</div>
+    </div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">4 · Stage 2 · Calm conclusion</div>
+  <h2 class="doc-h">What the "Answer" stage looks like on its own</h2>
+  <p class="doc-note">
+    Three small choices distinguish Stage 2 from a regular text block:
+    (1) a thin horizontal divider above it, (2) a tiny gold <em>Answer</em> kicker aligned to
+    the text rail, (3) a slightly taller line-height. No heavy borders, no boxed treatment —
+    the emphasis comes from <em>what is missing around it</em>, not ornament.
+  </p>
+
+  <div class="doc-card"><span class="doc-label proposed">.p2s-answer (Stage 1 collapsed above)</span>
+    <div class="messages doc-messages"><div class="messages-inner doc-inner">
+      <div class="msg-row assistant-turn" data-role="assistant">
+        <div class="msg-role assistant"><span class="role-icon assistant">H</span><span>Hermes</span></div>
+        <div class="assistant-turn-blocks"><div class="assistant-segment">
+
+          <div class="p2s-stage1 is-settled" data-p2s-toggle>
+            <div class="p2s-worklog">
+              <span class="p2s-worklog-dot"></span>
+              <span class="p2s-worklog-label">Worked for 1:42</span>
+              <span class="p2s-worklog-stats">
+                <span><b>4</b> tools</span>
+                <span><b>2</b> thinking</span>
+                <span><b>1</b> approval</span>
+              </span>
+              <span class="p2s-worklog-caret">▶</span>
+            </div>
+            <div class="p2s-stage1-body">
+              <div class="thinking-card"><div class="thinking-card-header"><span class="thinking-card-icon">💡</span><span class="thinking-card-label">Thought for 2.4s</span><span class="thinking-card-toggle">▶</span></div></div>
+              <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">📄</span><span class="tool-card-name">read_file</span><span class="tool-card-preview">api/streaming.py</span><span class="tool-card-toggle">▶</span></div></div></div>
+              <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">⚡</span><span class="tool-card-name">bash</span><span class="tool-card-preview">grep -rn "tool_call_id" api/</span><span class="tool-card-toggle">▶</span></div></div></div>
+              <div class="p2s-round-sep">Round 2</div>
+              <div class="thinking-card"><div class="thinking-card-header"><span class="thinking-card-icon">💡</span><span class="thinking-card-label">Thought for 1.8s</span><span class="thinking-card-toggle">▶</span></div></div>
+              <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">⚡</span><span class="tool-card-name">bash</span><span class="tool-card-preview">pytest -q · exit 0 · 2.4s</span><span class="tool-card-toggle">▶</span></div></div></div>
+              <div class="tool-card-row"><div class="tool-card"><div class="tool-card-header"><span class="tool-card-icon">✍️</span><span class="tool-card-name">edit_file</span><span class="tool-card-preview">api/streaming.py · +12 −3</span><span class="tool-card-toggle">▶</span></div></div></div>
+            </div>
+          </div>
+
+          <div class="p2s-transition"></div>
+          <div class="p2s-answer">
+            <div class="p2s-answer-kicker">Answer</div>
+            <div class="msg-body">
+              <p>Tool-call persistence was breaking because <code>session.tool_calls</code> was
+              written <em>after</em> <code>s.save()</code> in <code>api/streaming.py</code>.
+              I moved the attach step above the save, and added a fallback that reconstructs
+              ordering from live tool-progress events when <code>tool_call_id</code> is absent
+              on older sessions.</p>
+              <p>Net result:</p>
+              <ul>
+                <li>Reloading mid-stream now preserves every tool card with args + output snippet.</li>
+                <li>Last-turn reasoning survives reload.</li>
+                <li>No schema migration needed — old sessions degrade gracefully.</li>
+              </ul>
+              <p>Covered by the new regression in <code>tests/test_tool_call_persistence.py</code>.</p>
+            </div>
+            <div class="msg-foot" style="opacity:1;padding-left:var(--msg-rail);">
+              <span class="msg-time">11:42 AM · 2,481 tokens · 1.42s</span>
+              <span class="msg-actions">
+                <button class="msg-act" title="Copy">⧉</button>
+                <button class="msg-act" title="Regenerate">↻</button>
+              </span>
+            </div>
+          </div>
+
+        </div></div>
+      </div>
+    </div></div>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">5 · Open-question answers (picked defaults)</div>
+  <h2 class="doc-h">What this proposal commits to</h2>
+  <div class="doc-card" style="padding:16px 20px;">
+    <ul style="color:var(--muted);font-size:13px;line-height:1.85;list-style:disc;padding-left:22px;margin:0;">
+      <li><b style="color:var(--text);">Stage 1 on settle →</b> <em>partial</em> collapse to a
+        single worklog bar with counts. Click to re-expand. No "nuke to black box", no "keep
+        everything open forever".</li>
+      <li><b style="color:var(--text);">Final answer placement →</b> sits <em>beneath</em> Stage 1,
+        not replacing it. Visual distinction comes from the divider + kicker + spacing, not from
+        a two-panel layout.</li>
+      <li><b style="color:var(--text);">Clarify placement →</b> at the Stage 1 → Stage 2 seam.
+        Approvals stay inline with tools.</li>
+      <li><b style="color:var(--text);">Timer →</b> lives on Stage 1 only. Stops when the agent
+        emits the first Stage 2 token; final label becomes "Worked for N:NN".</li>
+      <li><b style="color:var(--text);">Signal for "answer has started" →</b> first assistant
+        text delta after all tool calls have resolved and no new <code>tool_use</code> is pending
+        in the current round. Already present in the SSE stream per maintainer comment.</li>
+    </ul>
+  </div>
+</section>
+
+<!-- ============================================================= -->
+<section class="doc-section">
+  <div class="doc-kicker">6 · DOM cheat-sheet</div>
+  <h2 class="doc-h">What changes vs index.html</h2>
+  <div class="doc-card" style="padding:14px 18px;">
+    <h3 style="font-size:13px;color:var(--text);margin:0 0 8px;">New wrappers</h3>
+    <ul style="color:var(--muted);font-size:12px;line-height:1.9;list-style:disc;padding-left:20px;">
+      <li><code>.p2s-stage1[is-live|is-settled][is-open]</code> — wraps the execution history inside an <code>.assistant-segment</code>.</li>
+      <li><code>.p2s-worklog</code> — header of Stage 1. Pulsing dot + label + counts + caret. Clickable when settled.</li>
+      <li><code>.p2s-stage1-body</code> — holds <code>.thinking-card</code> + <code>.tool-card-row</code> + <code>.p2s-round-sep</code>. Animated via <code>max-height</code>.</li>
+      <li><code>.p2s-round-sep</code> — inline horizontal separator between tool/reasoning rounds.</li>
+      <li><code>.p2s-transition</code> — thin gradient divider between Stage 1 and Stage 2.</li>
+      <li><code>.p2s-answer</code> — wraps the final <code>.msg-body</code> + <code>.msg-foot</code>.</li>
+      <li><code>.p2s-answer-kicker</code> — small gold <em>Answer</em> label.</li>
+      <li><code>.p2s-clarify-slot</code> — placement slot for <code>.clarify-card</code> at the Stage 1/2 seam.</li>
+    </ul>
+    <h3 style="font-size:13px;color:var(--text);margin:14px 0 8px;">Unchanged</h3>
+    <ul style="color:var(--muted);font-size:12px;line-height:1.9;list-style:disc;padding-left:20px;">
+      <li><code>.thinking-card</code>, <code>.tool-card</code>, <code>.approval-card</code>, <code>.clarify-card</code>, <code>.msg-body</code>, <code>.msg-foot</code> — all existing app CSS and existing markup.</li>
+      <li><code>.assistant-turn-blocks</code> and <code>.assistant-segment</code> remain the top-level wrappers.</li>
+      <li>Tool cards still live as <code>.tool-card-row</code> siblings — now nested <em>inside</em> <code>.p2s-stage1-body</code> rather than as direct children of <code>.messages-inner</code>.</li>
+    </ul>
+    <h3 style="font-size:13px;color:var(--text);margin:14px 0 8px;">Implementation notes</h3>
+    <ul style="color:var(--muted);font-size:12px;line-height:1.9;list-style:disc;padding-left:20px;">
+      <li>Renderer in <code>static/messages.js</code> wraps an assistant turn's non-final blocks in <code>.p2s-stage1-body</code> and appends the <code>.p2s-worklog</code> header once; toggles <code>is-live</code>/<code>is-settled</code> based on <code>data-live-assistant</code>.</li>
+      <li><code>static/boot.js</code> SSE handler ticks the timer while <code>is-live</code>, increments counts on each <code>tool_use</code>, and flips the class when the first Stage 2 delta arrives.</li>
+      <li>Persistence: no schema change needed — the worklog summary can be derived on reload from the existing persisted tool-call list + thinking rounds.</li>
+    </ul>
+  </div>
+</section>
+
+</main>
+
+<script>
+  // Theme picker (matches index.html)
+  document.querySelectorAll('[data-theme-btn]').forEach(btn => {
+    btn.addEventListener('click', () => {
+      const t = btn.dataset.themeBtn;
+      if (t === 'default') document.documentElement.removeAttribute('data-theme');
+      else document.documentElement.setAttribute('data-theme', t);
+      document.querySelectorAll('[data-theme-btn]').forEach(b => b.classList.toggle('on', b === btn));
+    });
+  });
+
+  // Existing thinking/tool cards click-to-toggle.
+  document.querySelectorAll('.thinking-card-header, .tool-card-header').forEach(h => {
+    h.addEventListener('click', (e) => {
+      e.stopPropagation();
+      h.parentElement.classList.toggle('open');
+    });
+  });
+
+  // Click the worklog bar on a settled Stage 1 to expand/collapse the history.
+  document.querySelectorAll('.p2s-stage1[data-p2s-toggle] .p2s-worklog').forEach(bar => {
+    bar.addEventListener('click', () => {
+      const stage = bar.closest('.p2s-stage1');
+      if (!stage.classList.contains('is-settled')) return;
+      stage.classList.toggle('is-open');
+    });
+  });
+
+  // Live timer demo in section 2 — ticks so the page feels alive.
+  (function(){
+    const el = document.getElementById('p2sTimer');
+    if (!el) return;
+    let [m, s] = el.textContent.split(':').map(Number);
+    setInterval(() => {
+      s = (s + 1) % 60;
+      if (s === 0) m += 1;
+      el.textContent = m + ':' + String(s).padStart(2,'0');
+    }, 1000);
+  })();
+</script>
+
+</body>
+</html>

server.py

@@ -3,19 +3,51 @@ Hermes Web UI -- Main server entry point.
 Thin routing shell: imports Handler, delegates to api/routes.py, runs server.
 All business logic lives in api/*.
 """
+import logging
+import socket
+import sys
 import time
 import traceback
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from urllib.parse import urlparse
 
+logger = logging.getLogger(__name__)
+
 from api.auth import check_auth
 from api.config import HOST, PORT, STATE_DIR, SESSION_DIR, DEFAULT_WORKSPACE
-from api.helpers import j
+from api.helpers import j, get_profile_cookie
+from api.profiles import set_request_profile, clear_request_profile
 from api.routes import handle_get, handle_post
+from api.startup import auto_install_agent_deps, fix_credential_permissions
+from api.updates import WEBUI_VERSION
+
+
+class QuietHTTPServer(ThreadingHTTPServer):
+    """Custom HTTP server that silently handles common network errors."""
+    
+    def handle_error(self, request, client_address):
+        """Override to suppress logging for common client disconnect errors."""
+        exc_type, exc_value, _ = sys.exc_info()
+        
+        # Silently ignore common connection errors caused by client disconnects
+        if exc_type in (ConnectionResetError, BrokenPipeError, ConnectionAbortedError):
+            return
+        
+        # Also handle socket errors that indicate client disconnect
+        if exc_type is socket.error:
+            # errno 54 is Connection reset by peer on macOS/BSD
+            # errno 104 is Connection reset by peer on Linux
+            if exc_value.errno in (54, 104, 32):  # ECONNRESET, EPIPE
+                return
+        
+        # For other errors, use default logging
+        super().handle_error(request, client_address)
 
 
 class Handler(BaseHTTPRequestHandler):
-    server_version = 'HermesWebUI/0.2'
+    timeout = 30  # seconds — kills idle/incomplete connections to prevent thread exhaustion
+    _ver_suffix = WEBUI_VERSION.removeprefix('v')
+    server_version = ('HermesWebUI/' + _ver_suffix) if _ver_suffix != 'unknown' else 'HermesWebUI'
     def log_message(self, fmt, *args): pass  # suppress default Apache-style log
 
     def log_request(self, code: str='-', size: str='-') -> None:
@@ -33,6 +65,10 @@ class Handler(BaseHTTPRequestHandler):
 
     def do_GET(self) -> None:
         self._req_t0 = time.time()
+        # Per-request profile context from cookie (issue #798)
+        cookie_profile = get_profile_cookie(self)
+        if cookie_profile:
+            set_request_profile(cookie_profile)
         try:
             parsed = urlparse(self.path)
             if not check_auth(self, parsed): return
@@ -42,9 +78,15 @@ class Handler(BaseHTTPRequestHandler):
         except Exception as e:
             print(f'[webui] ERROR {self.command} {self.path}\n' + traceback.format_exc(), flush=True)
             return j(self, {'error': 'Internal server error'}, status=500)
+        finally:
+            clear_request_profile()
 
     def do_POST(self) -> None:
         self._req_t0 = time.time()
+        # Per-request profile context from cookie (issue #798)
+        cookie_profile = get_profile_cookie(self)
+        if cookie_profile:
+            set_request_profile(cookie_profile)
         try:
             parsed = urlparse(self.path)
             if not check_auth(self, parsed): return
@@ -54,6 +96,8 @@ class Handler(BaseHTTPRequestHandler):
         except Exception as e:
             print(f'[webui] ERROR {self.command} {self.path}\n' + traceback.format_exc(), flush=True)
             return j(self, {'error': 'Internal server error'}, status=500)
+        finally:
+            clear_request_profile()
 
 
 def main() -> None:
@@ -61,23 +105,92 @@ def main() -> None:
 
     print_startup_config()
 
+    # Fix sensitive file permissions before doing anything else
+    fix_credential_permissions()
+
+    within_container = False
+    # Check for the "/.within_container" file to determine if we're running inside a container; this file is created in the Dockerfile
+    try:
+        with open('/.within_container', 'r') as f:
+            within_container = True
+    except FileNotFoundError:
+        pass
+
+    if within_container:
+        print('[ok] Running within container.', flush=True)
+
+    # Security: warn if binding non-loopback without authentication
+    from api.auth import is_auth_enabled
+    if HOST not in ('127.0.0.1', '::1', 'localhost') and not is_auth_enabled():
+        print(f'[!!] WARNING: Binding to {HOST} with NO PASSWORD SET.', flush=True)
+        print(f'     Anyone on the network can access your filesystem and agent.', flush=True)
+        print(f'     Set a password via Settings or HERMES_WEBUI_PASSWORD env var.', flush=True)
+        print(f'     To suppress: bind to 127.0.0.1 or set a password.', flush=True)
+        if within_container:
+            print(f'     Note: You are running within a container, must bind to 0.0.0.0 to publish the port.', flush=True)
+    elif not is_auth_enabled():
+        print(f'  [tip] No password set. Any process on this machine can read sessions', flush=True)
+        print(f'        and memory via the local API. Set HERMES_WEBUI_PASSWORD to', flush=True)
+        print(f'        enable authentication.', flush=True)
+
     ok, missing, errors = verify_hermes_imports()
     if not ok and _HERMES_FOUND:
         print(f'[!!] Warning: Hermes agent found but missing modules: {missing}', flush=True)
         for mod, err in errors.items():
             print(f'     {mod}: {err}', flush=True)
-        print('     Agent features may not work correctly.', flush=True)
+        print('     Attempting to install missing dependencies from agent requirements.txt...', flush=True)
+        auto_install_agent_deps()
+        ok, missing, errors = verify_hermes_imports()
+        if not ok:
+            print(f'[!!] Still missing after install attempt: {missing}', flush=True)
+            for mod, err in errors.items():
+                print(f'     {mod}: {err}', flush=True)
+            print('     Agent features may not work correctly.', flush=True)
+        else:
+            print('[ok] Agent dependencies installed successfully.', flush=True)
 
     STATE_DIR.mkdir(parents=True, exist_ok=True)
     SESSION_DIR.mkdir(parents=True, exist_ok=True)
     DEFAULT_WORKSPACE.mkdir(parents=True, exist_ok=True)
-    httpd = ThreadingHTTPServer((HOST, PORT), Handler)
-    print(f'  Hermes Web UI listening on http://{HOST}:{PORT}', flush=True)
-    if HOST == '127.0.0.1':
+
+    # Start the gateway session watcher for real-time SSE updates
+    try:
+        from api.gateway_watcher import start_watcher
+        start_watcher()
+    except Exception as e:
+        print(f'[!!] WARNING: Gateway watcher failed to start: {e}', flush=True)
+
+    httpd = QuietHTTPServer((HOST, PORT), Handler)
+
+    # ── TLS/HTTPS setup (optional) ─────────────────────────────────────────
+    from api.config import TLS_ENABLED, TLS_CERT, TLS_KEY
+    scheme = 'https' if TLS_ENABLED else 'http'
+    if TLS_ENABLED:
+        try:
+            import ssl
+            ctx = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
+            ctx.minimum_version = ssl.TLSVersion.TLSv1_2
+            ctx.load_cert_chain(TLS_CERT, TLS_KEY)
+            httpd.socket = ctx.wrap_socket(httpd.socket, server_side=True)
+            print(f'  TLS enabled: cert={TLS_CERT}, key={TLS_KEY}', flush=True)
+        except Exception as e:
+            print(f'[!!] WARNING: TLS setup failed ({e}), falling back to HTTP', flush=True)
+            scheme = 'http'
+
+    print(f'  Hermes Web UI listening on {scheme}://{HOST}:{PORT}', flush=True)
+    if HOST == '127.0.0.1' or within_container:
         print(f'  Remote access: ssh -N -L {PORT}:127.0.0.1:{PORT} <user>@<your-server>', flush=True)
-    print(f'  Then open:     http://localhost:{PORT}', flush=True)
+    print(f'  Then open:     {scheme}://localhost:{PORT}', flush=True)
     print('', flush=True)
-    httpd.serve_forever()
+    try:
+        httpd.serve_forever()
+    finally:
+        # Stop the gateway watcher on shutdown
+        try:
+            from api.gateway_watcher import stop_watcher
+            stop_watcher()
+        except Exception:
+            logger.debug("Failed to stop gateway watcher during shutdown")
 
 if __name__ == '__main__':
     main()

start.sh

@@ -1,260 +1,25 @@
 #!/usr/bin/env bash
-# ============================================================
-# Hermes Web UI -- portable bootstrap
-# Usage: ./start.sh [port]
-#
-# One-command startup. Discovers your Hermes install, sets up
-# a local virtualenv if needed, installs dependencies, then
-# launches the server and prints everything you need to know.
-#
-# Override any step with environment variables:
-#   HERMES_WEBUI_AGENT_DIR   path to hermes-agent checkout
-#   HERMES_WEBUI_PYTHON      python executable to use
-#   HERMES_WEBUI_PORT        port to listen on (default: 8787)
-#   HERMES_WEBUI_HOST        bind address (default: 127.0.0.1)
-#   HERMES_HOME              override ~/.hermes base
-#   HERMES_WEBUI_STATE_DIR   override state directory
-# ============================================================
-
 set -euo pipefail
 
-# ── Load .env if present (machine-local overrides, not committed) ─────────────
-_SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-if [[ -f "${_SCRIPT_DIR}/.env" ]]; then
-    set -a
-    # shellcheck source=/dev/null
-    source "${_SCRIPT_DIR}/.env"
-    set +a
-fi
-
-# ── Colours ──────────────────────────────────────────────────────────────────
-RED='\033[0;31m'; GREEN='\033[0;32m'; YELLOW='\033[1;33m'
-CYAN='\033[0;36m'; BOLD='\033[1m'; RESET='\033[0m'
-ok()   { echo -e "${GREEN}[ok]${RESET} $*"; }
-warn() { echo -e "${YELLOW}[!!]${RESET} $*"; }
-die()  { echo -e "${RED}[XX]${RESET} $*" >&2; exit 1; }
-info() { echo -e "${CYAN}[--]${RESET} $*"; }
-hdr()  { echo -e "\n${BOLD}$*${RESET}"; }
-
-# ── Resolve repo root (the directory this script lives in) ───────────────────
 REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-info "Repo root: ${REPO_ROOT}"
 
-# ── Port ─────────────────────────────────────────────────────────────────────
-PORT="${1:-${HERMES_WEBUI_PORT:-8787}}"
-export HERMES_WEBUI_PORT="${PORT}"
-
-# ── Python discovery ─────────────────────────────────────────────────────────
-hdr "Discovering Python..."
-
-_find_python() {
-    # 1. Explicit env var
-    if [[ -n "${HERMES_WEBUI_PYTHON:-}" ]]; then
-        echo "${HERMES_WEBUI_PYTHON}"; return
-    fi
-
-    # 2. Agent venv (discovered below -- call again after agent dir found)
-    # (handled after agent dir discovery)
-
-    # 3. Local .venv in repo
-    if [[ -x "${REPO_ROOT}/.venv/bin/python" ]]; then
-        echo "${REPO_ROOT}/.venv/bin/python"; return
-    fi
-
-    # 4. System python3
-    if command -v python3 &>/dev/null; then
-        echo "$(command -v python3)"; return
-    fi
-
-    echo ""
-}
-
-PYTHON="$(_find_python)"
-
-# ── Hermes agent discovery ────────────────────────────────────────────────────
-hdr "Discovering Hermes agent..."
-
-HERMES_HOME="${HERMES_HOME:-${HOME}/.hermes}"
-AGENT_DIR=""
-
-_find_agent() {
-    local candidates=(
-        "${HERMES_WEBUI_AGENT_DIR:-}"
-        "${HERMES_HOME}/hermes-agent"
-        "${REPO_ROOT}/../hermes-agent"
-        "${HOME}/.hermes/hermes-agent"
-        "${HOME}/hermes-agent"
-    )
-
-    for d in "${candidates[@]}"; do
-        [[ -z "$d" ]] && continue
-        d="$(cd "${d}" 2>/dev/null && pwd || true)"
-        if [[ -n "$d" && -f "${d}/run_agent.py" ]]; then
-            echo "$d"; return
-        fi
-    done
-    echo ""
-}
-
-AGENT_DIR="$(_find_agent)"
-
-if [[ -n "${AGENT_DIR}" ]]; then
-    ok "Hermes agent: ${AGENT_DIR}"
-    export HERMES_WEBUI_AGENT_DIR="${AGENT_DIR}"
-
-    # Now that we have agent dir, prefer its venv if we don't already have a python
-    if [[ -z "${HERMES_WEBUI_PYTHON:-}" && -x "${AGENT_DIR}/venv/bin/python" ]]; then
-        PYTHON="${AGENT_DIR}/venv/bin/python"
-    fi
-else
-    warn "Hermes agent not found. Agent features will not work."
-    warn "Fix with: export HERMES_WEBUI_AGENT_DIR=/path/to/hermes-agent"
+if [[ -f "${REPO_ROOT}/.env" ]]; then
+  set -a
+  # shellcheck source=/dev/null
+  source "${REPO_ROOT}/.env"
+  set +a
 fi
 
-if [[ -n "${PYTHON}" ]]; then
-    ok "Python: ${PYTHON}  ($(${PYTHON} --version 2>&1))"
-else
-    warn "No Python found. Attempting to install..."
-    if command -v apt-get &>/dev/null; then
-        sudo apt-get install -y python3 python3-venv python3-pip
-    elif command -v brew &>/dev/null; then
-        brew install python3
-    else
-        die "Could not find or install Python. Please install Python 3.8+ and re-run."
-    fi
+PYTHON="${HERMES_WEBUI_PYTHON:-}"
+if [[ -z "${PYTHON}" ]]; then
+  if command -v python3 >/dev/null 2>&1; then
     PYTHON="$(command -v python3)"
-    ok "Python installed: ${PYTHON}"
+  elif command -v python >/dev/null 2>&1; then
+    PYTHON="$(command -v python)"
+  else
+    echo "[XX] Python 3 is required to run bootstrap.py" >&2
+    exit 1
+  fi
 fi
 
-# ── Minimum Python version check ─────────────────────────────────────────────
-PY_VER="$(${PYTHON} -c 'import sys; print(f"{sys.version_info.major}.{sys.version_info.minor}")')"
-PY_MAJOR="$(echo "${PY_VER}" | cut -d. -f1)"
-PY_MINOR="$(echo "${PY_VER}" | cut -d. -f2)"
-if [[ "${PY_MAJOR}" -lt 3 || ( "${PY_MAJOR}" -eq 3 && "${PY_MINOR}" -lt 8 ) ]]; then
-    die "Python 3.8+ required. Found: ${PY_VER}"
-fi
-
-# ── Dependency check / local venv setup ──────────────────────────────────────
-hdr "Checking dependencies..."
-
-VENV_NEEDED=false
-VENV_PATH="${REPO_ROOT}/.venv"
-
-# If the chosen python is already the agent venv, its deps are already installed.
-# If it is a system python, check if we can import the webui deps, create a local
-# .venv if not.
-_check_deps() {
-    "${PYTHON}" -c "import yaml" 2>/dev/null
-}
-
-if ! _check_deps; then
-    info "PyYAML not found in ${PYTHON}. Creating local .venv..."
-
-    if [[ ! -d "${VENV_PATH}" ]]; then
-        "${PYTHON}" -m venv "${VENV_PATH}" || die "Failed to create virtualenv at ${VENV_PATH}"
-    fi
-
-    VENV_PY="${VENV_PATH}/bin/python"
-    "${VENV_PY}" -m pip install --quiet --upgrade pip
-
-    if [[ -f "${REPO_ROOT}/requirements.txt" ]]; then
-        info "Installing from requirements.txt..."
-        "${VENV_PY}" -m pip install --quiet -r "${REPO_ROOT}/requirements.txt"
-    else
-        info "Installing minimal deps (pyyaml)..."
-        "${VENV_PY}" -m pip install --quiet pyyaml
-    fi
-
-    PYTHON="${VENV_PY}"
-    ok "Local venv ready: ${VENV_PATH}"
-else
-    ok "Dependencies satisfied."
-fi
-
-# ── Kill any stale instance on the same port ─────────────────────────────────
-hdr "Checking for existing instances..."
-
-EXISTING=$(lsof -ti tcp:"${PORT}" 2>/dev/null || true)
-if [[ -n "${EXISTING}" ]]; then
-    warn "Killing existing process on port ${PORT} (PID ${EXISTING})"
-    kill "${EXISTING}" 2>/dev/null || true
-    sleep 0.5
-fi
-
-# Also kill any server.py process from this repo
-pkill -f "${REPO_ROOT}/server.py" 2>/dev/null || true
-
-# ── Set up working directory for Hermes imports ───────────────────────────────
-# server.py / api/config.py inject agent dir into sys.path at import time,
-# but we also cd into the agent dir so relative imports in run_agent work.
-if [[ -n "${AGENT_DIR}" ]]; then
-    WORKDIR="${AGENT_DIR}"
-else
-    WORKDIR="${REPO_ROOT}"
-fi
-
-# ── Launch ───────────────────────────────────────────────────────────────────
-hdr "Starting Hermes Web UI..."
-
-LOG="/tmp/hermes-webui-${PORT}.log"
-export HERMES_WEBUI_HOST="${HERMES_WEBUI_HOST:-127.0.0.1}"
-export HERMES_WEBUI_STATE_DIR="${HERMES_WEBUI_STATE_DIR:-${HERMES_HOME}/webui}"
-
-nohup "${PYTHON}" "${REPO_ROOT}/server.py" \
-    > "${LOG}" 2>&1 &
-PID=$!
-
-echo -e "\n${CYAN}  PID ${PID} starting...${RESET}"
-sleep 1.5
-
-# ── Health check ─────────────────────────────────────────────────────────────
-HEALTH_URL="http://${HERMES_WEBUI_HOST:-127.0.0.1}:${PORT}/health"
-MAX_WAIT=15
-ELAPSED=0
-while [[ $ELAPSED -lt $MAX_WAIT ]]; do
-    if curl -sf "${HEALTH_URL}" | grep -q '"status"' 2>/dev/null; then
-        break
-    fi
-    sleep 0.5
-    ELAPSED=$((ELAPSED + 1))
-done
-
-if ! curl -sf "${HEALTH_URL}" | grep -q '"status"' 2>/dev/null; then
-    warn "Health check did not pass within ${MAX_WAIT}s. Check log:"
-    tail -20 "${LOG}"
-    echo ""
-    warn "Server may still be starting. Try: curl ${HEALTH_URL}"
-else
-    ok "Server is healthy."
-fi
-
-# ── Print access instructions ─────────────────────────────────────────────────
-BIND_HOST="${HERMES_WEBUI_HOST:-127.0.0.1}"
-
-echo ""
-echo -e "${BOLD}========================================${RESET}"
-echo -e "${GREEN}  Hermes Web UI is running${RESET}"
-echo -e "${BOLD}========================================${RESET}"
-echo ""
-
-if [[ "${BIND_HOST}" == "127.0.0.1" || "${BIND_HOST}" == "localhost" ]]; then
-    # Server is bound to loopback -- detect if we are on a remote machine
-    # by checking if $SSH_CLIENT or $SSH_TTY is set
-    if [[ -n "${SSH_CLIENT:-}" || -n "${SSH_TTY:-}" ]]; then
-        SERVER_IP="$(hostname -I 2>/dev/null | awk '{print $1}' || echo "<your-server-ip>")"
-        echo -e "  You are on a remote machine. To access from your local browser:"
-        echo ""
-        echo -e "  ${CYAN}ssh -N -L ${PORT}:127.0.0.1:${PORT} \$(whoami)@${SERVER_IP}${RESET}"
-        echo ""
-        echo -e "  Then open: ${BOLD}http://localhost:${PORT}${RESET}"
-    else
-        echo -e "  Open: ${BOLD}http://localhost:${PORT}${RESET}"
-    fi
-else
-    echo -e "  Open: ${BOLD}http://${BIND_HOST}:${PORT}${RESET}"
-fi
-
-echo ""
-echo -e "  Log:  ${LOG}"
-echo -e "  PID:  ${PID}"
-echo ""
+exec "${PYTHON}" "${REPO_ROOT}/bootstrap.py" --no-browser "$@"

static/boot.js

@@ -2,13 +2,132 @@ async function cancelStream(){
   const streamId = S.activeStreamId;
   if(!streamId) return;
   try{
-    await fetch(new URL(`/api/chat/cancel?stream_id=${encodeURIComponent(streamId)}`,location.origin).href,{credentials:'include'});
-    const btn=$('btnCancel');if(btn)btn.style.display='none';
-    setStatus('Cancelling…');
-  }catch(e){setStatus('Cancel failed: '+e.message);}
+    await fetch(new URL(`api/chat/cancel?stream_id=${encodeURIComponent(streamId)}`,location.href).href,{credentials:'include'});
+  }catch(e){/* cancel request failed — cleanup below still runs */}
+  // Clear status unconditionally after the cancel request completes.
+  // The SSE cancel event may also fire, but if the connection is already
+  // closed it won't arrive — so we handle cleanup here as the guaranteed path.
+  const btn=$('btnCancel');if(btn)btn.style.display='none';
+  S.activeStreamId=null;
+  setBusy(false);
+  if(typeof setComposerStatus==='function') setComposerStatus('');
+  else setStatus('');
 }
 
 // ── Mobile navigation ──────────────────────────────────────────────────────
+let _workspacePanelMode='closed'; // 'closed' | 'browse' | 'preview'
+
+function _isCompactWorkspaceViewport(){
+  return window.matchMedia('(max-width: 900px)').matches;
+}
+
+function _workspacePanelEls(){
+  return {
+    layout: document.querySelector('.layout'),
+    panel: document.querySelector('.rightpanel'),
+    toggleBtn: $('btnWorkspacePanelToggle'),
+    collapseBtn: $('btnCollapseWorkspacePanel'),
+  };
+}
+
+function _hasWorkspacePreviewVisible(){
+  const preview=$('previewArea');
+  return !!(preview&&preview.classList.contains('visible'));
+}
+
+function _setWorkspacePanelMode(mode){
+  const {layout,panel}= _workspacePanelEls();
+  if(!layout||!panel)return;
+  _workspacePanelMode=(mode==='browse'||mode==='preview')?mode:'closed';
+  const open=_workspacePanelMode!=='closed';
+  document.documentElement.dataset.workspacePanel=open?'open':'closed';
+  // Persist open/closed across refreshes (browse/preview → open; closed → closed)
+  // Do NOT overwrite the user's "keep open" preference — only track runtime state
+  // so that toggleWorkspacePanel(false) from the toolbar doesn't clear the setting.
+  localStorage.setItem('hermes-webui-workspace-panel', open ? 'open' : 'closed');
+  layout.classList.toggle('workspace-panel-collapsed',!open);
+  if(_isCompactWorkspaceViewport()){
+    panel.classList.toggle('mobile-open',open);
+  }else{
+    panel.classList.remove('mobile-open');
+  }
+  syncWorkspacePanelUI();
+}
+
+function syncWorkspacePanelState(){
+  const hasPreview=_hasWorkspacePreviewVisible();
+  if(hasPreview){
+    if(_workspacePanelMode==='closed') _setWorkspacePanelMode('preview');
+    else syncWorkspacePanelUI();
+    return;
+  }
+  if(!S.session){
+    _setWorkspacePanelMode('closed');
+    return;
+  }
+  _setWorkspacePanelMode(_workspacePanelMode==='preview'?'closed':_workspacePanelMode);
+}
+
+function openWorkspacePanel(mode='browse'){
+  if(mode==='browse'&&!S.session&&!_hasWorkspacePreviewVisible())return;
+  if(mode==='preview'&&_workspacePanelMode==='browse'){
+    syncWorkspacePanelUI();
+    return;
+  }
+  _setWorkspacePanelMode(mode);
+}
+
+function closeWorkspacePanel(){
+  _setWorkspacePanelMode('closed');
+}
+
+function ensureWorkspacePreviewVisible(){
+  if(_workspacePanelMode==='closed') _setWorkspacePanelMode('preview');
+  else syncWorkspacePanelUI();
+}
+
+function handleWorkspaceClose(){
+  if(_hasWorkspacePreviewVisible()){
+    clearPreview();
+    return;
+  }
+  closeWorkspacePanel();
+}
+
+function syncWorkspacePanelUI(){
+  const {layout,panel,toggleBtn,collapseBtn}= _workspacePanelEls();
+  if(!layout||!panel)return;
+  const desktopOpen=_workspacePanelMode!=='closed';
+  const mobileOpen=panel.classList.contains('mobile-open');
+  const isCompact=_isCompactWorkspaceViewport();
+  const isOpen=isCompact?mobileOpen:desktopOpen;
+  const canBrowse=!!S.session||_hasWorkspacePreviewVisible();
+  const hasPreview=_hasWorkspacePreviewVisible();
+  if(toggleBtn){
+    toggleBtn.classList.toggle('active',isOpen);
+    toggleBtn.setAttribute('aria-pressed',isOpen?'true':'false');
+    toggleBtn.title=isOpen?'Hide workspace panel':'Show workspace panel';
+    toggleBtn.disabled=!canBrowse;
+  }
+  if(collapseBtn){
+    collapseBtn.title=isCompact?'Close workspace panel':'Hide workspace panel';
+  }
+  const hasSession=!!S.session;
+  ['btnUpDir','btnNewFile','btnNewFolder','btnRefreshPanel'].forEach(id=>{
+    const el=$(id);
+    if(el)el.disabled=!hasSession;
+  });
+  const clearBtn=$('btnClearPreview');
+  if(clearBtn){
+    clearBtn.disabled=!isOpen;
+    clearBtn.title=hasPreview?'Close preview':'Hide workspace panel';
+    // On desktop, only show the X button when a file preview is open.
+    // In browse mode the chevron (btnCollapseWorkspacePanel) already serves
+    // as the close control, so showing both produces a duplicate X.
+    if(!isCompact) clearBtn.style.display=hasPreview?'':'none';
+  }
+}
+
 function toggleMobileSidebar(){
   const sidebar=document.querySelector('.sidebar');
   const overlay=$('mobileOverlay');
@@ -24,16 +143,22 @@ function closeMobileSidebar(){
   if(overlay)overlay.classList.remove('visible');
 }
 function toggleMobileFiles(){
-  const panel=document.querySelector('.rightpanel');
+  toggleWorkspacePanel();
+}
+function toggleWorkspacePanel(force){
+  const {panel}= _workspacePanelEls();
   if(!panel)return;
-  panel.classList.toggle('mobile-open');
+  const currentlyOpen=_workspacePanelMode!=='closed';
+  const nextOpen=typeof force==='boolean'?force:!currentlyOpen;
+  if(!nextOpen){
+    closeWorkspacePanel();
+    return;
+  }
+  const nextMode=_hasWorkspacePreviewVisible()?'preview':'browse';
+  openWorkspacePanel(nextMode);
 }
 function mobileSwitchPanel(name){
-  // Switch the panel content view
   switchPanel(name);
-  // For non-chat panels (tasks, skills, memory, spaces), open the sidebar
-  // so the panel is visible. For 'chat', the content is in the main area —
-  // just close the sidebar so the chat view is unobstructed.
   if(name==='chat'){
     closeMobileSidebar();
   } else {
@@ -44,98 +169,220 @@ function mobileSwitchPanel(name){
       if(overlay)overlay.classList.add('visible');
     }
   }
-  // Update bottom nav active state
-  document.querySelectorAll('.mobile-nav-btn').forEach(btn=>{
-    btn.classList.toggle('active',btn.dataset.panel===name);
-  });
 }
 
-$('btnSend').onclick=()=>{if(window._micActive)_stopMic();send();};
+$('btnSend').onclick=()=>{
+  if(window._micActive){
+    window._micPendingSend=true;
+    _stopMic();
+    return;
+  }
+  send();
+};
 $('btnAttach').onclick=()=>$('fileInput').click();
 
-// ── Voice input (Web Speech API) ─────────────────────────────────────────
+// ── Voice input (Web Speech API + MediaRecorder fallback) ───────────────────
 (function(){
   const SpeechRecognition=window.SpeechRecognition||window.webkitSpeechRecognition;
-  if(!SpeechRecognition) return; // Browser unsupported — mic button stays hidden
+  const _canRecordAudio=!!(navigator.mediaDevices&&navigator.mediaDevices.getUserMedia&&window.MediaRecorder);
+  if(!SpeechRecognition&&!_canRecordAudio) return; // Browser unsupported — mic button stays hidden
+
+  // Persist SR failure across reloads (e.g. Tailscale/network error)
+  const _micForceMediaRecorderKey='mic_force_mediarecorder';
+  let _forceMediaRecorder=!SpeechRecognition||localStorage.getItem(_micForceMediaRecorderKey)==='1';
 
   const btn=$('btnMic');
   const status=$('micStatus');
   const ta=$('msg');
-  btn.style.display=''; // Show button — browser supports speech
-
-  const recognition=new SpeechRecognition();
-  recognition.continuous=false;
-  recognition.interimResults=true;
-  recognition.lang='en-US';
+  const statusText=status?status.querySelector('.status-text'):null;
+  btn.style.display=''; // Show button — browser supports speech recognition or recording fallback
 
+  let recognition=(!_forceMediaRecorder&&SpeechRecognition)?new SpeechRecognition():null;
+  let mediaRecorder=null;
+  let mediaStream=null;
+  let audioChunks=[];
   let _finalText='';
   let _prefix='';
+  let _isRecording=false;
 
   function _setRecording(on){
     window._micActive=on;
     btn.classList.toggle('recording',on);
     status.style.display=on?'':'none';
+    if(statusText) statusText.textContent=on?'Listening':'Listening';
     if(!on){ _finalText=''; _prefix=''; }
   }
 
-  recognition.onstart=()=>{ _finalText=''; };
-
-  recognition.onresult=(event)=>{
-    let interim='';
-    let final=_finalText;
-    for(let i=event.resultIndex;i<event.results.length;i++){
-      const t=event.results[i][0].transcript;
-      if(event.results[i].isFinal){ final+=t; _finalText=final; }
-      else{ interim+=t; }
-    }
-    // Append to whatever was already in the textarea before mic started
-    ta.value=_prefix+(final||interim);
-    autoResize();
-  };
-
-  recognition.onend=()=>{
-    // Commit: prefix + final transcription; trim trailing space if prefix was non-empty
-    const committed=_finalText
+  function _commitTranscript(text){
+    const clean=(text||'').trim();
+    const committed=clean
       ? (_prefix&&!_prefix.endsWith(' ')&&!_prefix.endsWith('\n')
-          ? _prefix+' '+_finalText.trimStart()
-          : _prefix+_finalText)
-      : ta.value; // no speech detected — leave whatever is there
-    _setRecording(false);
+          ? _prefix+' '+clean.trimStart()
+          : _prefix+clean)
+      : ta.value;
     ta.value=committed;
     autoResize();
-  };
+    if(window._micPendingSend){
+      window._micPendingSend=false;
+      send();
+    }
+  }
 
-  recognition.onerror=(event)=>{
-    _setRecording(false);
-    const msgs={
-      'not-allowed':'Microphone access denied. Check browser permissions.',
-      'no-speech':'No speech detected. Try again.',
-      'network':'Speech recognition unavailable.',
-    };
-    showToast(msgs[event.error]||'Voice input error: '+event.error);
-  };
+  async function _transcribeBlob(blob){
+    const ext=(blob.type&&blob.type.includes('ogg'))?'ogg':'webm';
+    const form=new FormData();
+    form.append('file',new File([blob],`voice-input.${ext}`,{type:blob.type||`audio/${ext}`}));
+    setComposerStatus('Transcribing…');
+    try{
+      const res=await fetch('api/transcribe',{method:'POST',body:form});
+      const data=await res.json().catch(()=>({}));
+      if(!res.ok) throw new Error(data.error||'Transcription failed');
+      _commitTranscript(data.transcript||'');
+    }catch(err){
+      window._micPendingSend=false;
+      showToast(err.message||t('mic_network'));
+    }finally{
+      setComposerStatus('');
+    }
+  }
+
+  function _stopTracks(){
+    if(mediaStream){
+      mediaStream.getTracks().forEach(track=>track.stop());
+      mediaStream=null;
+    }
+  }
 
   function _stopMic(){
-    if(window._micActive){ recognition.stop(); }
+    if(!window._micActive) return;
+    if(recognition){
+      recognition.stop();
+      return;
+    }
+    if(mediaRecorder&&mediaRecorder.state!=='inactive'){
+      mediaRecorder.stop();
+      return;
+    }
+    _setRecording(false);
+    _stopTracks();
   }
   window._stopMic=_stopMic; // expose for send-guard above
 
-  btn.onclick=()=>{
+  if(recognition && !_forceMediaRecorder){
+    recognition.continuous=false;
+    recognition.interimResults=true;
+    recognition.lang=(typeof _locale!=='undefined'&&_locale._speech)||'en-US';
+
+    recognition.onstart=()=>{ _finalText=''; };
+
+    recognition.onresult=(event)=>{
+      let interim='';
+      let final=_finalText;
+      for(let i=event.resultIndex;i<event.results.length;i++){
+        const t=event.results[i][0].transcript;
+        if(event.results[i].isFinal){ final+=t; _finalText=final; }
+        else{ interim+=t; }
+      }
+      ta.value=_prefix+(final||interim);
+      autoResize();
+    };
+
+    recognition.onend=()=>{
+      const committed=_finalText
+        ? (_prefix&&!_prefix.endsWith(' ')&&!_prefix.endsWith('\n')
+            ? _prefix+' '+_finalText.trimStart()
+            : _prefix+_finalText)
+        : ta.value;
+      _setRecording(false);
+      ta.value=committed;
+      autoResize();
+      if(window._micPendingSend){
+        window._micPendingSend=false;
+        send();
+      }
+    };
+
+    recognition.onerror=(event)=>{
+      _setRecording(false);
+      window._micPendingSend=false;
+      _isRecording=false;
+      if(event.error==='network'||event.error==='not-allowed'){
+        // Persist SR failure: next reload will skip SpeechRecognition
+        localStorage.setItem(_micForceMediaRecorderKey,'1');
+        _forceMediaRecorder=true;
+        recognition=null;
+      }
+      const msgs={
+        'not-allowed':t('mic_denied'),
+        'no-speech':t('mic_no_speech'),
+        'network':t('mic_network'),
+      };
+      showToast(msgs[event.error]||t('mic_error')+event.error);
+    };
+  }
+
+  btn.onclick=async()=>{
+    // Race-condition guard: ignore rapid double-clicks
+    if(_isRecording){
+      _stopMic();
+      _isRecording=false;
+      return;
+    }
     if(window._micActive){
-      recognition.stop();
-      // _setRecording(false) will be called by onend
-    } else {
-      _finalText='';
-      // Snapshot existing textarea content so we append rather than replace
-      _prefix=ta.value;
+      _stopMic();
+      return;
+    }
+    _isRecording=true;
+    _finalText='';
+    _prefix=ta.value;
+    if(recognition && !_forceMediaRecorder){
       recognition.start();
       _setRecording(true);
+      return;
+    }
+    if(!_canRecordAudio){
+      _isRecording=false;
+      showToast(t('mic_network'));
+      return;
+    }
+    try{
+      mediaStream=await navigator.mediaDevices.getUserMedia({audio:true});
+      const preferredTypes=['audio/webm;codecs=opus','audio/webm','audio/ogg;codecs=opus','audio/ogg'];
+      const mimeType=preferredTypes.find(type=>window.MediaRecorder.isTypeSupported?.(type))||'';
+      mediaRecorder=new MediaRecorder(mediaStream,mimeType?{mimeType}:undefined);
+      audioChunks=[];
+      mediaRecorder.ondataavailable=e=>{if(e.data&&e.data.size)audioChunks.push(e.data);};
+      mediaRecorder.onerror=()=>{
+        _isRecording=false;
+        _setRecording(false);
+        window._micPendingSend=false;
+        _stopTracks();
+        showToast(t('mic_network'));
+      };
+      mediaRecorder.onstop=async()=>{
+        _isRecording=false;
+        const blob=new Blob(audioChunks,{type:mediaRecorder.mimeType||mimeType||'audio/webm'});
+        _setRecording(false);
+        _stopTracks();
+        if(blob.size){ await _transcribeBlob(blob); }
+        else if(window._micPendingSend){
+          window._micPendingSend=false;
+        }
+      };
+      mediaRecorder.start();
+      _setRecording(true);
+    }catch(err){
+      _isRecording=false;
+      window._micPendingSend=false;
+      _stopTracks();
+      showToast(t('mic_denied'));
     }
   };
 })();
 window._micActive=window._micActive||false;
+window._micPendingSend=window._micPendingSend||false;
 $('fileInput').onchange=e=>{addFiles(Array.from(e.target.files));e.target.value='';};
-$('btnNewChat').onclick=async()=>{await newSession();await renderSessionList();$('msg').focus();};
+$('btnNewChat').onclick=async()=>{await newSession();await renderSessionList();closeMobileSidebar();$('msg').focus();};
 $('btnDownload').onclick=()=>{
   if(!S.session)return;
   const blob=new Blob([transcript()],{type:'text/markdown'});
@@ -160,14 +407,16 @@ $('importFileInput').onchange=async(e)=>{
     if(res.ok&&res.session){
       await loadSession(res.session.session_id);
       await renderSessionList();
-      showToast('Session imported');
+      if(_currentPanel==='settings') switchPanel('chat');
+      showToast(t('session_imported'));
     }
   }catch(err){
-    showToast('Import failed: '+(err.message||'Invalid JSON'));
+    showToast(t('import_failed')+(err.message||t('import_invalid_json')));
   }
 };
 // btnRefreshFiles is now panel-icon-btn in header (see HTML)
 function clearPreview(){
+  const closePanelAfter=_workspacePanelMode==='preview';
   const pa=$('previewArea');if(pa)pa.classList.remove('visible');
   const pi=$('previewImg');if(pi){pi.onerror=null;pi.src='';}
   const pm=$('previewMd');if(pm)pm.innerHTML='';
@@ -175,24 +424,48 @@ function clearPreview(){
   const pp=$('previewPathText');if(pp)pp.textContent='';
   const ft=$('fileTree');if(ft)ft.style.display='';
   _previewCurrentPath='';_previewCurrentMode='';_previewDirty=false;
+  // Restore directory breadcrumb after closing file preview
+  if(typeof renderBreadcrumb==='function') renderBreadcrumb();
+  if(closePanelAfter)closeWorkspacePanel();
+  else syncWorkspacePanelUI();
 }
-$('btnClearPreview').onclick=clearPreview;
+$('btnClearPreview').onclick=handleWorkspaceClose;
 // workspacePath click handler removed -- use topbar workspace chip dropdown instead
 $('modelSelect').onchange=async()=>{
   if(!S.session)return;
   const selectedModel=$('modelSelect').value;
+  if(typeof closeModelDropdown==='function') closeModelDropdown();
   localStorage.setItem('hermes-webui-model', selectedModel);
   await api('/api/session/update',{method:'POST',body:JSON.stringify({session_id:S.session.session_id,workspace:S.session.workspace,model:selectedModel})});
-  S.session.model=selectedModel;syncTopbar();
+  S.session.model=selectedModel;
+  if(typeof syncModelChip==='function') syncModelChip();
+  syncTopbar();
+  // Warn if selected model belongs to a different provider than what Hermes is configured for
+  if(typeof _checkProviderMismatch==='function'){
+    const warn=_checkProviderMismatch(selectedModel);
+    if(warn&&typeof showToast==='function') showToast(warn,4000);
+  }
+  // Notify user that model changes only take effect in the next conversation (#419)
+  if(S.messages && S.messages.length > 0 && typeof showToast==='function'){
+    showToast('Model change takes effect in your next conversation', 3000);
+  }
 };
 $('msg').addEventListener('input',()=>{
   autoResize();
   updateSendBtn();
   const text=$('msg').value;
   if(text.startsWith('/')&&text.indexOf('\n')===-1){
-    const prefix=text.slice(1);
-    const matches=getMatchingCommands(prefix);
-    if(matches.length)showCmdDropdown(matches); else hideCmdDropdown();
+    if(typeof getSlashAutocompleteMatches==='function'){
+      getSlashAutocompleteMatches(text).then(matches=>{
+        if(($('msg').value||'')!==text) return;
+        if(matches.length)showCmdDropdown(matches); else hideCmdDropdown();
+      });
+    }else{
+      const prefix=text.slice(1);
+      const matches=getMatchingCommands(prefix);
+      if(matches.length)showCmdDropdown(matches); else hideCmdDropdown();
+    }
+    if(typeof ensureSkillCommandsLoadedForAutocomplete==='function') ensureSkillCommandsLoadedForAutocomplete();
   } else {
     hideCmdDropdown();
   }
@@ -206,11 +479,22 @@ $('msg').addEventListener('keydown',e=>{
     if(e.key==='ArrowDown'){e.preventDefault();navigateCmdDropdown(1);return;}
     if(e.key==='Tab'){e.preventDefault();selectCmdDropdownItem();return;}
     if(e.key==='Escape'){e.preventDefault();hideCmdDropdown();return;}
-    if(e.key==='Enter'&&!e.shiftKey){e.preventDefault();selectCmdDropdownItem();return;}
+    if(e.key==='Enter'&&!e.shiftKey){
+      if(e.isComposing){return;}
+      e.preventDefault();
+      selectCmdDropdownItem();
+      return;
+    }
   }
-  // Send key: respect user preference
+  // Send key: respect user preference.
+  // On touch-primary devices (software keyboard), default to Enter = newline
+  // since there's no physical Shift key. Users send via the Send button.
+  // The 'ctrl+enter' setting also uses this behavior (Enter = newline).
+  // Users can override in Settings by explicitly choosing 'enter' mode.
   if(e.key==='Enter'){
-    if(window._sendKey==='ctrl+enter'){
+    if(e.isComposing){return;}
+    const _mobileDefault=matchMedia('(pointer:coarse)').matches&&window._sendKey==='enter';
+    if(window._sendKey==='ctrl+enter'||_mobileDefault){
       if(e.ctrlKey||e.metaKey){e.preventDefault();send();}
     } else {
       if(!e.shiftKey){e.preventDefault();send();}
@@ -219,14 +503,30 @@ $('msg').addEventListener('keydown',e=>{
 });
 // B14: Cmd/Ctrl+K creates a new chat from anywhere
 document.addEventListener('keydown',async e=>{
+  // Enter on approval card = Allow once (when a button inside the card is focused or
+  // card is visible and focus is not on an input/textarea/select)
+  if(e.key==='Enter'&&!e.metaKey&&!e.ctrlKey&&!e.shiftKey){
+    const card=$('approvalCard');
+    const tag=(document.activeElement||{}).tagName||'';
+    if(card&&card.classList.contains('visible')&&tag!=='TEXTAREA'&&tag!=='INPUT'&&tag!=='SELECT'){
+      e.preventDefault();
+      if(typeof respondApproval==='function') respondApproval('once');
+      return;
+    }
+  }
   if((e.metaKey||e.ctrlKey)&&e.key==='k'){
     e.preventDefault();
-    if(!S.busy){await newSession();await renderSessionList();$('msg').focus();}
+    if(!S.busy){await newSession();await renderSessionList();closeMobileSidebar();$('msg').focus();}
   }
   if(e.key==='Escape'){
-    // Close settings overlay if open
-    const settingsOverlay=$('settingsOverlay');
-    if(settingsOverlay&&settingsOverlay.style.display!=='none'){_closeSettingsPanel();return;}
+    // Close onboarding overlay if open (skip/dismiss the wizard)
+    const onboardingOverlay=$('onboardingOverlay');
+    if(onboardingOverlay&&onboardingOverlay.style.display!=='none'){
+      if(typeof skipOnboarding==='function') skipOnboarding();
+      return;
+    }
+    // Close settings panel if active
+    if(_currentPanel==='settings'){_closeSettingsPanel();return;}
     // Close workspace dropdown
     closeWsDropdown();
     // Clear session search
@@ -251,17 +551,21 @@ $('msg').addEventListener('paste',e=>{
     return new File([blob],`screenshot-${Date.now()}.${ext}`,{type:i.type});
   });
   addFiles(files);
-  setStatus(`Image pasted: ${files.map(f=>f.name).join(', ')}`);
+  setStatus(t('image_pasted')+files.map(f=>f.name).join(', '));
 });
 document.querySelectorAll('.suggestion').forEach(btn=>{
   btn.onclick=()=>{$('msg').value=btn.dataset.msg;send();};
 });
 
+window.addEventListener('resize',()=>{
+  syncWorkspacePanelState();
+});
+
 // Boot: restore last session or start fresh
 // ── Resizable panels ──────────────────────────────────────────────────────
 (function(){
   const SIDEBAR_MIN=180, SIDEBAR_MAX=420;
-  const PANEL_MIN=180,   PANEL_MAX=500;
+  const PANEL_MIN=180,   PANEL_MAX=1200;
 
   function initResize(handleId, targetEl, edge, minW, maxW, storageKey){
     const handle = $(handleId);
@@ -306,10 +610,227 @@ document.querySelectorAll('.suggestion').forEach(btn=>{
   };
 })();
 
+// ── Appearance helpers (theme = light/dark/system, skin = accent color) ──────
+const _SKINS=[
+  {name:'Default',  colors:['#FFD700','#FFBF00','#CD7F32']},
+  {name:'Ares',     colors:['#FF4444','#CC3333','#992222']},
+  {name:'Mono',     colors:['#CCCCCC','#999999','#666666']},
+  {name:'Slate',    colors:['#334155','#475569','#64748b']},
+  {name:'Poseidon', colors:['#0EA5E9','#0284C7','#0369A1']},
+  {name:'Sisyphus', colors:['#A78BFA','#8B5CF6','#7C3AED']},
+  {name:'Charizard',colors:['#FB923C','#F97316','#EA580C']},
+];
+const _VALID_THEMES=new Set(['system','dark','light']);
+const _VALID_SKINS=new Set((_SKINS||[]).map(s=>s.name.toLowerCase()));
+const _LEGACY_THEME_MAP={
+  slate:{theme:'dark',skin:'slate'},
+  solarized:{theme:'dark',skin:'poseidon'},
+  monokai:{theme:'dark',skin:'sisyphus'},
+  nord:{theme:'dark',skin:'slate'},
+  oled:{theme:'dark',skin:'default'},
+};
+let _systemThemeMq=null;
+let _onSystemThemeChange=null;
+
+function _normalizeAppearance(theme,skin){
+  const rawTheme=typeof theme==='string'?theme.trim().toLowerCase():'';
+  const rawSkin=typeof skin==='string'?skin.trim().toLowerCase():'';
+  const legacy=_LEGACY_THEME_MAP[rawTheme];
+  const nextTheme=legacy?legacy.theme:(_VALID_THEMES.has(rawTheme)?rawTheme:'dark');
+  const nextSkin=_VALID_SKINS.has(rawSkin)?rawSkin:(legacy?legacy.skin:'default');
+  return {theme:nextTheme,skin:nextSkin};
+}
+
+function _setResolvedTheme(isDark){
+  document.documentElement.classList.toggle('dark',!!isDark);
+  const link=document.getElementById('prism-theme');
+  if(!link) return;
+  const want=isDark
+    ?'https://cdn.jsdelivr.net/npm/prismjs@1.29.0/themes/prism-tomorrow.min.css'
+    :'https://cdn.jsdelivr.net/npm/prismjs@1.29.0/themes/prism.min.css';
+  const wantIntegrity=isDark
+    ?'sha384-wFjoQjtV1y5jVHbt0p35Ui8aV8GVpEZkyF99OXWqP/eNJDU93D3Ugxkoyh6Y2I4A'
+    :'sha384-rCCjoCPCsizaAAYVoz1Q0CmCTvnctK0JkfCSjx7IIxexTBg+uCKtFYycedUjMyA2';
+  if(link.href!==want){ link.integrity=wantIntegrity; link.href=want; }
+}
+
+function _applyTheme(name){
+  const normalized=_normalizeAppearance(name,'default');
+  if(_systemThemeMq&&_onSystemThemeChange){
+    _systemThemeMq.removeEventListener('change',_onSystemThemeChange);
+    _systemThemeMq=null;
+    _onSystemThemeChange=null;
+  }
+  if(normalized.theme==='system'){
+    _systemThemeMq=window.matchMedia('(prefers-color-scheme:dark)');
+    _onSystemThemeChange=()=>_setResolvedTheme(_systemThemeMq.matches);
+    _setResolvedTheme(_systemThemeMq.matches);
+    _systemThemeMq.addEventListener('change',_onSystemThemeChange);
+    return;
+  }
+  _setResolvedTheme(normalized.theme==='dark');
+}
+
+function _applySkin(name){
+  const key=(name||'default').toLowerCase();
+  if(key==='default') delete document.documentElement.dataset.skin;
+  else document.documentElement.dataset.skin=key;
+}
+
+function _pickTheme(name){
+  const currentSkin=localStorage.getItem('hermes-skin');
+  const appearance=_normalizeAppearance(name,currentSkin);
+  localStorage.setItem('hermes-theme',appearance.theme);
+  localStorage.setItem('hermes-skin',appearance.skin);
+  _applyTheme(appearance.theme);
+  _applySkin(appearance.skin);
+  _syncThemePicker(appearance.theme);
+  _syncSkinPicker(appearance.skin);
+  if(typeof _markSettingsDirty==='function') _markSettingsDirty();
+  const hidden=$('settingsTheme');
+  if(hidden) hidden.value=appearance.theme;
+  const skinHidden=$('settingsSkin');
+  if(skinHidden) skinHidden.value=appearance.skin;
+}
+
+function _pickSkin(name){
+  const appearance=_normalizeAppearance(localStorage.getItem('hermes-theme'),name);
+  localStorage.setItem('hermes-theme',appearance.theme);
+  localStorage.setItem('hermes-skin',appearance.skin);
+  _applyTheme(appearance.theme);
+  _applySkin(appearance.skin);
+  _syncThemePicker(appearance.theme);
+  _syncSkinPicker(appearance.skin);
+  if(typeof _markSettingsDirty==='function') _markSettingsDirty();
+  const hidden=$('settingsSkin');
+  if(hidden) hidden.value=appearance.skin;
+  const themeHidden=$('settingsTheme');
+  if(themeHidden) themeHidden.value=appearance.theme;
+}
+
+function _syncThemePicker(active){
+  document.querySelectorAll('#themePickerGrid .theme-pick-btn').forEach(btn=>{
+    const sel=btn.dataset.themeVal===active;
+    btn.style.borderColor=sel?'var(--accent)':'var(--border2)';
+    btn.style.boxShadow=sel?'0 0 0 1px var(--accent-bg-strong)':'none';
+  });
+}
+
+function _syncSkinPicker(active){
+  document.querySelectorAll('#skinPickerGrid .skin-pick-btn').forEach(btn=>{
+    const sel=btn.dataset.skinVal===active;
+    btn.style.borderColor=sel?'var(--accent)':'var(--border2)';
+    btn.style.boxShadow=sel?'0 0 0 1px var(--accent-bg-strong)':'none';
+  });
+}
+
+function _applyFontSize(size){
+  if(size&&size!=='default'){
+    document.documentElement.dataset.fontSize=size;
+  } else {
+    delete document.documentElement.dataset.fontSize;
+  }
+}
+
+function _pickFontSize(size){
+  localStorage.setItem('hermes-font-size',size);
+  _applyFontSize(size);
+  _syncFontSizePicker(size);
+  if(typeof _markSettingsDirty==='function') _markSettingsDirty();
+  const hidden=$('settingsFontSize');
+  if(hidden) hidden.value=size;
+}
+
+function _syncFontSizePicker(active){
+  document.querySelectorAll('#fontSizePickerGrid .font-size-pick-btn').forEach(btn=>{
+    const sel=btn.dataset.fontSizeVal===(active||'default');
+    btn.style.borderColor=sel?'var(--accent)':'var(--border2)';
+    btn.style.boxShadow=sel?'0 0 0 1px var(--accent-bg-strong)':'none';
+  });
+}
+
+function _buildSkinPicker(activeSkin){
+  const grid=$('skinPickerGrid');
+  if(!grid) return;
+  grid.innerHTML='';
+  for(const skin of _SKINS){
+    const key=skin.name.toLowerCase();
+    const btn=document.createElement('button');
+    btn.type='button';
+    btn.className='skin-pick-btn';
+    btn.dataset.skinVal=key;
+    btn.style.cssText='border:1px solid var(--border2);border-radius:8px;padding:8px 4px;text-align:center;cursor:pointer;background:none;transition:all .15s';
+    btn.onclick=()=>_pickSkin(skin.name);
+    const dots=skin.colors.map(c=>`<span style="display:inline-block;width:10px;height:10px;border-radius:50%;background:${c}"></span>`).join('');
+    btn.innerHTML=`<div style="display:flex;gap:3px;justify-content:center;margin-bottom:4px">${dots}</div><span style="font-size:11px;color:var(--text)">${skin.name}</span>`;
+    grid.appendChild(btn);
+  }
+  _syncSkinPicker((activeSkin||'default').toLowerCase());
+}
+
+function applyBotName(){
+  const name=window._botName||'Hermes';
+  document.title=name;
+  const sidebarH1=document.querySelector('.sidebar-header h1');
+  if(sidebarH1) sidebarH1.textContent=name;
+  const logo=document.querySelector('.sidebar-header .logo');
+  if(logo) logo.textContent=name.charAt(0).toUpperCase();
+  const topbarTitle=$('topbarTitle');
+  if(topbarTitle && (!S.session)) topbarTitle.textContent=name;
+  const msg=$('msg');
+  if(msg) msg.placeholder='Message '+name+'\u2026';
+}
+
 (async()=>{
   // Load send key preference
   let _bootSettings={};
-  try{const s=await api('/api/settings');_bootSettings=s;window._sendKey=s.send_key||'enter';window._showTokenUsage=!!s.show_token_usage;window._showCliSessions=!!s.show_cli_sessions;const _theme=s.theme||'dark';document.documentElement.dataset.theme=_theme;localStorage.setItem('hermes-theme',_theme);}catch(e){window._sendKey='enter';window._showTokenUsage=false;window._showCliSessions=false;_bootSettings={check_for_updates:false};}
+  try{
+    const s=await api('/api/settings');
+    _bootSettings=s;
+    window._sendKey=s.send_key||'enter';
+    window._showTokenUsage=!!s.show_token_usage;
+    window._showCliSessions=!!s.show_cli_sessions;
+    window._soundEnabled=!!s.sound_enabled;
+    window._notificationsEnabled=!!s.notifications_enabled;
+    window._showThinking=s.show_thinking!==false;
+    window._sidebarDensity=(s.sidebar_density==='detailed'?'detailed':'compact');
+    window._botName=s.bot_name||'Hermes';
+    if(s.default_model) window._defaultModel=s.default_model;
+    // Persist default workspace so the blank new-chat page can show it
+    // and workspace actions (New file/folder) work before the first session (#804).
+    if(s.default_workspace) S._profileDefaultWorkspace=s.default_workspace;
+    const appearance=_normalizeAppearance(s.theme,s.skin);
+    localStorage.setItem('hermes-theme',appearance.theme);
+    _applyTheme(appearance.theme);
+    localStorage.setItem('hermes-skin',appearance.skin);
+    _applySkin(appearance.skin);
+    if(typeof setLocale==='function'){
+      const _lang=typeof resolvePreferredLocale==='function'
+        ? resolvePreferredLocale(s.language, localStorage.getItem('hermes-lang'))
+        : (s.language || localStorage.getItem('hermes-lang') || 'en');
+      setLocale(_lang);
+      if(typeof applyLocaleToDOM==='function')applyLocaleToDOM();
+    }
+    applyBotName();
+  }catch(e){
+    window._sendKey='enter';
+    window._showTokenUsage=false;
+    window._showCliSessions=false;
+    window._soundEnabled=false;
+    window._notificationsEnabled=false;
+    window._showThinking=true;
+    window._sidebarDensity='compact';
+    window._botName='Hermes';
+    _bootSettings={check_for_updates:false};
+    if(typeof setLocale==='function'){
+      const _lang=typeof resolvePreferredLocale==='function'
+        ? resolvePreferredLocale(null, localStorage.getItem('hermes-lang'))
+        : (localStorage.getItem('hermes-lang') || 'en');
+      setLocale(_lang);
+      if(typeof applyLocaleToDOM==='function')applyLocaleToDOM();
+    }
+    applyBotName();
+  }
   // Non-blocking update check (fire-and-forget, once per tab session)
   // ?test_updates=1 in URL forces banner display for testing (bypasses sessionStorage guards)
   const _testUpdates=new URLSearchParams(location.search).get('test_updates')==='1';
@@ -322,25 +843,82 @@ document.querySelectorAll('.suggestion').forEach(btn=>{
   // Update profile chip label immediately
   const profileLabel=$('profileChipLabel');
   if(profileLabel) profileLabel.textContent=S.activeProfile||'default';
-  // Fetch available models from server and populate dropdown dynamically
-  await populateModelDropdown();
-  // Restore last-used model preference
-  const savedModel=localStorage.getItem('hermes-webui-model');
-  if(savedModel && $('modelSelect')){
-    $('modelSelect').value=savedModel;
-    // If the value didn't take (model not in list), clear the bad pref
-    if($('modelSelect').value!==savedModel) localStorage.removeItem('hermes-webui-model');
-  }
+  // Fetch available models without blocking session restore. The static HTML
+  // options are enough for first paint; the dynamic provider list can settle
+  // after the saved session is visible.
+  const _modelDropdownReady=populateModelDropdown().then(()=>{
+    const savedModel=localStorage.getItem('hermes-webui-model');
+    if(savedModel && $('modelSelect')){
+      $('modelSelect').value=savedModel;
+      // If the value didn't take (model not in list), clear the bad pref
+      if($('modelSelect').value!==savedModel) localStorage.removeItem('hermes-webui-model');
+      else if(typeof syncModelChip==='function') syncModelChip();
+    }
+    if(S.session) syncTopbar();
+  }).catch(()=>{});
+  window._modelDropdownReady=_modelDropdownReady;
   // Pre-load workspace list so sidebar name is correct from first render
   await loadWorkspaceList();
+  await loadOnboardingWizard();
   _initResizePanels();
+  // Workspace panel restore happens AFTER loadSession so we know if
+  // the session has a workspace — prevents the snap-open-then-closed flash (#576).
+  // Fix #822: clear any browser-restored value before first render. This
+  // covers fresh page loads and reloads. The bfcache restore case is handled
+  // separately below by a `pageshow` listener — the async IIFE here does NOT
+  // re-run when the browser restores the page from bfcache.
+  const _srch = document.getElementById('sessionSearch'); if (_srch) _srch.value = '';
   const saved=localStorage.getItem('hermes-webui-session');
   if(saved){
-    try{await loadSession(saved);await renderSessionList();await checkInflightOnBoot(saved);return;}
+    try{
+      await loadSession(saved);
+      // Restore the panel from localStorage when the session has a workspace.
+      // Preference key takes priority over runtime state so that closing
+      // the panel via toolbar X doesn't suppress the "keep open" setting.
+      const panelPref=localStorage.getItem('hermes-webui-workspace-panel-pref')==='open'
+        || localStorage.getItem('hermes-webui-workspace-panel')==='open';
+      if(S.session&&S.session.workspace&&panelPref){
+        _workspacePanelMode='browse';
+      }
+      S._bootReady=true;
+      syncTopbar();syncWorkspacePanelState();await renderSessionList();if(typeof startGatewaySSE==='function')startGatewaySSE();await checkInflightOnBoot(saved);return;}
     catch(e){localStorage.removeItem('hermes-webui-session');}
   }
   // no saved session - show empty state, wait for user to hit +
+  S._bootReady=true;
+  syncTopbar();
+  syncWorkspacePanelState();
   $('emptyState').style.display='';
   await renderSessionList();
+  // Start real-time gateway session sync if setting is enabled
+  if(typeof startGatewaySSE==='function') startGatewaySSE();
 })();
 
+// Fix #822 (bfcache path): when the browser restores the page from the
+// back-forward cache, the async boot IIFE above does NOT re-run, but the
+// DOM — including any stale value in #sessionSearch — IS restored.  A
+// prior search string would silently hide all sessions via the filter in
+// renderSessionListFromCache().  Clear the field and re-run the full layout
+// sync whenever the page is restored from cache (`event.persisted === true`).
+// Fix #1045: also re-run topbar/workspace/panel state so the rail and layout
+// chrome aren't left in the stale bfcache snapshot.
+window.addEventListener('pageshow', (event) => {
+  if (!event.persisted) return;  // fresh loads are handled by the IIFE above
+  const _srch = document.getElementById('sessionSearch');
+  if (_srch) _srch.value = '';
+  // Close any dropdowns/popovers that were open when the user navigated away.
+  // bfcache freezes DOM state, so a dropdown left open remains open on restore.
+  if (typeof closeModelDropdown === 'function') try { closeModelDropdown(); } catch (_) {}
+  if (typeof closeReasoningDropdown === 'function') try { closeReasoningDropdown(); } catch (_) {}
+  if (typeof closeWsDropdown === 'function') try { closeWsDropdown(); } catch (_) {}
+  if (typeof closeProfileDropdown === 'function') try { closeProfileDropdown(); } catch (_) {}
+  // Re-synchronise layout chrome that the boot IIFE sets up but bfcache
+  // doesn't re-run. Each call is guarded so missing helpers degrade silently.
+  if (typeof syncTopbar === 'function') try { syncTopbar(); } catch (_) {}
+  if (typeof syncWorkspacePanelState === 'function') try { syncWorkspacePanelState(); } catch (_) {}
+  if (typeof renderSessionListFromCache === 'function') {
+    try { renderSessionListFromCache(); } catch (_) {}
+  }
+  // Restart the gateway SSE watcher — the persisted connection is dead after bfcache
+  if (typeof startGatewaySSE === 'function') try { startGatewaySSE(); } catch (_) {}
+});

static/commands.js

@@ -3,16 +3,35 @@
 // (no round-trip to the agent) and shows feedback via toast or local message.
 
 const COMMANDS=[
-  {name:'help',      desc:'List available commands',             fn:cmdHelp},
-  {name:'clear',     desc:'Clear conversation messages',         fn:cmdClear},
-  {name:'compact',   desc:'Compress conversation context',       fn:cmdCompact},
-  {name:'model',     desc:'Switch model (e.g. /model gpt-4o)',  fn:cmdModel,     arg:'model_name'},
-  {name:'workspace', desc:'Switch workspace by name',            fn:cmdWorkspace, arg:'name'},
-  {name:'new',       desc:'Start a new chat session',            fn:cmdNew},
-  {name:'usage',     desc:'Toggle token usage display on/off',   fn:cmdUsage},
-  {name:'theme',     desc:'Switch theme (dark/light/slate/solarized/monokai/nord)', fn:cmdTheme, arg:'name'},
+  // noEcho:true = action-only commands that don't produce a chat response.
+  // Commands without noEcho get a user message echoed to the chat (#840).
+  {name:'help',      desc:t('cmd_help'),             fn:cmdHelp},
+  {name:'clear',     desc:t('cmd_clear'),         fn:cmdClear,     noEcho:true},
+  {name:'compress',  desc:t('cmd_compress'),       fn:cmdCompress, arg:'[focus topic]', noEcho:true},
+  {name:'compact',   desc:t('cmd_compact_alias'),       fn:cmdCompact, noEcho:true},
+  {name:'model',     desc:t('cmd_model'),  fn:cmdModel,     arg:'model_name', subArgs:'models', noEcho:true},
+  {name:'workspace', desc:t('cmd_workspace'),            fn:cmdWorkspace, arg:'name',           noEcho:true},
+  {name:'new',       desc:t('cmd_new'),            fn:cmdNew,       noEcho:true},
+  {name:'usage',     desc:t('cmd_usage'),   fn:cmdUsage,     noEcho:true},
+  {name:'theme',     desc:t('cmd_theme'), fn:cmdTheme, arg:'name',  noEcho:true},
+  {name:'personality', desc:t('cmd_personality'), fn:cmdPersonality, arg:'name', subArgs:'personalities'},
+  {name:'skills',    desc:t('cmd_skills'),   fn:cmdSkills,   arg:'query'},
+  {name:'stop',      desc:t('cmd_stop'),     fn:cmdStop,      noEcho:true},
+  {name:'title',     desc:t('cmd_title'),    fn:cmdTitle,    arg:'[title]'},
+  {name:'retry',     desc:t('cmd_retry'),    fn:cmdRetry,     noEcho:true},
+  {name:'undo',      desc:t('cmd_undo'),     fn:cmdUndo,      noEcho:true},
+  {name:'btw',       desc:t('cmd_btw'),      fn:cmdBtw,       arg:'question', noEcho:true},
+  {name:'background',desc:t('cmd_background'),fn:cmdBackground,arg:'prompt',  noEcho:true},
+  {name:'status',    desc:t('cmd_status'),   fn:cmdStatus},
+  {name:'voice',     desc:t('cmd_voice'),    fn:cmdVoice,     noEcho:true},
+  {name:'reasoning', desc:t('cmd_reasoning'), fn:cmdReasoning, arg:'show|hide|none|minimal|low|medium|high|xhigh', subArgs:['show','hide','none','minimal','low','medium','high','xhigh'], noEcho:true},
 ];
 
+const SLASH_SUBARG_SOURCES={
+  model:{desc:t('cmd_model'), subArgs:'models'},
+  personality:{desc:t('cmd_personality'), subArgs:'personalities'},
+};
+
 function parseCommand(text){
   if(!text.startsWith('/'))return null;
   const parts=text.slice(1).split(/\s+/);
@@ -23,42 +42,192 @@ function parseCommand(text){
 
 function executeCommand(text){
   const parsed=parseCommand(text);
-  if(!parsed)return false;
+  if(!parsed)return null;
   const cmd=COMMANDS.find(c=>c.name===parsed.name);
-  if(!cmd)return false;
-  cmd.fn(parsed.args);
-  return true;
+  if(!cmd)return null;
+  // A handler may return `false` to opt out of interception — e.g. /reasoning
+  // with an effort level falls through so the agent's own handler sees it,
+  // preserving the pre-existing pass-through behaviour for that subcommand.
+  if(cmd.fn(parsed.args)===false)return null;
+  // Return noEcho flag so send() knows whether to echo the command as a user message (#840).
+  return {noEcho:!!cmd.noEcho};
 }
 
 function getMatchingCommands(prefix){
   const q=prefix.toLowerCase();
-  return COMMANDS.filter(c=>c.name.startsWith(q));
+  const matches=COMMANDS.filter(c=>c.name.startsWith(q)).map(c=>({...c,source:'builtin'}));
+  const seen=new Set(matches.map(c=>c.name));
+  for(const [name, spec] of Object.entries(SLASH_SUBARG_SOURCES)){
+    if(!name.startsWith(q)||seen.has(name))continue;
+    matches.push({
+      name,
+      desc:spec.desc,
+      arg:'name',
+      source:'subarg-command',
+    });
+    seen.add(name);
+  }
+  for(const skill of _skillCommandCache){
+    if(!skill.name.startsWith(q)||seen.has(skill.name))continue;
+    matches.push(skill);
+    seen.add(skill.name);
+  }
+  return matches;
+}
+
+let _slashModelCache=null;
+let _slashModelCachePromise=null;
+let _slashPersonalityCache=null;
+let _slashPersonalityCachePromise=null;
+
+function _normalizeSlashSubArg(value){
+  return String(value||'').trim();
+}
+
+function _getSlashModelSubArgsFromDom(){
+  const sel=$('modelSelect');
+  if(!sel) return [];
+  const values=[];
+  for(const opt of Array.from(sel.options||[])){
+    const value=_normalizeSlashSubArg(opt.value||opt.textContent||'');
+    if(value) values.push(value);
+  }
+  return Array.from(new Set(values)).sort((a,b)=>a.localeCompare(b));
+}
+
+async function _loadSlashModelSubArgs(force=false){
+  const domValues=_getSlashModelSubArgsFromDom();
+  if(domValues.length&&!force){
+    _slashModelCache=domValues;
+    return domValues;
+  }
+  if(_slashModelCache&&!force) return _slashModelCache;
+  if(_slashModelCachePromise&&!force) return _slashModelCachePromise;
+  _slashModelCachePromise=(async()=>{
+    try{
+      const data=await api('/api/models');
+      const values=[];
+      for(const group of (data&&data.groups)||[]){
+        for(const model of (group&&group.models)||[]){
+          const id=_normalizeSlashSubArg(model&&model.id);
+          if(id) values.push(id);
+        }
+      }
+      const deduped=Array.from(new Set(values)).sort((a,b)=>a.localeCompare(b));
+      _slashModelCache=deduped;
+      return deduped;
+    }catch(_){
+      _slashModelCache=domValues;
+      return domValues;
+    }finally{
+      _slashModelCachePromise=null;
+    }
+  })();
+  return _slashModelCachePromise;
+}
+
+async function _loadSlashPersonalitySubArgs(force=false){
+  if(_slashPersonalityCache&&!force) return _slashPersonalityCache;
+  if(_slashPersonalityCachePromise&&!force) return _slashPersonalityCachePromise;
+  _slashPersonalityCachePromise=(async()=>{
+    try{
+      const data=await api('/api/personalities');
+      const values=['none'];
+      for(const p of (data&&data.personalities)||[]){
+        const name=_normalizeSlashSubArg(p&&p.name);
+        if(name) values.push(name);
+      }
+      const deduped=Array.from(new Set(values)).sort((a,b)=>a.localeCompare(b));
+      _slashPersonalityCache=deduped;
+      return deduped;
+    }catch(_){
+      _slashPersonalityCache=['none'];
+      return _slashPersonalityCache;
+    }finally{
+      _slashPersonalityCachePromise=null;
+    }
+  })();
+  return _slashPersonalityCachePromise;
+}
+
+function _getSlashSubArgOptions(spec){
+  if(Array.isArray(spec)) return Promise.resolve(spec.slice());
+  if(spec==='models') return _loadSlashModelSubArgs();
+  if(spec==='personalities') return _loadSlashPersonalitySubArgs();
+  return Promise.resolve([]);
+}
+
+function _parseSlashAutocomplete(text){
+  if(!text.startsWith('/')||text.indexOf('\n')!==-1) return null;
+  const raw=text.slice(1);
+  const hasSpace=/\s/.test(raw);
+  const parts=raw.split(/\s+/);
+  const cmdName=(parts[0]||'').toLowerCase();
+  const command=COMMANDS.find(c=>c.name===cmdName);
+  const subArgSource=(command&&command.subArgs)?command:SLASH_SUBARG_SOURCES[cmdName];
+  if(!hasSpace||!subArgSource){
+    return {kind:'commands', query:raw};
+  }
+  const argText=raw.slice(cmdName.length).replace(/^\s+/,'');
+  return {kind:'subargs', command:{name:cmdName, desc:subArgSource.desc, subArgs:subArgSource.subArgs}, query:argText.toLowerCase(), rawQuery:argText};
+}
+
+async function getSlashAutocompleteMatches(text){
+  const parsed=_parseSlashAutocomplete(text);
+  if(!parsed) return [];
+  if(parsed.kind==='commands') return getMatchingCommands(parsed.query);
+  const options=await _getSlashSubArgOptions(parsed.command.subArgs);
+  return options
+    .filter(opt=>String(opt).toLowerCase().startsWith(parsed.query))
+    .map(opt=>({
+      name:parsed.command.name,
+      value:String(opt),
+      desc:parsed.command.desc,
+      source:'subarg',
+      parent:parsed.command.name,
+    }));
+}
+
+function _compressionAnchorMessageKey(m){
+  if(!m||!m.role||m.role==='tool') return null;
+  let content='';
+  try{
+    content=typeof msgContent==='function' ? String(msgContent(m)||'') : String(m.content||'');
+  }catch(_){
+    content=String(m.content||'');
+  }
+  const norm=content.replace(/\s+/g,' ').trim().slice(0,160);
+  const ts=m._ts||m.timestamp||null;
+  const attachments=Array.isArray(m.attachments)?m.attachments.length:0;
+  if(!norm && !attachments && !ts) return null;
+  return {role:String(m.role||''), ts, text:norm, attachments};
 }
 
 // ── Command handlers ────────────────────────────────────────────────────────
 
 function cmdHelp(){
   const lines=COMMANDS.map(c=>{
-    const usage=c.arg?` <${c.arg}>`:'';
+    const usage=c.arg ? (String(c.arg).startsWith('[') ? ` ${c.arg}` : ` <${c.arg}>`) : '';
     return `  /${c.name}${usage} — ${c.desc}`;
   });
-  const msg={role:'assistant',content:'**Available commands:**\n'+lines.join('\n')};
+  const msg={role:'assistant',content:t('available_commands')+'\n'+lines.join('\n')};
   S.messages.push(msg);
   renderMessages();
-  showToast('Type / to see commands');
+  showToast(t('type_slash'));
 }
 
 function cmdClear(){
   if(!S.session)return;
   S.messages=[];S.toolCalls=[];
   clearLiveToolCards();
+  if(typeof clearCompressionUi==='function') clearCompressionUi();
   renderMessages();
   $('emptyState').style.display='';
-  showToast('Conversation cleared');
+  showToast(t('conversation_cleared'));
 }
 
 async function cmdModel(args){
-  if(!args){showToast('Usage: /model <name>');return;}
+  if(!args){showToast(t('model_usage'));return;}
   const sel=$('modelSelect');
   if(!sel)return;
   const q=args.toLowerCase();
@@ -69,45 +238,158 @@ async function cmdModel(args){
       match=opt.value;break;
     }
   }
-  if(!match){showToast(`No model matching "${args}"`);return;}
+  if(!match){showToast(t('no_model_match')+`"${args}"`);return;}
   sel.value=match;
   await sel.onchange();
-  showToast(`Switched to ${match}`);
+  showToast(t('switched_to')+match);
 }
 
 async function cmdWorkspace(args){
-  if(!args){showToast('Usage: /workspace <name>');return;}
+  if(!args){showToast(t('workspace_usage'));return;}
   try{
     const data=await api('/api/workspaces');
     const q=args.toLowerCase();
     const ws=(data.workspaces||[]).find(w=>
       (w.name||'').toLowerCase().includes(q)||w.path.toLowerCase().includes(q)
     );
-    if(!ws){showToast(`No workspace matching "${args}"`);return;}
-    if(!S.session)return;
-    await api('/api/session/update',{method:'POST',body:JSON.stringify({
-      session_id:S.session.session_id,workspace:ws.path,model:S.session.model
-    })});
-    S.session.workspace=ws.path;
-    syncTopbar();await loadDir('.');
-    showToast(`Switched to workspace: ${ws.name||ws.path}`);
-  }catch(e){showToast('Workspace switch failed: '+e.message);}
+    if(!ws){showToast(t('no_workspace_match')+`"${args}"`);return;}
+    if(typeof switchToWorkspace==='function') await switchToWorkspace(ws.path, ws.name||ws.path);
+    else showToast(t('switched_workspace')+(ws.name||ws.path));
+  }catch(e){showToast(t('workspace_switch_failed')+e.message);}
 }
 
 async function cmdNew(){
+  if(typeof clearCompressionUi==='function') clearCompressionUi();
   await newSession();
   await renderSessionList();
   $('msg').focus();
-  showToast('New session created');
+  showToast(t('new_session'));
 }
 
-function cmdCompact(){
-  // Send as a regular message to the agent -- the agent's run_conversation
-  // preflight will detect the high token count and trigger _compress_context.
-  // We send a user message so it appears in the conversation.
-  $('msg').value='Please compress and summarize the conversation context to free up space.';
-  send();
-  showToast('Requesting context compression...');
+async function _runManualCompression(focusTopic){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  let visibleCount=0;
+  try{
+    const sid=S.session.session_id;
+    // Preflight: verify the viewed session still exists before compressing.
+    // This avoids a confusing "not found" toast when the UI is stale.
+    try{
+      const live=await api(`/api/session?session_id=${encodeURIComponent(sid)}`);
+      if(!live||!live.session||live.session.session_id!==sid){
+        throw new Error('session no longer available');
+      }
+      S.session=live.session;
+      S.messages=live.session.messages||[];
+      S.toolCalls=live.session.tool_calls||[];
+    }catch(preflightErr){
+      if(typeof clearCompressionUi==='function') clearCompressionUi();
+      if(typeof _setCompressionSessionLock==='function') _setCompressionSessionLock(null);
+      if(typeof setBusy==='function') setBusy(false);
+      if(typeof setComposerStatus==='function') setComposerStatus('');
+      renderMessages();
+      showToast('Compression failed: '+(preflightErr.message||'session no longer available'));
+      return;
+    }
+    if(typeof setBusy==='function') setBusy(true);
+    const body={session_id:sid};
+    if(focusTopic) body.focus_topic=focusTopic;
+    const visibleMessages=(S.messages||[]).filter(m=>{
+      if(!m||!m.role||m.role==='tool') return false;
+      if(m.role==='assistant'){
+        const hasTc=Array.isArray(m.tool_calls)&&m.tool_calls.length>0;
+        const hasTu=Array.isArray(m.content)&&m.content.some(p=>p&&p.type==='tool_use');
+        if(hasTc||hasTu|| (typeof _messageHasReasoningPayload==='function' && _messageHasReasoningPayload(m))) return true;
+      }
+      return typeof msgContent==='function' ? !!msgContent(m) || !!m.attachments?.length : !!m.content || !!m.attachments?.length;
+    });
+    visibleCount=visibleMessages.length;
+    const anchorVisibleIdx=Math.max(0, visibleCount - 1);
+    const anchorMessageKey=_compressionAnchorMessageKey(visibleMessages[visibleMessages.length-1]||null);
+    const commandText=focusTopic?`/compress ${focusTopic}`:'/compress';
+    if(typeof setCompressionUi==='function'){
+      setCompressionUi({
+        sessionId:S.session.session_id,
+        phase:'running',
+        focusTopic:focusTopic||'',
+        commandText,
+        beforeCount:visibleCount,
+        anchorVisibleIdx,
+        anchorMessageKey,
+      });
+    }
+    if(typeof setComposerStatus==='function') setComposerStatus(t('compressing'));
+    renderMessages();
+    const data=await api('/api/session/compress',{method:'POST',body:JSON.stringify(body)});
+    if(data&&data.session){
+      const currentSid=S.session&&S.session.session_id;
+      if(data.session.session_id&&data.session.session_id!==currentSid){
+        await loadSession(data.session.session_id);
+      }else{
+        S.session=data.session;
+        S.messages=data.session.messages||[];
+        S.toolCalls=data.session.tool_calls||[];
+        clearLiveToolCards();
+        localStorage.setItem('hermes-webui-session',S.session.session_id);
+        syncTopbar();
+        renderMessages();
+        await renderSessionList();
+        updateQueueBadge(S.session.session_id);
+      }
+    }
+    const summary=data&&data.summary;
+    if(typeof setCompressionUi==='function'&&S.session){
+      const referenceMsg=(S.messages||[]).find(m=>typeof _isContextCompactionMessage==='function'&&_isContextCompactionMessage(m));
+      const messageRef=referenceMsg?msgContent(referenceMsg)||String(referenceMsg.content||''):'';
+      const summaryRef=summary&&typeof summary.reference_message==='string' ? String(summary.reference_message||'').trim() : '';
+      // Prefer the persisted compaction handoff when it already exists in session state.
+      // The short summary fallback is only for environments where that message is unavailable.
+      const referenceText=messageRef || summaryRef;
+      const effectiveFocus=(data&&data.focus_topic)||focusTopic||'';
+      setCompressionUi({
+        sessionId:S.session.session_id,
+        phase:'done',
+        focusTopic:effectiveFocus,
+        commandText:effectiveFocus?`/compress ${effectiveFocus}`:'/compress',
+        beforeCount:visibleCount,
+        summary:summary||null,
+        referenceText,
+        anchorVisibleIdx: data?.session?.compression_anchor_visible_idx,
+        anchorMessageKey: data?.session?.compression_anchor_message_key||null,
+      });
+    }
+    if(typeof setComposerStatus==='function') setComposerStatus('');
+    renderMessages();
+    if(typeof _setCompressionSessionLock==='function') _setCompressionSessionLock(null);
+  }catch(e){
+    if(typeof setCompressionUi==='function'){
+      const currentSid=S.session&&S.session.session_id;
+      setCompressionUi({
+        sessionId:currentSid||'',
+        phase:'error',
+        focusTopic:(focusTopic||'').trim(),
+        commandText:focusTopic?`/compress ${focusTopic}`:'/compress',
+        beforeCount:(S.messages||[]).filter(m=>m&&m.role&&m.role!=='tool').length,
+        errorText:`Compression failed: ${e.message}`,
+        anchorVisibleIdx: Math.max(0, visibleCount - 1),
+        anchorMessageKey:null,
+      });
+    }
+    if(typeof _setCompressionSessionLock==='function') _setCompressionSessionLock(null);
+    if(typeof setBusy==='function') setBusy(false);
+    if(typeof setComposerStatus==='function') setComposerStatus('');
+    renderMessages();
+    showToast('Compression failed: '+e.message);
+    return;
+  }
+  if(typeof setBusy==='function') setBusy(false);
+}
+
+async function cmdCompress(args){
+  await _runManualCompression((args||'').trim());
+}
+
+async function cmdCompact(args){
+  await _runManualCompression((args||'').trim());
 }
 
 async function cmdUsage(){
@@ -120,23 +402,316 @@ async function cmdUsage(){
   const cb=$('settingsShowTokenUsage');
   if(cb) cb.checked=next;
   renderMessages();
-  showToast('Token usage '+(next?'on':'off'));
+  showToast(next?t('token_usage_on'):t('token_usage_off'));
 }
 
 async function cmdTheme(args){
-  const themes=['dark','light','slate','solarized','monokai','nord'];
-  if(!args||!themes.includes(args.toLowerCase())){
-    showToast('Usage: /theme '+themes.join('|'));
+  const themes=['system','dark','light'];
+  const skins=(_SKINS||[]).map(s=>s.name.toLowerCase());
+  const legacyThemes=Object.keys(_LEGACY_THEME_MAP||{});
+  const val=(args||'').toLowerCase().trim();
+  // Check if it's a theme
+  if(themes.includes(val)||legacyThemes.includes(val)){
+    const appearance=_normalizeAppearance(
+      val,
+      legacyThemes.includes(val)?null:localStorage.getItem('hermes-skin')
+    );
+    localStorage.setItem('hermes-theme',appearance.theme);
+    localStorage.setItem('hermes-skin',appearance.skin);
+    _applyTheme(appearance.theme);
+    _applySkin(appearance.skin);
+    try{await api('/api/settings',{method:'POST',body:JSON.stringify({theme:appearance.theme,skin:appearance.skin})});}catch(e){}
+    const sel=$('settingsTheme');
+    if(sel)sel.value=appearance.theme;
+    const skinSel=$('settingsSkin');
+    if(skinSel)skinSel.value=appearance.skin;
+    if(typeof _syncThemePicker==='function') _syncThemePicker(appearance.theme);
+    if(typeof _syncSkinPicker==='function') _syncSkinPicker(appearance.skin);
+    showToast(t('theme_set')+appearance.theme+(legacyThemes.includes(val)?` + ${appearance.skin}`:''));
     return;
   }
-  const t=args.toLowerCase();
-  document.documentElement.dataset.theme=t;
-  localStorage.setItem('hermes-theme',t);
-  try{await api('/api/settings',{method:'POST',body:JSON.stringify({theme:t})});}catch(e){}
-  // Update settings dropdown if panel is open
-  const sel=$('settingsTheme');
-  if(sel)sel.value=t;
-  showToast('Theme: '+t);
+  // Check if it's a skin
+  if(skins.includes(val)){
+    const appearance=_normalizeAppearance(localStorage.getItem('hermes-theme'),val);
+    localStorage.setItem('hermes-theme',appearance.theme);
+    localStorage.setItem('hermes-skin',appearance.skin);
+    _applyTheme(appearance.theme);
+    _applySkin(appearance.skin);
+    try{await api('/api/settings',{method:'POST',body:JSON.stringify({theme:appearance.theme,skin:appearance.skin})});}catch(e){}
+    const sel=$('settingsSkin');
+    if(sel)sel.value=appearance.skin;
+    const themeSel=$('settingsTheme');
+    if(themeSel)themeSel.value=appearance.theme;
+    if(typeof _syncThemePicker==='function') _syncThemePicker(appearance.theme);
+    if(typeof _syncSkinPicker==='function') _syncSkinPicker(appearance.skin);
+    showToast(t('theme_set')+appearance.skin);
+    return;
+  }
+  showToast(t('theme_usage')+themes.join('|')+' | '+skins.join('|')+' | legacy:'+legacyThemes.join('|'));
+}
+
+async function cmdSkills(args){
+  try{
+    const data = await api('/api/skills');
+    let skills = data.skills || [];
+    if(args){
+      const q = args.toLowerCase();
+      skills = skills.filter(s =>
+        (s.name||'').toLowerCase().includes(q) ||
+        (s.description||'').toLowerCase().includes(q) ||
+        (s.category||'').toLowerCase().includes(q)
+      );
+    }
+    if(!skills.length){
+      const msg = {role:'assistant', content: args ? `No skills matching "${args}".` : 'No skills found.'};
+      S.messages.push(msg); renderMessages(); return;
+    }
+    // Group by category
+    const byCategory = {};
+    skills.forEach(s => {
+      const cat = s.category || 'General';
+      if(!byCategory[cat]) byCategory[cat] = [];
+      byCategory[cat].push(s);
+    });
+    const lines = [];
+    for(const [cat, items] of Object.entries(byCategory).sort()){
+      lines.push(`**${cat}**`);
+      items.forEach(s => {
+        const desc = s.description ? ` — ${s.description.slice(0,80)}${s.description.length>80?'...':''}` : '';
+        lines.push(`  \`${s.name}\`${desc}`);
+      });
+      lines.push('');
+    }
+    const header = args
+      ? `Skills matching "${args}" (${skills.length}):\n\n`
+      : `Available skills (${skills.length}):\n\n`;
+    S.messages.push({role:'assistant', content: header + lines.join('\n')});
+    renderMessages();
+    showToast(t('type_slash'));
+  }catch(e){
+    showToast('Failed to load skills: '+e.message);
+  }
+}
+
+async function cmdPersonality(args){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  if(!args){
+    // List available personalities
+    try{
+      const data=await api('/api/personalities');
+      if(!data.personalities||!data.personalities.length){
+        showToast(t('no_personalities'));
+        return;
+      }
+      const list=data.personalities.map(p=>`  **${p.name}**${p.description?' — '+p.description:''}`).join('\n');
+      S.messages.push({role:'assistant',content:t('available_personalities')+'\n\n'+list+t('personality_switch_hint')});
+      renderMessages();
+    }catch(e){showToast(t('personalities_load_failed'));}
+    return;
+  }
+  const name=args.trim();
+  if(name.toLowerCase()==='none'||name.toLowerCase()==='default'||name.toLowerCase()==='clear'){
+    try{
+      await api('/api/personality/set',{method:'POST',body:JSON.stringify({session_id:S.session.session_id,name:''})});
+      showToast(t('personality_cleared'));
+    }catch(e){showToast(t('failed_colon')+e.message);}
+    return;
+  }
+  try{
+    const res=await api('/api/personality/set',{method:'POST',body:JSON.stringify({session_id:S.session.session_id,name})});
+    S.messages.push({role:'assistant',content:t('personality_set')+`**${name}**`});
+    renderMessages();
+    showToast(t('personality_set')+name);
+  }catch(e){showToast(t('failed_colon')+e.message);}
+}
+
+async function cmdStop(){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  if(!S.activeStreamId){showToast(t('no_active_task'));return;}
+  if(typeof cancelStream==='function'){await cancelStream();showToast(t('stream_stopped'));}
+  else showToast(t('cancel_unavailable'));
+}
+async function cmdTitle(args){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  const name=(args||'').trim();
+  if(!name){
+    S.messages.push({role:'assistant',content:`${t('title_current')}: **${S.session.title||t('untitled')}**\n\n${t('title_change_hint')}`});
+    renderMessages();return;
+  }
+  try{
+    const r=await api('/api/session/rename',{method:'POST',body:JSON.stringify({session_id:S.session.session_id,title:name})});
+    if(r&&r.error){showToast(r.error);return;}
+    S.session.title=(r&&r.session&&r.session.title)||name;
+    if(typeof syncTopbar==='function')syncTopbar();
+    if(typeof renderSessionList==='function')renderSessionList();
+    showToast(`${t('title_set')} "${S.session.title}"`);
+    S.messages.push({role:'assistant',content:`${t('title_set')} **${S.session.title}**`});
+    renderMessages();
+  }catch(e){showToast(t('failed_colon')+e.message);}
+}
+async function cmdRetry(){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  if(S.session.is_cli_session){showToast(t('cmd_webui_only_session'));return;}
+  const activeSid=S.session.session_id;
+  try{
+    const r=await api('/api/session/retry',{method:'POST',body:JSON.stringify({session_id:activeSid})});
+    if(r&&r.error){showToast(r.error);return;}
+    if(!S.session||S.session.session_id!==activeSid)return;
+    const data=await api('/api/session?session_id='+encodeURIComponent(activeSid));
+    if(data&&data.session){S.messages=data.session.messages||[];S.toolCalls=[];if(typeof clearLiveToolCards==='function')clearLiveToolCards();renderMessages();}
+    $('msg').value=r.last_user_text||'';if(typeof autoResize==='function')autoResize();await send();
+  }catch(e){showToast(t('retry_failed')+e.message);}
+}
+async function cmdUndo(){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  if(S.session.is_cli_session){showToast(t('cmd_webui_only_session'));return;}
+  const activeSid=S.session.session_id;
+  try{
+    const r=await api('/api/session/undo',{method:'POST',body:JSON.stringify({session_id:activeSid})});
+    if(r&&r.error){showToast(r.error);return;}
+    if(!S.session||S.session.session_id!==activeSid)return;
+    const data=await api('/api/session?session_id='+encodeURIComponent(activeSid));
+    if(data&&data.session){S.messages=data.session.messages||[];S.toolCalls=[];if(typeof clearLiveToolCards==='function')clearLiveToolCards();renderMessages();}
+    showToast(`↩ ${t('undid_n_messages')} ${r.removed_count} ${t('undid_messages_suffix')}`);
+  }catch(e){showToast(t('undo_failed')+e.message);}
+}
+async function undoLastExchange(){await cmdUndo();}
+async function cmdBtw(args){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  const question=(args||'').trim();
+  if(!question){showToast(t('cmd_btw_usage'));return;}
+  showToast(t('btw_asking'));
+  const activeSid=S.session.session_id;
+  try{
+    const r=await api('/api/btw',{method:'POST',body:JSON.stringify({session_id:activeSid,question})});
+    if(r&&r.error){showToast(r.error);return;}
+    // Connect to the ephemeral SSE stream
+    const streamId=r.stream_id;
+    const parentSid=r.parent_session_id;
+    if(typeof attachBtwStream==='function') attachBtwStream(parentSid,streamId,question);
+  }catch(e){showToast(t('btw_failed')+e.message);}
+}
+async function cmdBackground(args){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  const prompt=(args||'').trim();
+  if(!prompt){showToast(t('cmd_background_usage'));return;}
+  showToast(t('bg_running'));
+  const activeSid=S.session.session_id;
+  try{
+    const r=await api('/api/background',{method:'POST',body:JSON.stringify({session_id:activeSid,prompt})});
+    if(r&&r.error){showToast(r.error);return;}
+    // Show background badge and start polling
+    if(typeof showBackgroundBadge==='function') showBackgroundBadge(r.task_id);
+    if(typeof startBackgroundPolling==='function') startBackgroundPolling(activeSid,r.task_id,prompt);
+  }catch(e){showToast(t('bg_failed')+e.message);}
+}
+async function cmdStatus(){
+  if(!S.session){showToast(t('no_active_session'));return;}
+  try{
+    const r=await api('/api/session/status?session_id='+encodeURIComponent(S.session.session_id));
+    if(r&&r.error){showToast(r.error);return;}
+    S.messages.push({role:'assistant',content:[`**${t('status_heading')}**`,'',`**${t('status_session_id')}:** \`${r.session_id}\``,`**${t('status_title')}:** ${r.title||t('untitled')}`,`**${t('status_model')}:** ${r.model||t('usage_default_model')}`,`**${t('status_workspace')}:** ${r.workspace}`,`**${t('status_personality')}:** ${r.personality||t('usage_personality_none')}`,`**${t('status_messages')}:** ${r.message_count}`,`**${t('status_agent_running')}:** ${r.agent_running?t('status_yes'):t('status_no')}`,].join('\n')});
+    renderMessages();
+  }catch(e){showToast(t('status_load_failed')+e.message);}
+}
+function cmdReasoning(args){
+  const arg=(args||'').trim().toLowerCase();
+  const BRAIN='\uD83E\uDDE0';
+  // Matches hermes_constants.VALID_REASONING_EFFORTS + 'none' (CLI parity).
+  const EFFORTS=['none','minimal','low','medium','high','xhigh'];
+  // Shared status renderer used by the no-args branch and as a fallback.
+  function _fmtStatus(st){
+    const vis=(st && st.show_reasoning===false)?'off':'on';
+    const eff=(st && st.reasoning_effort)||'default';
+    return BRAIN+' Reasoning effort: '+eff+' \u00B7 display: '+vis
+      +'  |  /reasoning show|hide|none|minimal|low|medium|high|xhigh';
+  }
+  if(!arg){
+    // Status — read from the same config.yaml keys the CLI uses.
+    api('/api/reasoning').then(function(st){showToast(_fmtStatus(st));})
+      .catch(function(){showToast(BRAIN+' /reasoning — status unavailable');});
+    return true;
+  }
+  if(arg==='show'||arg==='on'||arg==='hide'||arg==='off'){
+    const on=(arg==='show'||arg==='on');
+    // Update the UI render gate immediately for responsiveness.
+    window._showThinking=on;
+    if(typeof renderMessages==='function') renderMessages();
+    // Persist via /api/reasoning → config.yaml display.show_reasoning
+    // (CLI reads the same key).  Also mirror into WebUI settings.json
+    // show_thinking so boot.js picks it up on reload without hitting
+    // /api/reasoning on every page load.
+    api('/api/reasoning',{method:'POST',body:JSON.stringify({display:arg})}).catch(function(){});
+    api('/api/settings',{method:'POST',body:JSON.stringify({show_thinking:on})}).catch(function(){});
+    showToast(BRAIN+' Thinking blocks: '+(on?'on':'off')+' (saved)');
+    return true;
+  }
+  if(EFFORTS.includes(arg)){
+    // Persist via /api/reasoning → config.yaml agent.reasoning_effort.
+    // Takes effect on the NEXT session/turn (agent re-reads config at
+    // construction time), matching CLI semantics where `/reasoning high`
+    // also forces an agent re-init.
+    api('/api/reasoning',{method:'POST',body:JSON.stringify({effort:arg})})
+      .then(function(st){
+        const eff=(st && st.reasoning_effort)||arg;
+        showToast(BRAIN+' Reasoning effort: '+eff+' (saved; applies to next turn)');
+        if(typeof _applyReasoningChip==='function') _applyReasoningChip(eff);
+      })
+      .catch(function(e){
+        showToast(BRAIN+' Failed to set effort: '+(e && e.message ? e.message : arg));
+      });
+    return true;
+  }
+  showToast('Unknown argument: '+arg+' \u2014 use show|hide|'+EFFORTS.join('|'));
+  return true;
+}
+function cmdVoice(){
+  const mic=document.getElementById('btnMic');
+  if(mic&&mic.style.display!=='none'&&!mic.disabled){try{mic.click();return;}catch(_){}}
+  showToast(t('cmd_voice_use_mic'));
+}
+let _skillCommandCache=[];
+let _skillCommandLoadPromise=null;
+let _skillCommandCacheReady=false;
+function _skillCommandSlug(name){
+  const raw=String(name||'').trim().toLowerCase();
+  if(!raw)return'';
+  return raw.replace(/[\s_]+/g,'-').replace(/[^a-z0-9-]/g,'').replace(/-{2,}/g,'-').replace(/^-+|-+$/g,'');
+}
+function _buildSkillCommandEntry(skill){
+  const skillName=String(skill&&skill.name||'').trim();
+  const slug=_skillCommandSlug(skillName);
+  if(!slug)return null;
+  if(COMMANDS.some(c=>c.name===slug)) return null;
+  return{name:slug,desc:String(skill&&skill.description||'').trim()||t('slash_skill_desc'),source:'skill',skillName};
+}
+async function loadSkillCommands(force=false){
+  if(_skillCommandCacheReady&&!force)return _skillCommandCache;
+  if(_skillCommandLoadPromise&&!force)return _skillCommandLoadPromise;
+  _skillCommandLoadPromise=(async()=>{
+    try{
+      const data=await api('/api/skills');
+      const deduped=new Map();
+      for(const skill of (data&&data.skills)||[]){const entry=_buildSkillCommandEntry(skill);if(entry&&!deduped.has(entry.name))deduped.set(entry.name,entry);}
+      _skillCommandCache=Array.from(deduped.values()).sort((a,b)=>a.name.localeCompare(b.name));
+    }catch(_){_skillCommandCache=[];}
+    finally{_skillCommandCacheReady=true;_skillCommandLoadPromise=null;}
+    return _skillCommandCache;
+  })();
+  return _skillCommandLoadPromise;
+}
+function refreshSlashCommandDropdown(){
+  const ta=$('msg');if(!ta)return;
+  const text=ta.value||'';
+  if(!text.startsWith('/')||text.indexOf('\n')!==-1){hideCmdDropdown();return;}
+  getSlashAutocompleteMatches(text).then(matches=>{
+    if(($('msg').value||'')!==text) return;
+    if(matches.length)showCmdDropdown(matches);else hideCmdDropdown();
+  });
+}
+function ensureSkillCommandsLoadedForAutocomplete(){
+  if(_skillCommandCacheReady||_skillCommandLoadPromise)return;
+  loadSkillCommands().then(()=>{refreshSlashCommandDropdown();});
 }
 
 // ── Autocomplete dropdown ───────────────────────────────────────────────────
@@ -147,19 +722,36 @@ function showCmdDropdown(matches){
   const dd=$('cmdDropdown');
   if(!dd)return;
   dd.innerHTML='';
-  _cmdSelectedIdx=-1;
+  _cmdSelectedIdx=matches.length?0:-1;
   for(let i=0;i<matches.length;i++){
     const c=matches[i];
     const el=document.createElement('div');
     el.className='cmd-item';
+    if(i===_cmdSelectedIdx) el.classList.add('selected');
     el.dataset.idx=i;
-    const usage=c.arg?` <span class="cmd-item-arg">${esc(c.arg)}</span>`:'';
-    el.innerHTML=`<div class="cmd-item-name">/${esc(c.name)}${usage}</div><div class="cmd-item-desc">${esc(c.desc)}</div>`;
+    const isSubArg=c.source==='subarg';
+    const usage=(!isSubArg&&c.arg)?` <span class="cmd-item-arg">${esc(c.arg)}</span>`:'';
+    const badge=c.source==='skill'?`<span class="cmd-item-badge cmd-item-badge-skill">${esc(t('slash_skill_badge'))}</span>`:'';
+    if(c.source==='skill') el.classList.add('cmd-item-skill');
+    const nameHtml=isSubArg
+      ? `<div class="cmd-item-name"><span class="cmd-item-parent">/${esc(c.parent)}</span> <span class="cmd-item-subarg">${esc(c.value)}</span></div>`
+      : `<div class="cmd-item-name">/${esc(c.name)}${usage}${badge}</div>`;
+    const descHtml=`<div class="cmd-item-desc">${esc(c.desc)}</div>`;
+    el.innerHTML=`${nameHtml}${descHtml}`;
     el.onmousedown=(e)=>{
       e.preventDefault();
-      $('msg').value='/'+c.name+(c.arg?' ':'');
-      hideCmdDropdown();
+      const nextValue=isSubArg?('/'+c.parent+' '+c.value):('/'+c.name+(c.arg?' ':''));
+      $('msg').value=nextValue;
       $('msg').focus();
+      if(!isSubArg&&c.source!=='skill'&&nextValue.endsWith(' ')&&typeof getSlashAutocompleteMatches==='function'){
+        getSlashAutocompleteMatches(nextValue).then(matches=>{
+          if(($('msg').value||'')!==nextValue) return;
+          if(matches.length) showCmdDropdown(matches);
+          else hideCmdDropdown();
+        });
+      }else{
+        hideCmdDropdown();
+      }
     };
     dd.appendChild(el);
   }
@@ -182,6 +774,9 @@ function navigateCmdDropdown(dir){
   if(_cmdSelectedIdx<0)_cmdSelectedIdx=items.length-1;
   if(_cmdSelectedIdx>=items.length)_cmdSelectedIdx=0;
   items[_cmdSelectedIdx].classList.add('selected');
+  // Scroll the newly highlighted item into view so it stays visible when the
+  // dropdown overflows and the user navigates with keyboard (#838).
+  items[_cmdSelectedIdx].scrollIntoView({block:'nearest'});
 }
 
 function selectCmdDropdownItem(){
@@ -195,3 +790,9 @@ function selectCmdDropdownItem(){
   }
   hideCmdDropdown();
 }
+
+// ── Handler aliases (for test-discoverable command registration) ──────────────
+// The COMMANDS array above is the authoritative dispatch table. These aliases
+// allow tooling and tests to discover command handlers by name independently.
+const HANDLERS = {};
+HANDLERS.skills = cmdSkills;

static/favicon.svg

@@ -0,0 +1,20 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 64 64">
+  <rect width="64" height="64" rx="12" fill="#1a1a1a"/>
+  <defs>
+    <linearGradient id="g" x1="0%" y1="0%" x2="0%" y2="100%">
+      <stop offset="0%" style="stop-color:#F5C542;stop-opacity:1"/>
+      <stop offset="100%" style="stop-color:#D4961C;stop-opacity:1"/>
+    </linearGradient>
+  </defs>
+  <rect x="30" y="10" width="4" height="46" rx="2" fill="url(#g)"/>
+  <path d="M30 18 C24 14, 14 14, 10 18 C14 16, 22 16, 28 20" fill="#F5C542" opacity="0.9"/>
+  <path d="M30 22 C26 19, 18 19, 14 22 C18 20, 24 20, 28 24" fill="#D4961C" opacity="0.8"/>
+  <path d="M34 18 C40 14, 50 14, 54 18 C50 16, 42 16, 36 20" fill="#F5C542" opacity="0.9"/>
+  <path d="M34 22 C38 19, 46 19, 50 22 C46 20, 40 20, 36 24" fill="#D4961C" opacity="0.8"/>
+  <path d="M32 48 C22 44, 20 38, 26 34 C20 36, 18 42, 24 46 C18 40, 22 30, 30 28 C24 32, 22 38, 28 42"
+        fill="none" stroke="#F5C542" stroke-width="2.5" stroke-linecap="round"/>
+  <path d="M32 48 C42 44, 44 38, 38 34 C44 36, 46 42, 40 46 C46 40, 42 30, 34 28 C40 32, 42 38, 36 42"
+        fill="none" stroke="#D4961C" stroke-width="2.5" stroke-linecap="round"/>
+  <circle cx="32" cy="10" r="4" fill="#F5C542"/>
+  <circle cx="32" cy="10" r="2" fill="#FFF8E1" opacity="0.7"/>
+</svg>

static/icons.js

@@ -0,0 +1,78 @@
+// ── Lucide icon library (self-hosted SVG paths, no CDN dependency) ──────────
+// All icons are 24×24 viewBox, stroke-based, currentColor.
+// Usage: li('folder')  →  returns a ready-to-embed SVG string
+// The returned SVG uses display:inline-block + vertical-align so it sits
+// neatly beside text in both HTML templates and innerHTML assignments.
+
+const LI_PATHS = {
+  // Navigation tabs
+  'message-square':  '<path d="M21 15a2 2 0 0 1-2 2H7l-4 4V5a2 2 0 0 1 2-2h14a2 2 0 0 1 2 2z"/>',
+  'calendar':        '<rect x="3" y="4" width="18" height="18" rx="2"/><line x1="16" y1="2" x2="16" y2="6"/><line x1="8" y1="2" x2="8" y2="6"/><line x1="3" y1="10" x2="21" y2="10"/>',
+  'layers':          '<path d="M12 2L2 7l10 5 10-5-10-5z"/><path d="M2 17l10 5 10-5"/><path d="M2 12l10 5 10-5"/>',
+  'lightbulb':       '<path d="M12 2a7 7 0 0 1 7 7c0 2.5-1.3 4.7-3.2 6H8.2C6.3 13.7 5 11.5 5 9a7 7 0 0 1 7-7z"/><line x1="9" y1="17" x2="15" y2="17"/><line x1="10" y1="20" x2="14" y2="20"/>',
+  'folder':          '<path d="M22 19a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h5l2 3h9a2 2 0 0 1 2 2z"/>',
+  'list-todo':       '<rect x="3" y="5" width="6" height="6" rx="1"/><path d="m3 17 2 2 4-4"/><path d="M13 6h8"/><path d="M13 12h8"/><path d="M13 18h8"/>',
+  // Editing / actions
+  'pencil':          '<path d="M17 3a2.85 2.83 0 1 1 4 4L7.5 20.5 2 22l1.5-5.5Z"/>',
+  'save':            '<path d="M19 21H5a2 2 0 0 1-2-2V5a2 2 0 0 1 2-2h11l5 5v11a2 2 0 0 1-2 2z"/><polyline points="17 21 17 13 7 13 7 21"/><polyline points="7 3 7 8 15 8"/>',
+  'chevron-down':    '<polyline points="6 9 12 15 18 9"/>',
+  'chevron-right':   '<polyline points="9 18 15 12 9 6"/>',
+  'download':        '<path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4"/><polyline points="7 10 12 15 17 10"/><line x1="12" y1="15" x2="12" y2="3"/>',
+  'upload':          '<path d="M21 15v4a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2v-4"/><polyline points="17 8 12 3 7 8"/><line x1="12" y1="3" x2="12" y2="15"/>',
+  'braces':          '<path d="M8 3H7a2 2 0 0 0-2 2v5a2 2 0 0 1-2 2 2 2 0 0 1 2 2v5c0 1.1.9 2 2 2h1"/><path d="M16 3h1a2 2 0 0 1 2 2v5a2 2 0 0 0 2 2 2 2 0 0 0-2 2v5a2 2 0 0 1-2 2h-1"/>',
+  'trash-2':         '<path d="M3 6h18"/><path d="M19 6v14c0 1-1 2-2 2H7c-1 0-2-1-2-2V6"/><path d="M8 6V4c0-1 1-2 2-2h4c1 0 2 1 2 2v2"/><line x1="10" y1="11" x2="10" y2="17"/><line x1="14" y1="11" x2="14" y2="17"/>',
+  'settings':        '<circle cx="12" cy="12" r="3"/><path d="M19.4 15a1.65 1.65 0 0 0 .33 1.82l.06.06a2 2 0 0 1-2.83 2.83l-.06-.06a1.65 1.65 0 0 0-1.82-.33 1.65 1.65 0 0 0-1 1.51V21a2 2 0 0 1-4 0v-.09A1.65 1.65 0 0 0 9 19.4a1.65 1.65 0 0 0-1.82.33l-.06.06a2 2 0 0 1-2.83-2.83l.06-.06A1.65 1.65 0 0 0 4.68 15a1.65 1.65 0 0 0-1.51-1H3a2 2 0 0 1 0-4h.09A1.65 1.65 0 0 0 4.6 9a1.65 1.65 0 0 0-.33-1.82l-.06-.06a2 2 0 0 1 2.83-2.83l.06.06A1.65 1.65 0 0 0 9 4.68a1.65 1.65 0 0 0 1-1.51V3a2 2 0 0 1 4 0v.09a1.65 1.65 0 0 0 1 1.51 1.65 1.65 0 0 0 1.82-.33l.06-.06a2 2 0 0 1 2.83 2.83l-.06.06A1.65 1.65 0 0 0 19.4 9a1.65 1.65 0 0 0 1.51 1H21a2 2 0 0 1 0 4h-.09a1.65 1.65 0 0 0-1.51 1z"/>',
+  'alert-triangle':  '<path d="M10.29 3.86L1.82 18a2 2 0 0 0 1.71 3h16.94a2 2 0 0 0 1.71-3L13.71 3.86a2 2 0 0 0-3.42 0z"/><line x1="12" y1="9" x2="12" y2="13"/><line x1="12" y1="17" x2="12.01" y2="17"/>',
+  'refresh-cw':      '<polyline points="23 4 23 10 17 10"/><polyline points="1 20 1 14 7 14"/><path d="M3.51 9a9 9 0 0 1 14.85-3.36L23 10M1 14l4.64 4.36A9 9 0 0 0 20.49 15"/>',
+  'undo':            '<path d="M9 14 4 9l5-5"/><path d="M4 9h10.5a5.5 5.5 0 0 1 5.5 5.5v0a5.5 5.5 0 0 1-5.5 5.5H11"/>',
+  'check':           '<polyline points="20 6 9 17 4 12"/>',
+  'lock':            '<rect x="3" y="11" width="18" height="11" rx="2" ry="2"/><path d="M7 11V7a5 5 0 0 1 10 0v4"/>',
+  'star':            '<polygon points="12 2 15.09 8.26 22 9.27 17 14.14 18.18 21.02 12 17.77 5.82 21.02 7 14.14 2 9.27 8.91 8.26 12 2"/>',
+  'x':               '<line x1="18" y1="6" x2="6" y2="18"/><line x1="6" y1="6" x2="18" y2="18"/>',
+  'square':          '<rect x="3" y="3" width="18" height="18" rx="2" ry="2"/>',
+  'plus':            '<line x1="12" y1="5" x2="12" y2="19"/><line x1="5" y1="12" x2="19" y2="12"/>',
+  'arrow-up':        '<line x1="12" y1="19" x2="12" y2="5"/><polyline points="5 12 12 5 19 12"/>',
+  'arrow-right':     '<line x1="5" y1="12" x2="19" y2="12"/><polyline points="12 5 19 12 12 19"/>',
+  'loader':          '<line x1="12" y1="2" x2="12" y2="6"/><line x1="12" y1="18" x2="12" y2="22"/><line x1="4.93" y1="4.93" x2="7.76" y2="7.76"/><line x1="16.24" y1="16.24" x2="19.07" y2="19.07"/><line x1="2" y1="12" x2="6" y2="12"/><line x1="18" y1="12" x2="22" y2="12"/><line x1="4.93" y1="19.07" x2="7.76" y2="16.24"/><line x1="16.24" y1="7.76" x2="19.07" y2="4.93"/>',
+  'pause':           '<rect x="6" y="4" width="4" height="16" rx="1"/><rect x="14" y="4" width="4" height="16" rx="1"/>',
+  // Tool icons
+  'terminal':        '<polyline points="4 17 10 11 4 5"/><line x1="12" y1="19" x2="20" y2="19"/>',
+  'file-text':       '<path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z"/><polyline points="14 2 14 8 20 8"/><line x1="16" y1="13" x2="8" y2="13"/><line x1="16" y1="17" x2="8" y2="17"/><polyline points="10 9 9 9 8 9"/>',
+  'file-pen':        '<path d="M12 22h6a2 2 0 0 0 2-2V7l-5-5H6a2 2 0 0 0-2 2v10"/><path d="M14 2v4a2 2 0 0 0 2 2h4"/><path d="M10.4 19.4 14 16l-4-1 .4 4.4z"/><path d="m14 16 1.5-1.5a2.12 2.12 0 0 1 3 3L17 19"/>',
+  'search':          '<circle cx="11" cy="11" r="8"/><line x1="21" y1="21" x2="16.65" y2="16.65"/>',
+  'globe':           '<circle cx="12" cy="12" r="10"/><line x1="2" y1="12" x2="22" y2="12"/><path d="M12 2a15.3 15.3 0 0 1 4 10 15.3 15.3 0 0 1-4 10 15.3 15.3 0 0 1-4-10 15.3 15.3 0 0 1 4-10z"/>',
+  'play':            '<polygon points="5 3 19 12 5 21 5 3"/>',
+  'wrench':          '<path d="M14.7 6.3a1 1 0 0 0 0 1.4l1.6 1.6a1 1 0 0 0 1.4 0l3.77-3.77a6 6 0 0 1-7.94 7.94l-6.91 6.91a2.12 2.12 0 0 1-3-3l6.91-6.91a6 6 0 0 1 7.94-7.94l-3.76 3.76z"/>',
+  'brain':           '<path d="M9.5 2A2.5 2.5 0 0 1 12 4.5v15a2.5 2.5 0 0 1-4.96-.44 2.5 2.5 0 0 1-2.96-3.08 3 3 0 0 1-.34-5.58 2.5 2.5 0 0 1 1.32-4.24 2.5 2.5 0 0 1 1.98-3A2.5 2.5 0 0 1 9.5 2z"/><path d="M14.5 2A2.5 2.5 0 0 0 12 4.5v15a2.5 2.5 0 0 0 4.96-.44 2.5 2.5 0 0 0 2.96-3.08 3 3 0 0 0 .34-5.58 2.5 2.5 0 0 0-1.32-4.24 2.5 2.5 0 0 0-1.98-3A2.5 2.5 0 0 0 14.5 2z"/>',
+  'book-open':       '<path d="M2 3h6a4 4 0 0 1 4 4v14a3 3 0 0 0-3-3H2z"/><path d="M22 3h-6a4 4 0 0 0-4 4v14a3 3 0 0 1 3-3h7z"/>',
+  'clock':           '<circle cx="12" cy="12" r="10"/><polyline points="12 6 12 12 16 14"/>',
+  'bot':             '<rect x="3" y="11" width="18" height="10" rx="2"/><circle cx="12" cy="5" r="2"/><path d="M12 7v4"/><line x1="8" y1="16" x2="8" y2="16"/><line x1="16" y1="16" x2="16" y2="16"/>',
+  'eye':             '<path d="M1 12s4-8 11-8 11 8 11 8-4 8-11 8-11-8-11-8z"/><circle cx="12" cy="12" r="3"/>',
+  'shuffle':         '<polyline points="16 3 21 3 21 8"/><line x1="4" y1="20" x2="21" y2="3"/><polyline points="21 16 21 21 16 21"/><line x1="15" y1="15" x2="21" y2="21"/><line x1="4" y1="4" x2="9" y2="9"/>',
+  'paperclip':       '<path d="m21.44 11.05-9.19 9.19a6 6 0 0 1-8.49-8.49l9.19-9.19a4 4 0 0 1 5.66 5.66l-9.2 9.19a2 2 0 0 1-2.82-2.82l8.48-8.48"/>',
+  'copy':            '<rect x="9" y="9" width="13" height="13" rx="2" ry="2"/><path d="M5 15H4a2 2 0 0 1-2-2V4a2 2 0 0 1 2-2h9a2 2 0 0 1 2 2v1"/>',
+  'rotate-ccw':      '<path d="M3 2v6h6"/><path d="M3 8a9 9 0 1 0 2.64-4.36L3 8"/>',
+  'user':            '<path d="M20 21a8 8 0 0 0-16 0"/><circle cx="12" cy="7" r="4"/>',
+  // File-type icons
+  'image':           '<rect x="3" y="3" width="18" height="18" rx="2" ry="2"/><circle cx="8.5" cy="8.5" r="1.5"/><polyline points="21 15 16 10 5 21"/>',
+  'file-code':       '<path d="M14 2H6a2 2 0 0 0-2 2v16a2 2 0 0 0 2 2h12a2 2 0 0 0 2-2V8z"/><polyline points="14 2 14 8 20 8"/><polyline points="10 13 8 15 10 17"/><polyline points="14 13 16 15 14 17"/>',
+  'zap':             '<polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"/>',
+  // Suggestion buttons
+  'clipboard-list':  '<path d="M16 4h2a2 2 0 0 1 2 2v14a2 2 0 0 1-2 2H6a2 2 0 0 1-2-2V6a2 2 0 0 1 2-2h2"/><rect x="8" y="2" width="8" height="4" rx="1" ry="1"/><line x1="9" y1="12" x2="15" y2="12"/><line x1="9" y1="16" x2="12" y2="16"/>',
+  'map':             '<polygon points="1 6 1 22 8 18 16 22 23 18 23 2 16 6 8 2 1 6"/><line x1="8" y1="2" x2="8" y2="18"/><line x1="16" y1="6" x2="16" y2="22"/>',
+};
+
+/**
+ * Returns a Lucide SVG string for the given icon name.
+ * @param {string} name  – key in LI_PATHS (e.g. 'folder', 'trash-2')
+ * @param {number} size  – width/height in px (default 16)
+ * @returns {string} SVG element string ready for innerHTML
+ */
+function li(name, size = 16) {
+  const p = LI_PATHS[name];
+  if (!p) { console.warn('li(): unknown icon', name); return ''; }
+  return `<svg width="${size}" height="${size}" viewBox="0 0 24 24" fill="none" `
+       + `stroke="currentColor" stroke-width="2" stroke-linecap="round" `
+       + `stroke-linejoin="round" aria-hidden="true" `
+       + `style="display:inline-block;vertical-align:-0.15em;flex-shrink:0">${p}</svg>`;
+}

static/login.js

@@ -0,0 +1,69 @@
+/* Login page — external script, no inline handlers.
+ * Loaded by the /login route. Reads data attributes from the form for
+ * i18n strings so the server does not need to inject JS literals.
+ */
+document.addEventListener('DOMContentLoaded', function () {
+  var form = document.getElementById('login-form');
+  var input = document.getElementById('pw');
+
+  if (!form || !input) return;
+
+  var invalidPw = form.getAttribute('data-invalid-pw') || 'Invalid password';
+  var connFailed = form.getAttribute('data-conn-failed') || 'Connection failed';
+
+  function showErr(msg) {
+    var err = document.getElementById('err');
+    if (err) { err.textContent = msg; err.style.display = 'block'; }
+  }
+
+  function hideErr() {
+    var err = document.getElementById('err');
+    if (err) { err.style.display = 'none'; }
+  }
+
+  // Return the ?next= redirect path if present and safe, otherwise './'
+  // Guards against open-redirect: rejects protocol-relative (//evil.com),
+  // absolute URLs, backslash variants, and control characters.
+  function _safeNextPath() {
+    try {
+      var raw = new URL(window.location.href).searchParams.get('next');
+      if (!raw) return './';
+      if (raw.charAt(0) !== '/') return './';             // must be path-absolute
+      if (raw.charAt(1) === '/' || raw.charAt(1) === '\\') return './'; // reject // and \\
+      if (/[\x00-\x1f\x7f\s]/.test(raw)) return './';  // reject control chars / whitespace
+      return raw;
+    } catch (_) { return './'; }
+  }
+
+  async function doLogin(e) {
+    e.preventDefault();
+    var pw = input.value;
+    hideErr();
+    try {
+      var res = await fetch('api/auth/login', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({ password: pw }),
+        credentials: 'include',
+      });
+      var data = {};
+      try { data = await res.json(); } catch (_) {}
+      if (res.ok && data.ok) {
+        window.location.href = _safeNextPath();
+      } else {
+        showErr(data.error || invalidPw);
+      }
+    } catch (ex) {
+      showErr(connFailed);
+    }
+  }
+
+  form.addEventListener('submit', doLogin);
+
+  input.addEventListener('keydown', function (e) {
+    if (e.key === 'Enter') {
+      e.preventDefault();
+      doLogin(e);
+    }
+  });
+});

static/manifest.json

@@ -0,0 +1,23 @@
+{
+  "name": "Hermes",
+  "short_name": "Hermes",
+  "description": "Hermes AI Agent Web UI",
+  "start_url": "./",
+  "display": "standalone",
+  "background_color": "#1a1a1a",
+  "theme_color": "#1a1a1a",
+  "orientation": "portrait-primary",
+  "icons": [
+    {
+      "src": "static/favicon.svg",
+      "sizes": "any",
+      "type": "image/svg+xml",
+      "purpose": "any maskable"
+    },
+    {
+      "src": "static/favicon-32.png",
+      "sizes": "32x32",
+      "type": "image/png"
+    }
+  ]
+}

static/onboarding.js

@@ -0,0 +1,411 @@
+const ONBOARDING={status:null,step:0,steps:['system','setup','workspace','password','finish'],form:{provider:'openrouter',workspace:'',model:'',password:'',apiKey:'',baseUrl:''},active:false};
+
+function _getOnboardingSetupProviders(){
+  return (((ONBOARDING.status||{}).setup||{}).providers)||[];
+}
+
+function _getOnboardingSetupProvider(id){
+  return _getOnboardingSetupProviders().find(p=>p.id===id)||null;
+}
+
+function _getOnboardingSetupCategories(){
+  return (((ONBOARDING.status||{}).setup||{}).categories)||[];
+}
+
+/** Render the provider <select> with <optgroup> per category. */
+function _renderProviderSelectOptions(selectedId){
+  const providers=_getOnboardingSetupProviders();
+  const categories=_getOnboardingSetupCategories();
+  const provMap={};
+  providers.forEach(p=>{provMap[p.id]=p;});
+  if(!categories.length){
+    // Fallback: flat list when no categories are available.
+    return providers.map(p=>`<option value="${esc(p.id)}">${esc(p.label)}${p.quick?' — '+esc(t('onboarding_quick_setup_badge')):''}</option>`).join('');
+  }
+  return categories.map(cat=>{
+    const opts=cat.providers.map(pid=>{
+      const p=provMap[pid];
+      if(!p)return '';
+      return `<option value="${esc(p.id)}"${p.id===selectedId?' selected':''}>${esc(p.label)}${p.quick?' — '+esc(t('onboarding_quick_setup_badge')):''}</option>`;
+    }).join('');
+    return `<optgroup label="${esc(t('provider_category_'+cat.id)||cat.label)}">${opts}</optgroup>`;
+  }).join('');
+}
+
+function _getOnboardingCurrentSetup(){
+  return (((ONBOARDING.status||{}).setup||{}).current)||{};
+}
+
+function _onboardingStepMeta(key){
+  return ({
+    system:{title:t('onboarding_step_system_title'),desc:t('onboarding_step_system_desc')},
+    setup:{title:t('onboarding_step_setup_title'),desc:t('onboarding_step_setup_desc')},
+    workspace:{title:t('onboarding_step_workspace_title'),desc:t('onboarding_step_workspace_desc')},
+    password:{title:t('onboarding_step_password_title'),desc:t('onboarding_step_password_desc')},
+    finish:{title:t('onboarding_step_finish_title'),desc:t('onboarding_step_finish_desc')}
+  })[key];
+}
+
+function _renderOnboardingSteps(){
+  const wrap=$('onboardingSteps');
+  if(!wrap)return;
+  wrap.innerHTML='';
+  ONBOARDING.steps.forEach((key,idx)=>{
+    const meta=_onboardingStepMeta(key);
+    const item=document.createElement('div');
+    item.className='onboarding-step'+(idx===ONBOARDING.step?' active':idx<ONBOARDING.step?' done':'');
+    item.innerHTML=`<div class="onboarding-step-index">${idx+1}</div><div><div class="onboarding-step-title">${meta.title}</div><div class="onboarding-step-desc">${meta.desc}</div></div>`;
+    wrap.appendChild(item);
+  });
+}
+
+function _setOnboardingNotice(msg,kind='info'){
+  const el=$('onboardingNotice');
+  if(!el)return;
+  if(!msg){el.style.display='none';el.textContent='';el.className='onboarding-status';return;}
+  el.style.display='block';
+  el.className='onboarding-status '+kind;
+  el.textContent=msg;
+}
+
+function _getOnboardingWorkspaceChoices(){
+  const items=((ONBOARDING.status||{}).workspaces||{}).items||[];
+  return items.length?items:[{name:'Home',path:ONBOARDING.form.workspace||''}];
+}
+
+function _getOnboardingProviderModelChoices(){
+  const provider=_getOnboardingSetupProvider(ONBOARDING.form.provider);
+  return provider?(provider.models||[]):[];
+}
+
+function _getOnboardingSelectedModel(){
+  return ONBOARDING.form.model||'';
+}
+
+function _renderOnboardingModelField(){
+  const choices=_getOnboardingProviderModelChoices();
+  if(ONBOARDING.form.provider==='custom'){
+    return `<label class="onboarding-field"><span>${t('onboarding_model_label')}</span><input id="onboardingModelInput" value="${esc(_getOnboardingSelectedModel())}" placeholder="${t('onboarding_custom_model_placeholder')}" oninput="ONBOARDING.form.model=this.value"></label><p class="onboarding-copy">${t('onboarding_custom_model_help')}</p>`;
+  }
+  const options=choices.map(m=>`<option value="${esc(m.id)}">${esc(m.label)}</option>`).join('');
+  return `<label class="onboarding-field"><span>${t('onboarding_model_label')}</span><select id="onboardingModelSelect" onchange="ONBOARDING.form.model=this.value">${options}</select></label><p class="onboarding-copy">${t('onboarding_workspace_help')}</p>`;
+}
+
+function _providerStatusLabel(system){
+  if(system.chat_ready) return t('onboarding_check_provider_ready');
+  if(system.provider_configured) return t('onboarding_check_provider_partial');
+  return t('onboarding_check_provider_pending');
+}
+
+function _renderOnboardingBody(){
+  const body=$('onboardingBody');
+  if(!body||!ONBOARDING.status)return;
+  const key=ONBOARDING.steps[ONBOARDING.step];
+  const system=ONBOARDING.status.system||{};
+  const settings=ONBOARDING.status.settings||{};
+  const setup=ONBOARDING.status.setup||{};
+  const nextBtn=$('onboardingNextBtn');
+  const backBtn=$('onboardingBackBtn');
+  if(backBtn) backBtn.style.display=ONBOARDING.step>0?'':'none';
+  if(nextBtn) nextBtn.textContent=key==='finish'?t('onboarding_open'):t('onboarding_continue');
+
+  if(key==='system'){
+    const hermesOk=system.hermes_found&&system.imports_ok;
+    const setupOk=!!system.chat_ready;
+    _setOnboardingNotice(system.provider_note|| (setupOk?t('onboarding_notice_system_ready'):t('onboarding_notice_system_unavailable')),setupOk?'success':(hermesOk?'info':'warn'));
+    body.innerHTML=`
+      <div class="onboarding-panel-grid">
+        <div class="onboarding-check ${hermesOk?'ok':'warn'}"><strong>${t('onboarding_check_agent')}</strong><span>${hermesOk?t('onboarding_check_agent_ready'):t('onboarding_check_agent_missing')}</span></div>
+        <div class="onboarding-check ${(setupOk?'ok':system.provider_configured?'warn':'muted')}"><strong>${t('onboarding_check_provider')}</strong><span>${_providerStatusLabel(system)}</span></div>
+        <div class="onboarding-check ${(settings.password_enabled?'ok':'muted')}"><strong>${t('onboarding_check_password')}</strong><span>${settings.password_enabled?t('onboarding_check_password_enabled'):t('onboarding_check_password_disabled')}</span></div>
+      </div>
+      <div class="onboarding-copy">
+        <p><strong>${t('onboarding_config_file')}</strong> ${esc(system.config_path||t('onboarding_unknown'))}</p>
+        <p><strong>${t('onboarding_env_file')}</strong> ${esc(system.env_path||t('onboarding_unknown'))}</p>
+        <p>${esc(system.provider_note||'')}</p>
+        ${system.current_provider?`<p><strong>${t('onboarding_current_provider')}</strong> ${esc(system.current_provider)}${system.current_model?` — ${esc(system.current_model)}`:''}</p>`:''}
+        ${system.current_base_url?`<p><strong>${t('onboarding_base_url_label')}</strong> ${esc(system.current_base_url)}</p>`:''}
+        ${system.missing_modules&&system.missing_modules.length?`<p><strong>${t('onboarding_missing_imports')}</strong> ${esc(system.missing_modules.join(', '))}</p>`:''}
+      </div>`;
+    return;
+  }
+
+  if(key==='setup'){
+    const selectedId=ONBOARDING.form.provider;
+    const groupedOptions=_renderProviderSelectOptions(selectedId);
+    const provider=_getOnboardingSetupProvider(selectedId)||_getOnboardingSetupProviders()[0]||null;
+    const showBaseUrl=provider&&provider.requires_base_url;
+    const keyHelp=provider?`${t('onboarding_api_key_help_prefix')} ${esc(provider.env_var)}.`:'';
+
+    // OAuth provider path: configured via CLI, no API key input needed.
+    const currentIsOauth=!!(ONBOARDING.status.setup||{}).current_is_oauth;
+    const currentProviderName=((ONBOARDING.status.setup||{}).current||{}).provider||'';
+    if(currentIsOauth){
+      const isReady=!!(ONBOARDING.status.system||{}).chat_ready;
+      const providerLabel=esc(currentProviderName);
+      if(isReady){
+        _setOnboardingNotice(t('onboarding_notice_setup_already_ready'),'success');
+        body.innerHTML=`
+          <div class="onboarding-oauth-card onboarding-oauth-ready">
+            <div class="onboarding-oauth-icon">✓</div>
+            <div>
+              <strong>${t('onboarding_oauth_provider_ready_title')}</strong>
+              <p>${t('onboarding_oauth_provider_ready_body').replace('{provider}',providerLabel)}</p>
+            </div>
+          </div>
+          <p class="onboarding-copy" style="margin-top:20px">${t('onboarding_oauth_switch_hint')}</p>
+          <label class="onboarding-field">
+            <span>${t('onboarding_provider_label')}</span>
+            <select id="onboardingProviderSelect" onchange="syncOnboardingProvider(this.value)">${groupedOptions}</select>
+          </label>
+          <label class="onboarding-field" id="onboardingApiKeyField">
+            <span>${t('onboarding_api_key_label')}</span>
+            <input id="onboardingApiKeyInput" type="password" value="${esc(ONBOARDING.form.apiKey||'')}" placeholder="${t('onboarding_api_key_placeholder')}" oninput="ONBOARDING.form.apiKey=this.value">
+          </label>
+          ${showBaseUrl?`<label class="onboarding-field"><span>${t('onboarding_base_url_label')}</span><input id="onboardingBaseUrlInput" value="${esc(ONBOARDING.form.baseUrl||'')}" placeholder="${t('onboarding_base_url_placeholder')}" oninput="ONBOARDING.form.baseUrl=this.value"></label>`:''}
+          <p class="onboarding-copy">${keyHelp}</p>`;
+      } else {
+        _setOnboardingNotice(t('onboarding_notice_setup_required'),'warn');
+        body.innerHTML=`
+          <div class="onboarding-oauth-card onboarding-oauth-pending">
+            <div class="onboarding-oauth-icon">⚠</div>
+            <div>
+              <strong>${t('onboarding_oauth_provider_not_ready_title')}</strong>
+              <p>${t('onboarding_oauth_provider_not_ready_body').replace('{provider}',providerLabel)}</p>
+            </div>
+          </div>
+          <p class="onboarding-copy" style="margin-top:20px">${t('onboarding_oauth_switch_hint')}</p>
+          <label class="onboarding-field">
+            <span>${t('onboarding_provider_label')}</span>
+            <select id="onboardingProviderSelect" onchange="syncOnboardingProvider(this.value)">${groupedOptions}</select>
+          </label>
+          <label class="onboarding-field" id="onboardingApiKeyField">
+            <span>${t('onboarding_api_key_label')}</span>
+            <input id="onboardingApiKeyInput" type="password" value="${esc(ONBOARDING.form.apiKey||'')}" placeholder="${t('onboarding_api_key_placeholder')}" oninput="ONBOARDING.form.apiKey=this.value">
+          </label>
+          ${showBaseUrl?`<label class="onboarding-field"><span>${t('onboarding_base_url_label')}</span><input id="onboardingBaseUrlInput" value="${esc(ONBOARDING.form.baseUrl||'')}" placeholder="${t('onboarding_base_url_placeholder')}" oninput="ONBOARDING.form.baseUrl=this.value"></label>`:''}
+          <p class="onboarding-copy">${keyHelp}</p>`;
+      }
+      return;
+    }
+
+    _setOnboardingNotice(system.chat_ready?t('onboarding_notice_setup_already_ready'):t('onboarding_notice_setup_required'),system.chat_ready?'success':'info');
+    body.innerHTML=`
+      <label class="onboarding-field">
+        <span>${t('onboarding_provider_label')}</span>
+        <select id="onboardingProviderSelect" onchange="syncOnboardingProvider(this.value)">${groupedOptions}</select>
+      </label>
+      <label class="onboarding-field">
+        <span>${t('onboarding_api_key_label')}</span>
+        <input id="onboardingApiKeyInput" type="password" value="${esc(ONBOARDING.form.apiKey||'')}" placeholder="${t('onboarding_api_key_placeholder')}" oninput="ONBOARDING.form.apiKey=this.value">
+      </label>
+      ${showBaseUrl?`<label class="onboarding-field"><span>${t('onboarding_base_url_label')}</span><input id="onboardingBaseUrlInput" value="${esc(ONBOARDING.form.baseUrl||'')}" placeholder="${t('onboarding_base_url_placeholder')}" oninput="ONBOARDING.form.baseUrl=this.value"></label>`:''}
+      <p class="onboarding-copy">${keyHelp}</p>
+      ${showBaseUrl?`<p class="onboarding-copy">${t('onboarding_base_url_help')}</p>`:''}
+      <p class="onboarding-copy">${esc(setup.unsupported_note||'')||''}</p>`;
+    return;
+  }
+
+  if(key==='workspace'){
+    const workspaceOptions=_getOnboardingWorkspaceChoices().map(ws=>`<option value="${esc(ws.path)}">${esc(ws.name||ws.path)} — ${esc(ws.path)}</option>`).join('');
+    _setOnboardingNotice(t('onboarding_notice_workspace'), 'info');
+    body.innerHTML=`
+      <label class="onboarding-field">
+        <span>${t('onboarding_workspace_label')}</span>
+        <select id="onboardingWorkspaceSelect" onchange="syncOnboardingWorkspaceSelect(this.value)">${workspaceOptions}</select>
+      </label>
+      <label class="onboarding-field">
+        <span>${t('onboarding_workspace_or_path')}</span>
+        <input id="onboardingWorkspaceInput" value="${esc(ONBOARDING.form.workspace||'')}" placeholder="${t('onboarding_workspace_placeholder')}" oninput="ONBOARDING.form.workspace=this.value">
+      </label>
+      ${_renderOnboardingModelField()}`;
+    const wsSel=$('onboardingWorkspaceSelect');
+    if(wsSel && ONBOARDING.form.workspace) wsSel.value=ONBOARDING.form.workspace;
+    const modelSel=$('onboardingModelSelect');
+    if(modelSel && ONBOARDING.form.model) modelSel.value=ONBOARDING.form.model;
+    return;
+  }
+
+  if(key==='password'){
+    _setOnboardingNotice(settings.password_enabled?t('onboarding_notice_password_enabled'):t('onboarding_notice_password_recommended'), settings.password_enabled?'success':'info');
+    body.innerHTML=`
+      <label class="onboarding-field">
+        <span>${t('onboarding_password_label')}</span>
+        <input id="onboardingPasswordInput" type="password" value="${esc(ONBOARDING.form.password||'')}" placeholder="${t('onboarding_password_placeholder')}" oninput="ONBOARDING.form.password=this.value">
+      </label>
+      <p class="onboarding-copy">${t('onboarding_password_help')}</p>`;
+    return;
+  }
+
+  const provider=_getOnboardingSetupProvider(ONBOARDING.form.provider);
+  _setOnboardingNotice(t('onboarding_notice_finish'), 'success');
+  body.innerHTML=`
+    <div class="onboarding-summary">
+      <div><strong>${t('onboarding_provider_label')}</strong><span>${esc((provider&&provider.label)||ONBOARDING.form.provider||t('onboarding_not_set'))}</span></div>
+      <div><strong>${t('onboarding_model_label')}</strong><span>${esc(_getOnboardingSelectedModel()||t('onboarding_not_set'))}</span></div>
+      <div><strong>${t('onboarding_workspace_label')}</strong><span>${esc(ONBOARDING.form.workspace||t('onboarding_not_set'))}</span></div>
+      <div><strong>${t('onboarding_check_password')}</strong><span>${t(_getOnboardingPasswordSummaryKey(settings))}</span></div>
+    </div>
+    ${ONBOARDING.form.baseUrl?`<p class="onboarding-copy"><strong>${t('onboarding_base_url_label')}</strong> ${esc(ONBOARDING.form.baseUrl)}</p>`:''}
+    <p class="onboarding-copy">${t('onboarding_finish_help')}</p>`;
+}
+
+function _getOnboardingPasswordSummaryKey(settings){
+  const hasExistingPassword=!!(settings&&settings.password_enabled);
+  const hasNewPassword=!!((ONBOARDING.form.password||'').trim());
+  if(hasNewPassword) return hasExistingPassword?'onboarding_password_will_replace':'onboarding_password_will_enable';
+  return hasExistingPassword?'onboarding_password_keep_existing':'onboarding_password_remains_disabled';
+}
+
+function syncOnboardingWorkspaceSelect(value){
+  ONBOARDING.form.workspace=value;
+  const input=$('onboardingWorkspaceInput');
+  if(input) input.value=value;
+}
+
+function syncOnboardingProvider(value){
+  const provider=_getOnboardingSetupProvider(value);
+  ONBOARDING.form.provider=value;
+  if(provider){
+    if(!ONBOARDING.form.model || !_getOnboardingProviderModelChoices().some(m=>m.id===ONBOARDING.form.model) || value==='custom'){
+      ONBOARDING.form.model=provider.default_model||'';
+    }
+    if(provider.requires_base_url){
+      ONBOARDING.form.baseUrl=ONBOARDING.form.baseUrl||provider.default_base_url||'';
+    }else{
+      ONBOARDING.form.baseUrl=provider.default_base_url||'';
+    }
+  }
+  _renderOnboardingBody();
+}
+
+async function loadOnboardingWizard(){
+  try{
+    const status=await api('/api/onboarding/status');
+    ONBOARDING.status=status;
+    const current=((status.setup||{}).current)||{};
+    ONBOARDING.form.provider=current.provider||'openrouter';
+    ONBOARDING.form.workspace=(status.workspaces&&status.workspaces.last)||status.settings.default_workspace||'';
+    ONBOARDING.form.model=status.settings.default_model||current.model||'';
+    ONBOARDING.form.password='';
+    ONBOARDING.form.apiKey='';
+    ONBOARDING.form.baseUrl=current.base_url||'';
+    ONBOARDING.active=!status.completed;
+    if(!ONBOARDING.active) return false;
+    $('onboardingOverlay').style.display='flex';
+    _renderOnboardingSteps();
+    _renderOnboardingBody();
+    return true;
+  }catch(e){
+    console.warn('onboarding status failed',e);
+    return false;
+  }
+}
+
+function prevOnboardingStep(){
+  if(ONBOARDING.step===0)return;
+  ONBOARDING.step--;
+  _renderOnboardingSteps();
+  _renderOnboardingBody();
+}
+
+async function _saveOnboardingProviderSetup(){
+  const provider=(ONBOARDING.form.provider||'').trim();
+  const model=(ONBOARDING.form.model||'').trim();
+  const apiKey=(ONBOARDING.form.apiKey||'').trim();
+  const baseUrl=(ONBOARDING.form.baseUrl||'').trim();
+  const current=_getOnboardingCurrentSetup();
+  const isUnchanged=current.provider===provider&&((current.model||'')===model)&&((current.base_url||'')===baseUrl);
+  // Skip the POST when nothing changed.  We also skip when the provider is
+  // unsupported/OAuth-based and already working — chat_ready may be false for
+  // providers not in the quick-setup list (e.g. minimax-cn) even though they are
+  // fully configured.  Posting in that case would either be a no-op (the server
+  // just marks complete for unsupported providers) or could silently overwrite
+  // config.yaml if the user accidentally changed the provider dropdown.
+  const currentIsOauth=!!(ONBOARDING.status&&ONBOARDING.status.setup&&ONBOARDING.status.setup.current_is_oauth);
+  if(isUnchanged && !apiKey && ((ONBOARDING.status.system||{}).chat_ready || currentIsOauth)) return;
+  const body={provider,model};
+  if(apiKey) body.api_key=apiKey;
+  if(baseUrl) body.base_url=baseUrl;
+  const status=await api('/api/onboarding/setup',{method:'POST',body:JSON.stringify(body)});
+  ONBOARDING.status=status;
+}
+
+async function _saveOnboardingDefaults(){
+  const workspace=(ONBOARDING.form.workspace||'').trim();
+  const model=(ONBOARDING.form.model||'').trim();
+  const password=(ONBOARDING.form.password||'').trim();
+  if(!workspace) throw new Error(t('onboarding_error_choose_workspace'));
+  if(!model) throw new Error(t('onboarding_error_choose_model'));
+  const known=_getOnboardingWorkspaceChoices().some(ws=>ws.path===workspace);
+  if(!known){
+    await api('/api/workspaces/add',{method:'POST',body:JSON.stringify({path:workspace})});
+  }
+  // Model persisted by /api/onboarding/setup — no /api/default-model call needed here
+  const body={default_workspace:workspace};
+  if(password) body._set_password=password;
+  const saved=await api('/api/settings',{method:'POST',body:JSON.stringify(body)});
+  if(ONBOARDING.status){
+    ONBOARDING.status.settings={...(ONBOARDING.status.settings||{}),password_enabled:!!saved.auth_enabled};
+  }
+  localStorage.setItem('hermes-webui-model',model);
+  if($('modelSelect')) _applyModelToDropdown(model,$('modelSelect'));
+}
+
+async function _finishOnboarding(){
+  await _saveOnboardingProviderSetup();
+  await _saveOnboardingDefaults();
+  const done=await api('/api/onboarding/complete',{method:'POST',body:'{}'});
+  ONBOARDING.status=done;
+  ONBOARDING.active=false;
+  $('onboardingOverlay').style.display='none';
+  showToast(t('onboarding_complete'));
+  await loadWorkspaceList();
+  if(typeof renderSessionList==='function') await renderSessionList();
+  if(!S.session && typeof newSession==='function'){
+    await newSession(true);
+    await renderSessionList();
+  }
+}
+
+async function skipOnboarding(){
+  try{
+    // Mark onboarding completed server-side without changing any config
+    await api('/api/onboarding/complete',{method:'POST',body:'{}'});
+    ONBOARDING.active=false;
+    $('onboardingOverlay').style.display='none';
+    showToast(t('onboarding_skipped')||'Setup skipped');
+  }catch(e){
+    _setOnboardingNotice((e.message||String(e)),'warn');
+  }
+}
+
+async function nextOnboardingStep(){
+  try{
+    if(ONBOARDING.steps[ONBOARDING.step]==='setup'){
+      ONBOARDING.form.provider=(($('onboardingProviderSelect')||{}).value||ONBOARDING.form.provider||'').trim();
+      ONBOARDING.form.apiKey=(($('onboardingApiKeyInput')||{}).value||'').trim();
+      ONBOARDING.form.baseUrl=(($('onboardingBaseUrlInput')||{}).value||ONBOARDING.form.baseUrl||'').trim();
+      if(!ONBOARDING.form.provider) throw new Error(t('onboarding_error_provider_required'));
+      if(ONBOARDING.form.provider==='custom' && !ONBOARDING.form.baseUrl) throw new Error(t('onboarding_error_base_url_required'));
+    }
+    if(ONBOARDING.steps[ONBOARDING.step]==='workspace'){
+      ONBOARDING.form.workspace=(($('onboardingWorkspaceInput')||{}).value||ONBOARDING.form.workspace||'').trim();
+      ONBOARDING.form.model=(($('onboardingModelInput')||{}).value||($('onboardingModelSelect')||{}).value||ONBOARDING.form.model||'').trim();
+      if(!ONBOARDING.form.workspace) throw new Error(t('onboarding_error_workspace_required'));
+      if(!ONBOARDING.form.model) throw new Error(t('onboarding_error_model_required'));
+    }
+    if(ONBOARDING.steps[ONBOARDING.step]==='password'){
+      ONBOARDING.form.password=(($('onboardingPasswordInput')||{}).value||'').trim();
+    }
+    if(ONBOARDING.step===ONBOARDING.steps.length-1){
+      await _finishOnboarding();
+      return;
+    }
+    ONBOARDING.step++;
+    _renderOnboardingSteps();
+    _renderOnboardingBody();
+  }catch(e){
+    _setOnboardingNotice(e.message||String(e),'warn');
+  }
+}

static/sw.js

@@ -0,0 +1,106 @@
+/**
+ * Hermes WebUI Service Worker
+ * Minimal PWA service worker — enables "Add to Home Screen".
+ * No offline caching of API responses (the UI requires a live backend).
+ * Caches only static shell assets so the app shell loads fast on repeat visits.
+ */
+
+// Cache version is injected by the server at request time (routes.py /sw.js handler).
+// Bumps automatically whenever the git commit changes — no manual edits needed.
+const CACHE_NAME = 'hermes-shell-__CACHE_VERSION__';
+
+// Static assets that form the app shell
+const SHELL_ASSETS = [
+  './',
+  './static/style.css',
+  './static/boot.js',
+  './static/ui.js',
+  './static/messages.js',
+  './static/sessions.js',
+  './static/panels.js',
+  './static/commands.js',
+  './static/icons.js',
+  './static/i18n.js',
+  './static/workspace.js',
+  './static/onboarding.js',
+  './static/favicon.svg',
+  './static/favicon-32.png',
+  './manifest.json',
+];
+
+// Install: pre-cache the app shell
+self.addEventListener('install', (event) => {
+  event.waitUntil(
+    caches.open(CACHE_NAME).then((cache) => {
+      return cache.addAll(SHELL_ASSETS).catch((err) => {
+        // Non-fatal: if any asset fails, still activate
+        console.warn('[sw] Shell pre-cache partial failure:', err);
+      });
+    })
+  );
+  self.skipWaiting();
+});
+
+// Activate: clean up old caches
+self.addEventListener('activate', (event) => {
+  event.waitUntil(
+    caches.keys().then((keys) =>
+      Promise.all(
+        keys.filter((k) => k !== CACHE_NAME).map((k) => caches.delete(k))
+      )
+    )
+  );
+  self.clients.claim();
+});
+
+// Fetch strategy:
+// - API calls (/api/*, /stream) → always network (never cache)
+// - Shell assets → cache-first with network fallback
+// - Everything else → network-first, fall back to offline page
+self.addEventListener('fetch', (event) => {
+  const url = new URL(event.request.url);
+
+  // Never intercept cross-origin requests
+  if (url.origin !== self.location.origin) return;
+
+  // API and streaming endpoints — always go to network
+  if (
+    url.pathname.startsWith('/api/') ||
+    url.pathname.includes('/stream') ||
+    url.pathname.startsWith('/health')
+  ) {
+    return; // let browser handle normally
+  }
+
+  // Shell assets: cache-first
+  event.respondWith(
+    caches.match(event.request).then((cached) => {
+      if (cached) return cached;
+      return fetch(event.request).then((response) => {
+        // Cache successful GET responses for shell assets
+        if (
+          event.request.method === 'GET' &&
+          response.status === 200
+        ) {
+          const clone = response.clone();
+          caches.open(CACHE_NAME).then((cache) => cache.put(event.request, clone));
+        }
+        return response;
+      }).catch(() => {
+        // Offline fallback for navigation requests.
+        // Note: caches.match() returns a Promise (always truthy in a `||` check),
+        // so we must await/then to unwrap it — otherwise the `new Response(...)`
+        // branch is dead code and the browser falls back to its default offline page.
+        if (event.request.mode === 'navigate') {
+          return caches.match('./').then((cached) => cached || new Response(
+            '<html><body style="font-family:sans-serif;padding:2rem;background:#1a1a1a;color:#ccc">' +
+            '<h2>You are offline</h2>' +
+            '<p>Hermes requires a server connection. Please check your network and try again.</p>' +
+            '</body></html>',
+            { headers: { 'Content-Type': 'text/html' } }
+          ));
+        }
+      });
+    })
+  );
+});

static/vendor/smd.min.js

@@ -0,0 +1,29 @@
+var D=2,C=3,h=4,b=5,B=6,U=7,G=8,S=9,x=10,m=11,H=12,K=13,M=14,Q=15,w=16,q=17,W=18,P=19,Y=20,y=21,F=22,$=23,v=24,X=25,j=26,z=27,J=28,V=29,Z=30,p=31;var I=1,k=2,L=4,T=8,f=16;function ee(e){switch(e){case I:return"href";case k:return"src";case L:return"class";case T:return"checked";case f:return"start"}}var ne=e=>{switch(e){case 1:return 3;case 2:return 4;case 3:return 5;case 4:return 6;case 5:return 7;default:return 8}},te=ne;var O=24;function ae(e){let c=new Uint32Array(O);return c[0]=1,{renderer:e,text:"",pending:"",tokens:c,len:0,token:1,fence_end:0,blockquote_idx:0,hr_char:"",hr_chars:0,fence_start:0,spaces:new Uint8Array(O),indent:"",indent_len:0,table_state:0}}function ce(e){e.pending.length>0&&o(e,`
+`)}function a(e){e.text.length!==0&&(e.renderer.add_text(e.renderer.data,e.text),e.text="")}function _(e){e.len-=1,e.token=e.tokens[e.len],e.renderer.end_token(e.renderer.data)}function i(e,c){(e.tokens[e.len]===24||e.tokens[e.len]===23)&&c!==25&&_(e),e.len+=1,e.tokens[e.len]=c,e.token=c,e.renderer.add_token(e.renderer.data,c)}function re(e,c,n){for(;n<=e.len;){if(e.tokens[n]===c)return n;n+=1}return-1}function l(e,c){for(e.fence_start=0;e.len>c;)_(e)}function u(e,c){let n=0;for(let t=0;t<=e.len&&(c-=e.spaces[t],!(c<0));t+=1)switch(e.tokens[t]){case 9:case 10:case 20:case 25:n=t;break}for(;e.len>n;)_(e);return c}function A(e,c){let n=-1,t=-1;for(let s=e.blockquote_idx+1;s<=e.len;s+=1)if(e.tokens[s]===25){if(e.indent_len<e.spaces[s]){t=-1;break}t=s}else e.tokens[s]===c&&(n=s);return t===-1?n===-1?(l(e,e.blockquote_idx),i(e,c),!0):(l(e,n),!1):(l(e,t),i(e,c),!0)}function g(e,c){i(e,25),e.spaces[e.len]=e.indent_len+c,E(e),e.token=103}function E(e){e.indent="",e.indent_len=0,e.pending=""}function N(e){switch(e){case 48:case 49:case 50:case 51:case 52:case 53:case 54:case 55:case 56:case 57:return!0;default:return!1}}function ie(e){switch(e){case 32:case 58:case 59:case 41:case 44:case 33:case 46:case 63:case 93:case 10:return!0;default:return!1}}function se(e){return N(e)||ie(e)}function o(e,c){for(let n of c){if(e.token===101){switch(n){case" ":e.indent_len+=1;continue;case"	":e.indent_len+=4;continue}let s=u(e,e.indent_len);e.indent_len=0,e.token=e.tokens[e.len],s>0&&o(e," ".repeat(s))}let t=e.pending+n;switch(e.token){case 21:case 1:case 20:case 24:case 23:switch(e.pending[0]){case void 0:e.pending=n;continue;case" ":e.pending=n,e.indent+=" ",e.indent_len+=1;continue;case"	":e.pending=n,e.indent+="	",e.indent_len+=4;continue;case`
+`:if(e.tokens[e.len]===25&&e.token===21){_(e),E(e),e.pending=n;continue}l(e,e.blockquote_idx),E(e),e.blockquote_idx=0,e.fence_start=0,e.pending=n;continue;case"#":switch(n){case"#":if(e.pending.length<6){e.pending=t;continue}break;case" ":u(e,e.indent_len),i(e,te(e.pending.length)),E(e);continue}break;case">":{let r=re(e,20,e.blockquote_idx+1);r===-1?(l(e,e.blockquote_idx),e.blockquote_idx+=1,e.fence_start=0,i(e,20)):e.blockquote_idx=r,E(e),e.pending=n;continue}case"-":case"*":case"_":if(e.hr_chars===0&&(e.hr_chars=1,e.hr_char=e.pending),e.hr_chars>0){switch(n){case e.hr_char:e.hr_chars+=1,e.pending=t;continue;case" ":e.pending=t;continue;case`
+`:if(e.hr_chars<3)break;u(e,e.indent_len),e.renderer.add_token(e.renderer.data,22),e.renderer.end_token(e.renderer.data),E(e),e.hr_chars=0;continue}e.hr_chars=0}if(e.pending[0]!=="_"&&e.pending[1]===" "){A(e,23),g(e,2),o(e,t.slice(2));continue}break;case"`":if(e.pending.length<3){if(n==="`"){e.pending=t,e.fence_start=t.length;continue}e.fence_start=0;break}switch(n){case"`":e.pending.length===e.fence_start?(e.pending=t,e.fence_start=t.length):(i(e,2),E(e),e.fence_start=0,o(e,t));continue;case`
+`:{u(e,e.indent_len),i(e,10),e.pending.length>e.fence_start&&e.renderer.set_attr(e.renderer.data,L,e.pending.slice(e.fence_start)),E(e),e.token=101;continue}default:e.pending=t;continue}case"+":if(n!==" ")break;A(e,23),g(e,2);continue;case"0":case"1":case"2":case"3":case"4":case"5":case"6":case"7":case"8":case"9":if(e.pending[e.pending.length-1]==="."){if(n!==" ")break;A(e,24)&&e.pending!=="1."&&e.renderer.set_attr(e.renderer.data,f,e.pending.slice(0,-1)),g(e,e.pending.length+1);continue}else{let r=n.charCodeAt(0);if(r===46||N(r)){e.pending=t;continue}}break;case"|":l(e,e.blockquote_idx),i(e,27),i(e,28),e.pending="",o(e,n);continue}let s=t;if(e.token===21)e.token=e.tokens[e.len],e.renderer.add_token(e.renderer.data,21),e.renderer.end_token(e.renderer.data);else if(e.indent_len>=4){let r=0;for(;r<4;r+=1)if(e.indent[r]==="	"){r=r+1;break}s=e.indent.slice(r)+t,i(e,9)}else i(e,2);E(e),o(e,s);continue;case 27:if(e.table_state===1)switch(n){case"-":case" ":case"|":case":":e.pending=t;continue;case`
+`:e.table_state=2,e.pending="";continue;default:_(e),e.table_state=0;break}else switch(e.pending){case"|":i(e,28),e.pending="",o(e,n);continue;case`
+`:_(e),e.pending="",e.table_state=0,o(e,n);continue}break;case 28:switch(e.pending){case"":break;case"|":i(e,29),_(e),e.pending="",o(e,n);continue;case`
+`:_(e),e.table_state=Math.min(e.table_state+1,2),e.pending="",o(e,n);continue;default:i(e,29),o(e,n);continue}break;case 29:if(e.pending==="|"){a(e),_(e),e.pending="",o(e,n);continue}break;case 9:switch(t){case`
+    `:case`
+   	`:case`
+  	`:case`
+ 	`:case`
+	`:e.text+=`
+`,e.pending="";continue;case`
+`:case`
+ `:case`
+  `:case`
+   `:e.pending=t;continue;default:e.pending.length!==0?(a(e),_(e),e.pending=n):e.text+=n;continue}case 10:switch(n){case"`":e.pending=t;continue;case`
+`:if(t.length===e.fence_start+e.fence_end+1){a(e),_(e),e.pending="",e.fence_start=0,e.fence_end=0,e.token=101;continue}e.token=101;break;case" ":if(e.pending[0]===`
+`){e.pending=t,e.fence_end+=1;continue}break}e.text+=e.pending,e.pending=n,e.fence_end=1;continue;case 11:switch(n){case"`":t.length===e.fence_start+ +(e.pending[0]===" ")?(a(e),_(e),e.pending="",e.fence_start=0):e.pending=t;continue;case`
+`:e.text+=e.pending,e.pending="",e.token=21,e.blockquote_idx=0,a(e);continue;case" ":e.text+=e.pending,e.pending=n;continue;default:e.text+=t,e.pending="";continue}case 103:switch(e.pending.length){case 0:if(n!=="[")break;e.pending=t;continue;case 1:if(n!==" "&&n!=="x")break;e.pending=t;continue;case 2:if(n!=="]")break;e.pending=t;continue;case 3:if(n!==" ")break;e.renderer.add_token(e.renderer.data,26),e.pending[1]==="x"&&e.renderer.set_attr(e.renderer.data,T,""),e.renderer.end_token(e.renderer.data),e.pending=" ";continue}e.token=e.tokens[e.len],e.pending="",o(e,t);continue;case 14:case 15:{let r="*",d=12;if(e.token===15&&(r="_",d=13),r===e.pending){if(a(e),r===n){_(e),e.pending="";continue}i(e,d),e.pending=n;continue}break}case 12:case 13:{let r="*",d=14;switch(e.token===13&&(r="_",d=15),e.pending){case r:r===n?e.tokens[e.len-1]===d?e.pending=t:(a(e),i(e,d),e.pending=""):(a(e),_(e),e.pending=n);continue;case r+r:let R=e.token;a(e),_(e),_(e),r!==n?(i(e,R),e.pending=n):e.pending="";continue}break}case 16:if(t==="~~"){a(e),_(e),e.pending="";continue}break;case 105:n===`
+`?(a(e),i(e,30),e.pending=""):(e.token=e.tokens[e.len],e.pending[0]==="\\"?e.text+="[":e.text+="$$",e.pending="",o(e,n));continue;case 30:if(t==="\\]"||t==="$$"){a(e),_(e),e.pending="";continue}break;case 31:if(t==="\\)"||e.pending[0]==="$"){a(e),_(e),n===")"?e.pending="":e.pending=n;continue}break;case 102:t==="http://"||t==="https://"?(a(e),i(e,18),e.pending=t,e.text=t):"http:/"[e.pending.length]===n||"https:/"[e.pending.length]===n?e.pending=t:(e.token=e.tokens[e.len],o(e,n));continue;case 17:case 19:if(e.pending==="]"){a(e),n==="("?e.pending=t:(_(e),e.pending=n);continue}if(e.pending[0]==="]"&&e.pending[1]==="("){if(n===")"){let r=e.token===17?I:k,d=e.pending.slice(2);e.renderer.set_attr(e.renderer.data,r,d),_(e),e.pending=""}else e.pending+=n;continue}break;case 18:n===" "||n===`
+`||n==="\\"?(e.renderer.set_attr(e.renderer.data,I,e.pending),a(e),_(e),e.pending=n):(e.text+=n,e.pending=t);continue;case 104:if(t.startsWith("<br")){if(t.length===3||n===" "||n==="/"&&(t.length===4||e.pending[e.pending.length-1]===" ")){e.pending=t;continue}if(n===">"){a(e),e.token=e.tokens[e.len],e.renderer.add_token(e.renderer.data,21),e.renderer.end_token(e.renderer.data),e.pending="";continue}}e.token=e.tokens[e.len],e.text+="<",e.pending=e.pending.slice(1),o(e,n);continue}switch(e.pending[0]){case"\\":if(e.token===19||e.token===30||e.token===31)break;switch(n){case"(":a(e),i(e,31),e.pending="";continue;case"[":e.token=105,e.pending=t;continue;case`
+`:e.pending=n;continue;default:let s=n.charCodeAt(0);e.pending="",e.text+=N(s)||s>=65&&s<=90||s>=97&&s<=122?t:n;continue}case`
+`:switch(e.token){case 19:case 30:case 31:break;case 3:case 4:case 5:case 6:case 7:case 8:a(e),l(e,e.blockquote_idx),e.blockquote_idx=0,e.pending=n;continue;default:a(e),e.pending=n,e.token=21,e.blockquote_idx=0;continue}break;case"<":if(e.token!==19&&e.token!==30&&e.token!==31){a(e),e.pending=t,e.token=104;continue}break;case"`":if(e.token===19)break;n==="`"?(e.fence_start+=1,e.pending=t):(e.fence_start+=1,a(e),i(e,11),e.text=n===" "||n===`
+`?"":n,e.pending="");continue;case"_":case"*":{if(e.token===19||e.token===30||e.token===31||e.token===14)break;let s=12,r=14,d=e.pending[0];if(d==="_"&&(s=13,r=15),e.pending.length===1){if(d===n){e.pending=t;continue}if(n!==" "&&n!==`
+`){a(e),i(e,s),e.pending=n;continue}}else{if(d===n){a(e),i(e,r),i(e,s),e.pending="";continue}if(n!==" "&&n!==`
+`){a(e),i(e,r),e.pending=n;continue}}break}case"~":if(e.token!==19&&e.token!==16){if(e.pending==="~"){if(n==="~"){e.pending=t;continue}}else if(n!==" "&&n!==`
+`){a(e),i(e,16),e.pending=n;continue}}break;case"$":if(e.token!==19&&e.token!==16&&e.pending==="$")if(n==="$"){e.token=105,e.pending=t;continue}else{if(se(n.charCodeAt(0)))break;a(e),i(e,31),e.pending=n;continue}break;case"[":if(e.token!==19&&e.token!==17&&e.token!==30&&e.token!==31&&n!=="]"){a(e),i(e,17),e.pending=n;continue}break;case"!":if(e.token!==19&&n==="["){a(e),i(e,19),e.pending="";continue}break;case" ":if(e.pending.length===1&&n===" ")continue;break}if(e.token!==19&&e.token!==17&&e.token!==30&&e.token!==31&&n==="h"&&(e.pending===" "||e.pending==="")){e.text+=e.pending,e.pending=n,e.token=102;continue}e.text+=e.pending,e.pending=n}a(e)}function _e(e){return{add_token:oe,end_token:de,add_text:Ee,set_attr:le,data:{nodes:[e,,,,,],index:0}}}function oe(e,c){let n=e.nodes[e.index],t;switch(c){case 1:return;case 20:t=document.createElement("blockquote");break;case 2:t=document.createElement("p");break;case 21:t=document.createElement("br");break;case 22:t=document.createElement("hr");break;case 3:t=document.createElement("h1");break;case 4:t=document.createElement("h2");break;case 5:t=document.createElement("h3");break;case 6:t=document.createElement("h4");break;case 7:t=document.createElement("h5");break;case 8:t=document.createElement("h6");break;case 12:case 13:t=document.createElement("em");break;case 14:case 15:t=document.createElement("strong");break;case 16:t=document.createElement("s");break;case 11:t=document.createElement("code");break;case 18:case 17:t=document.createElement("a");break;case 19:t=document.createElement("img");break;case 23:t=document.createElement("ul");break;case 24:t=document.createElement("ol");break;case 25:t=document.createElement("li");break;case 26:let s=t=document.createElement("input");s.type="checkbox",s.disabled=!0;break;case 9:case 10:n=n.appendChild(document.createElement("pre")),t=document.createElement("code");break;case 27:t=document.createElement("table");break;case 28:switch(n.children.length){case 0:n=n.appendChild(document.createElement("thead"));break;case 1:n=n.appendChild(document.createElement("tbody"));break;default:n=n.children[1]}t=document.createElement("tr");break;case 29:t=document.createElement(n.parentElement?.tagName==="THEAD"?"th":"td");break;case 30:t=document.createElement("equation-block");break;case 31:t=document.createElement("equation-inline");break}e.nodes[++e.index]=n.appendChild(t)}function de(e){e.index-=1}function Ee(e,c){e.nodes[e.index].appendChild(document.createTextNode(c))}function le(e,c,n){e.nodes[e.index].setAttribute(ee(c),n)}export{Y as BLOCKQUOTE,j as CHECKBOX,T as CHECKED,S as CODE_BLOCK,x as CODE_FENCE,m as CODE_INLINE,Z as EQUATION_BLOCK,p as EQUATION_INLINE,C as HEADING_1,h as HEADING_2,b as HEADING_3,B as HEADING_4,U as HEADING_5,G as HEADING_6,I as HREF,P as IMAGE,H as ITALIC_AST,K as ITALIC_UND,L as LANG,y as LINE_BREAK,q as LINK,X as LIST_ITEM,v as LIST_ORDERED,$ as LIST_UNORDERED,D as PARAGRAPH,W as RAW_URL,F as RULE,k as SRC,f as START,w as STRIKE,M as STRONG_AST,Q as STRONG_UND,z as TABLE,V as TABLE_CELL,J as TABLE_ROW,_e as default_renderer,ae as parser,ce as parser_end,o as parser_write};

static/workspace.js

@@ -1,7 +1,13 @@
 async function api(path,opts={}){
-  const url=new URL(path,location.origin);
+  // Strip leading slash so URL resolves relative to location.href (supports subpath mounts)
+  const rel = path.startsWith('/') ? path.slice(1) : path;
+  const url=new URL(rel,location.href);
   const res=await fetch(url.href,{credentials:'include',headers:{'Content-Type':'application/json'},...opts});
   if(!res.ok){
+    // 401 means the auth session expired. Redirect to /login so the user can
+    // re-authenticate. This is especially important for iOS PWA (standalone mode)
+    // where a server-side 302 → /login opens in Safari instead of within the PWA.
+    if(res.status===401){window.location.href='/login?next='+encodeURIComponent(window.location.pathname+window.location.search);return;}
     const text=await res.text();
     // Parse JSON error body and surface the human-readable message,
     // rather than showing raw JSON like {"error":"Profile 'x' does not exist."}
@@ -54,7 +60,7 @@ async function loadDir(path){
     }
     if(typeof clearPreview==='function'){
       if(typeof _previewDirty!=='undefined'&&_previewDirty){
-        if(confirm('You have unsaved changes in the preview. Discard and navigate?'))clearPreview();
+        showConfirmDialog({title:t('unsaved_confirm'),message:'',confirmLabel:'Discard',danger:true,focusCancel:true}).then(ok=>{if(ok)clearPreview();});
       }else{
         clearPreview();
       }
@@ -95,6 +101,7 @@ function navigateUp(){
 // File extension sets for preview routing (must match server-side sets)
 const IMAGE_EXTS  = new Set(['.png','.jpg','.jpeg','.gif','.svg','.webp','.ico','.bmp']);
 const MD_EXTS     = new Set(['.md','.markdown','.mdown']);
+const HTML_EXTS   = new Set(['.html','.htm']);
 // Binary formats that should download rather than preview
 const DOWNLOAD_EXTS = new Set([
   '.docx','.doc','.xlsx','.xls','.pptx','.ppt','.odt','.ods','.odp',
@@ -108,21 +115,25 @@ const DOWNLOAD_EXTS = new Set([
 function fileExt(p){ const i=p.lastIndexOf('.'); return i>=0?p.slice(i).toLowerCase():''; }
 
 let _previewCurrentPath = '';  // relative path of currently previewed file
-let _previewCurrentMode = '';  // 'code' | 'md' | 'image'
+let _previewCurrentMode = '';  // 'code' | 'md' | 'image' | 'html'
 let _previewDirty = false;     // true when edits are unsaved
 
 function showPreview(mode){
-  // mode: 'code' | 'image' | 'md'
+  // mode: 'code' | 'image' | 'md' | 'html'
   $('previewCode').style.display     = mode==='code'  ? '' : 'none';
   $('previewImgWrap').style.display  = mode==='image' ? '' : 'none';
   $('previewMd').style.display       = mode==='md'    ? '' : 'none';
+  $('previewHtmlWrap').style.display = mode==='html'  ? '' : 'none';
   $('previewEditArea').style.display = 'none';  // start in read-only
   const badge=$('previewBadge');
   badge.className='preview-badge '+mode;
-  badge.textContent = mode==='image'?'image':mode==='md'?'md':fileExt($('previewPathText').textContent)||'text';
+  badge.textContent = mode==='image'?'image':mode==='md'?'md':mode==='html'?'html':fileExt($('previewPathText').textContent)||'text';
   _previewCurrentMode = mode;
   _previewDirty = false;
   updateEditBtn();
+  // Show "Open in browser" button only for HTML mode
+  const openBtn=$('btnOpenInBrowser');
+  if(openBtn) openBtn.style.display = mode==='html'?'inline-flex':'none';
 }
 
 function updateEditBtn(){
@@ -131,8 +142,8 @@ function updateEditBtn(){
   const editable = _previewCurrentMode==='code'||_previewCurrentMode==='md';
   btn.style.display = editable?'':'none';
   const editing = $('previewEditArea').style.display!=='none';
-  btn.innerHTML = editing ? '💾 Save' : '✎ Edit';
-  btn.title = editing ? 'Save changes' : 'Edit this file';
+  btn.innerHTML = editing ? `💾 ${t('save')}` : `✎ ${t('edit')}`;
+  btn.title = editing ? t('save_title') : t('edit_title');
   btn.style.color = editing ? 'var(--blue)' : '';
   if(_previewDirty) btn.innerHTML = '💾 Save*';
 }
@@ -150,12 +161,12 @@ async function toggleEditMode(){
       _previewDirty=false;
       // Update read-only views
       if(_previewCurrentMode==='code') $('previewCode').textContent=content;
-      else $('previewMd').innerHTML=renderMd(content);
+      else { $('previewMd').innerHTML=renderMd(content); requestAnimationFrame(()=>{if(typeof renderKatexBlocks==='function')renderKatexBlocks();}); }
       $('previewEditArea').style.display='none';
       if(_previewCurrentMode==='code') $('previewCode').style.display='';
       else $('previewMd').style.display='';
-      showToast('Saved');
-    }catch(e){setStatus('Save failed: '+e.message);}
+      showToast(t('saved'));
+    }catch(e){setStatus(t('save_failed')+e.message);}
   }else{
     // Enter edit mode: populate textarea with current content
     const currentText = _previewCurrentMode==='code'
@@ -200,13 +211,14 @@ async function openFile(path){
   $('fileTree').style.display='none';
 
   _previewCurrentPath = path;
+  renderFileBreadcrumb(path);
   if(IMAGE_EXTS.has(ext)){
     // Image: load via raw endpoint, show as <img>
     showPreview('image');
-    const url=`/api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(path)}`;
+    const url=`api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(path)}`;
     $('previewImg').alt=path;
     $('previewImg').src=url;
-    $('previewImg').onerror=()=>setStatus('Could not load image');
+    $('previewImg').onerror=()=>setStatus(t('image_load_failed'));
   } else if(MD_EXTS.has(ext)){
     // Markdown: fetch text, render with renderMd, display as formatted HTML
     try{
@@ -214,7 +226,24 @@ async function openFile(path){
       showPreview('md');
       _previewRawContent = data.content;
       $('previewMd').innerHTML=renderMd(data.content);
-    }catch(e){setStatus('Could not open file');}
+      requestAnimationFrame(()=>{if(typeof renderKatexBlocks==='function')renderKatexBlocks();});
+    }catch(e){setStatus(t('file_open_failed'));}
+  } else if(HTML_EXTS.has(ext)){
+    // HTML: render in sandboxed iframe via raw endpoint.
+    // SECURITY TRADEOFF: We use sandbox="allow-scripts" which lets inline JS run
+    // but prevents access to the parent frame (origin isolation). This is a
+    // deliberate choice — the user is previewing their own workspace files, so
+    // blocking scripts entirely would break most HTML documents. The sandbox
+    // still prevents the preview from navigating the parent, accessing cookies,
+    // or reading other origin data. If a stricter mode is needed, remove
+    // allow-scripts (or add sandbox="") to disable all JS execution.
+    showPreview('html');
+    const url=`api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(path)}&inline=1`;
+    const iframe=$('previewHtmlIframe');
+    if(iframe){
+      iframe.src=''; // clear first to avoid stale content
+      iframe.src=url;
+    }
   } else {
     // Plain code / text -- but fall back to download if server signals binary
     try{
@@ -236,12 +265,56 @@ async function openFile(path){
 function downloadFile(path){
   if(!S.session)return;
   // Trigger browser download via the raw file endpoint with content-disposition attachment
-  const url=`/api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(path)}&download=1`;
+  const url=`api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(path)}&download=1`;
   const filename=path.split('/').pop();
   const a=document.createElement('a');
   a.href=url;a.download=filename;
   document.body.appendChild(a);a.click();
   setTimeout(()=>document.body.removeChild(a),100);
-  showToast(`Downloading ${filename}\u2026`,2000);
+  showToast(t('downloading',filename),2000);
 }
 
+
+// ── Render breadcrumb for file preview mode ──────────────────────────────────
+function renderFileBreadcrumb(filePath) {
+  const bar = $('breadcrumbBar');
+  if (!bar) return;
+  bar.style.display = 'flex';
+  const upBtn = $('btnUpDir');
+  if (upBtn) upBtn.style.display = '';
+
+  bar.innerHTML = '';
+  // Root
+  const root = document.createElement('span');
+  root.className = 'breadcrumb-seg breadcrumb-link';
+  root.textContent = '~';
+  root.onclick = () => { clearPreview(); loadDir('.'); };
+  bar.appendChild(root);
+
+  const parts = filePath.split('/');
+  let accumulated = '';
+  for (let i = 0; i < parts.length; i++) {
+    const sep = document.createElement('span');
+    sep.className = 'breadcrumb-sep';
+    sep.textContent = '/';
+    bar.appendChild(sep);
+
+    accumulated += (accumulated ? '/' : '') + parts[i];
+    const seg = document.createElement('span');
+    seg.textContent = parts[i];
+    if (i < parts.length - 1) {
+      seg.className = 'breadcrumb-seg breadcrumb-link';
+      const target = accumulated;
+      seg.onclick = () => { clearPreview(); loadDir(target); };
+    } else {
+      seg.className = 'breadcrumb-seg breadcrumb-current';
+    }
+    bar.appendChild(seg);
+  }
+}
+
+function openInBrowser(){
+  if(!_previewCurrentPath||!S.session) return;
+  const url=`api/file/raw?session_id=${encodeURIComponent(S.session.session_id)}&path=${encodeURIComponent(_previewCurrentPath)}`;
+  window.open(url,'_blank');
+}

tests/_pytest_port.py

@@ -0,0 +1,46 @@
+"""
+Shared test server constants for use in individual test files.
+
+Instead of hardcoding ``BASE = "http://127.0.0.1:8788"`` in every test file,
+import from here so the port and state dir are always consistent with
+what conftest.py computed for this worktree.
+
+Usage::
+
+    from tests._pytest_port import BASE
+
+conftest.py publishes ``HERMES_WEBUI_TEST_PORT`` and
+``HERMES_WEBUI_TEST_STATE_DIR`` to ``os.environ`` at module level
+(before any test file is imported), so this module always reads the
+correct values.  The auto-derivation fallback matches conftest's logic
+exactly, so standalone imports also work correctly.
+"""
+import hashlib
+import os
+import pathlib
+
+def _auto_test_port(repo_root: pathlib.Path) -> int:
+    h = int(hashlib.md5(str(repo_root).encode()).hexdigest(), 16)
+    return 20000 + (h % 10000)
+
+def _auto_state_dir_name(repo_root: pathlib.Path) -> str:
+    h = hashlib.md5(str(repo_root).encode()).hexdigest()[:8]
+    return f"webui-test-{h}"
+
+_TESTS_DIR   = pathlib.Path(__file__).parent.resolve()
+_REPO_ROOT   = _TESTS_DIR.parent.resolve()
+_HERMES_HOME = pathlib.Path(os.getenv('HERMES_HOME',
+                             str(pathlib.Path.home() / '.hermes')))
+
+TEST_PORT = int(os.environ.get('HERMES_WEBUI_TEST_PORT',
+                               str(_auto_test_port(_REPO_ROOT))))
+BASE = f"http://127.0.0.1:{TEST_PORT}"
+
+TEST_STATE_DIR = pathlib.Path(os.environ.get(
+    'HERMES_WEBUI_TEST_STATE_DIR',
+    str(_HERMES_HOME / _auto_state_dir_name(_REPO_ROOT))
+))
+
+# Default model injected by conftest — tests that mutate the default model
+# must restore to this value so later tests see a consistent baseline.
+TEST_DEFAULT_MODEL = os.environ.get('HERMES_WEBUI_DEFAULT_MODEL', 'openai/gpt-5.4-mini')

tests/conftest.py

@@ -31,14 +31,37 @@ HOME       = pathlib.Path.home()
 HERMES_HOME = pathlib.Path(os.getenv('HERMES_HOME', str(HOME / '.hermes')))
 
 # ── Test server config ────────────────────────────────────────────────────
-TEST_PORT      = int(os.getenv('HERMES_WEBUI_TEST_PORT', '8788'))
+# Port and state dir auto-derive from the repo path when no env var is set,
+# giving every worktree its own isolated port (8800-8899) and state directory.
+# Override with HERMES_WEBUI_TEST_PORT / HERMES_WEBUI_TEST_STATE_DIR to pin.
+
+def _auto_test_port(repo_root) -> int:
+    """Map repo path to a unique port in 20000-29999 (10k range = near-zero collisions).
+    Far from system port ranges and Linux ephemeral ports (32768+).
+    Override with HERMES_WEBUI_TEST_PORT to use a specific port."""
+    import hashlib
+    h = int(hashlib.md5(str(repo_root).encode()).hexdigest(), 16)
+    return 20000 + (h % 10000)
+
+def _auto_state_dir_name(repo_root) -> str:
+    import hashlib
+    h = hashlib.md5(str(repo_root).encode()).hexdigest()[:8]
+    return f"webui-test-{h}"
+
+TEST_PORT      = int(os.getenv('HERMES_WEBUI_TEST_PORT',
+                               str(_auto_test_port(REPO_ROOT))))
 TEST_BASE      = f"http://127.0.0.1:{TEST_PORT}"
 TEST_STATE_DIR = pathlib.Path(os.getenv(
     'HERMES_WEBUI_TEST_STATE_DIR',
-    str(HERMES_HOME / 'webui-mvp-test')
+    str(HERMES_HOME / _auto_state_dir_name(REPO_ROOT))
 ))
 TEST_WORKSPACE = TEST_STATE_DIR / 'test-workspace'
 
+# Publish at module level so _pytest_port.py (imported at collection time)
+# and any test file using os.environ sees the right values immediately.
+os.environ.setdefault('HERMES_WEBUI_TEST_PORT', str(TEST_PORT))
+os.environ.setdefault('HERMES_WEBUI_TEST_STATE_DIR', str(TEST_STATE_DIR))
+
 # ── Server script: always relative to repo root ───────────────────────────
 SERVER_SCRIPT = REPO_ROOT / 'server.py'
 if not SERVER_SCRIPT.exists():
@@ -153,6 +176,8 @@ def pytest_collection_modifyitems(config, items):
         # Agent backend (need running AIAgent)
         'test_chat_stream_opens_successfully',
         'test_approval_submit_and_respond',
+        # Security redaction (flaky — session state varies across test ordering)
+        'test_api_sessions_list_redacts_titles',
         # Workspace path (macOS /tmp -> /private/tmp symlink)
         'test_new_session_inherits_workspace',
         'test_workspace_add_valid',
@@ -170,7 +195,7 @@ def pytest_collection_modifyitems(config, items):
             skipped += 1
 
     if skipped:
-        print(f"\n⚠️  hermes-agent not found — {skipped} agent-dependent tests will be skipped\n")
+        print(f"\nWARNING: hermes-agent not found; {skipped} agent-dependent tests will be skipped\n")
 
 
 # ── Helpers ──────────────────────────────────────────────────────────────────
@@ -210,6 +235,18 @@ def test_server():
     Start an isolated test server on TEST_PORT with a clean state directory.
     Paths are discovered dynamically -- no hardcoded absolute path assumptions.
     """
+    # Kill any leftover process on the test port before starting.
+    # Stale servers from QA harness runs or prior test sessions cause
+    # conftest to think the server is already up, producing false failures.
+    try:
+        import subprocess as _sp
+        _sp.run(['fuser', '-k', f'{TEST_PORT}/tcp'],
+                capture_output=True, timeout=5)
+    except Exception:
+        pass
+    import time as _time
+    _time.sleep(0.5)  # brief pause to let the port release
+
     # Clean slate
     if TEST_STATE_DIR.exists():
         shutil.rmtree(TEST_STATE_DIR)
@@ -226,7 +263,25 @@ def test_server():
     # Isolated cron state
     (TEST_STATE_DIR / 'cron').mkdir(parents=True, exist_ok=True)
 
+    # Expose TEST_STATE_DIR to the test process itself so that tests which write
+    # directly to state.db (e.g. test_gateway_sync.py) always use the same path
+    # as the server.  Other test files (test_auth_sessions.py) may override
+    # HERMES_WEBUI_STATE_DIR for their own purposes, but HERMES_WEBUI_TEST_STATE_DIR
+    # is reserved for this mapping and is never overridden by individual test files.
+    # Export both port and state-dir as env vars so individual test files
+    # can read them without importing conftest (avoids circular imports).
+    os.environ.setdefault('HERMES_WEBUI_TEST_PORT', str(TEST_PORT))
+    # os.environ already set at module level above; no-op here.
+
     env = os.environ.copy()
+    # Strip real provider keys so test subprocess never inherits production credentials.
+    # The test server uses a mock/isolated config — no real API calls are made.
+    for _k in list(env):
+        if any(_k.startswith(p) for p in (
+            'OPENROUTER_API_KEY', 'OPENAI_API_KEY', 'ANTHROPIC_API_KEY',
+            'GOOGLE_API_KEY', 'DEEPSEEK_API_KEY',
+        )):
+            del env[_k]
     env.update({
         "HERMES_WEBUI_PORT":              str(TEST_PORT),
         "HERMES_WEBUI_HOST":              "127.0.0.1",
@@ -234,6 +289,14 @@ def test_server():
         "HERMES_WEBUI_DEFAULT_WORKSPACE": str(TEST_WORKSPACE),
         "HERMES_WEBUI_DEFAULT_MODEL":     "openai/gpt-5.4-mini",
         "HERMES_HOME":                    str(TEST_STATE_DIR),
+        # Belt-and-suspenders: HERMES_BASE_HOME hard-locks _DEFAULT_HERMES_HOME
+        # in api/profiles.py to the test state dir regardless of profile switching
+        # or any os.environ mutation that happens inside the server process.
+        # Without this, a profile switch or active_profile file in the real
+        # ~/.hermes can redirect _get_active_hermes_home() out of the sandbox,
+        # causing onboarding writes (config.yaml, .env) to land in the production
+        # ~/.hermes/profiles/webui/ and overwrite real API keys.
+        "HERMES_BASE_HOME":               str(TEST_STATE_DIR),
     })
 
     # Pass agent dir if discovered so server.py doesn't have to re-discover
@@ -279,6 +342,33 @@ def base_url():
     return TEST_BASE
 
 
+# ── Per-test model cache invalidation ────────────────────────────────────────
+# The TTL cache for get_available_models() persists across tests within the
+# same process. Tests that modify cfg in-memory won't trigger the mtime path,
+# so the cache must be explicitly invalidated after each test that exercises
+# provider/model detection.
+
+@pytest.fixture(autouse=True)
+def _invalidate_models_cache_after_test():
+    """Force the TTL cache to be cleared before and after every test.
+
+    This prevents state bleed where a test that calls get_available_models()
+    populates the cache with a particular config, and the next test sees stale
+    results even though it has mutated _cfg_cache in-memory.
+    """
+    try:
+        from api.config import invalidate_models_cache
+        invalidate_models_cache()
+    except Exception:
+        pass
+    yield
+    try:
+        from api.config import invalidate_models_cache
+        invalidate_models_cache()
+    except Exception:
+        pass
+
+
 # ── Per-test session cleanup ──────────────────────────────────────────────────
 
 @pytest.fixture(autouse=True)

tests/test_1038_pwa_auth_redirect.py

@@ -0,0 +1,106 @@
+"""
+Tests for issue #1038 — iOS PWA auth-expiry redirect.
+
+When a 401 is returned by any API endpoint, the client-side JS should redirect
+to /login rather than showing a raw error toast. On iOS PWA standalone mode a
+server-side 302→/login breaks out of the PWA shell into Safari, so the fix is
+client-side: workspace.js api() intercepts 401 before throwing and calls
+window.location.href = '/login'.
+
+These are static regression tests that verify the JS source contains the
+correct guard patterns.
+"""
+
+import re
+from pathlib import Path
+
+ROOT = Path(__file__).parent.parent
+
+
+def _workspace_js() -> str:
+    return (ROOT / "static" / "workspace.js").read_text(encoding="utf-8")
+
+
+def _ui_js() -> str:
+    return (ROOT / "static" / "ui.js").read_text(encoding="utf-8")
+
+
+class TestPWAAuthRedirect:
+    def test_workspace_js_has_401_redirect(self):
+        """api() in workspace.js must redirect to /login on 401."""
+        src = _workspace_js()
+        # Guard must appear inside the !res.ok block, before throwing
+        assert "res.status===401" in src, \
+            "workspace.js api() must check res.status===401"
+        assert "window.location.href='/login" in src or 'window.location.href="/login' in src, \
+            "workspace.js api() must redirect to /login on 401"
+
+    def test_workspace_js_401_before_throw(self):
+        """The 401 redirect must come before the generic error throw."""
+        src = _workspace_js()
+        idx_401 = src.find("res.status===401")
+        idx_throw = src.find("throw new Error")
+        assert idx_401 != -1, "401 guard not found in workspace.js"
+        assert idx_throw != -1, "throw not found in workspace.js"
+        assert idx_401 < idx_throw, \
+            "401 redirect must appear before the generic throw in workspace.js"
+
+    def test_ui_js_has_redirect_helper(self):
+        """ui.js must define _redirectIfUnauth helper."""
+        src = _ui_js()
+        assert "_redirectIfUnauth" in src, \
+            "ui.js must define _redirectIfUnauth helper function"
+
+    def test_ui_js_models_fetch_uses_redirect(self):
+        """populateModelDropdown() must call _redirectIfUnauth on the api/models response."""
+        src = _ui_js()
+        # The helper must be called after the api/models fetch
+        assert "_redirectIfUnauth(_modelsRes)" in src, \
+            "populateModelDropdown() must check 401 on api/models fetch"
+
+    def test_ui_js_live_models_fetch_uses_redirect(self):
+        """loadLiveModels() must call _redirectIfUnauth on the api/models/live response."""
+        src = _ui_js()
+        assert "_redirectIfUnauth(_liveRes)" in src, \
+            "loadLiveModels() must check 401 on api/models/live fetch"
+
+    def test_ui_js_upload_fetch_uses_redirect(self):
+        """File upload must call _redirectIfUnauth on the api/upload response."""
+        src = _ui_js()
+        assert "_redirectIfUnauth(res)" in src, \
+            "upload fetch must call _redirectIfUnauth"
+
+
+class TestLoginJsSafeNextPath:
+    """login.js _safeNextPath() must honor ?next= but reject open-redirect payloads."""
+
+    @staticmethod
+    def _login_js():
+        return (Path(__file__).parent.parent / "static" / "login.js").read_text(encoding="utf-8")
+
+    def test_safe_next_path_function_exists(self):
+        """login.js must define _safeNextPath() to honor the ?next= redirect."""
+        assert "_safeNextPath" in self._login_js(), (
+            "login.js must define _safeNextPath() to use the ?next= redirect after login"
+        )
+
+    def test_login_uses_safe_next_path(self):
+        """doLogin success handler must redirect to _safeNextPath(), not hardcoded './'."""
+        src = self._login_js()
+        assert "_safeNextPath()" in src, (
+            "doLogin must call _safeNextPath() instead of hardcoding './'"
+        )
+
+    def test_safe_next_path_rejects_protocol_relative(self):
+        """_safeNextPath guard must reject '//' prefix (protocol-relative open-redirect)."""
+        src = self._login_js()
+        assert "charAt(1) === '/'" in src or "startsWith('//')" in src, (
+            "_safeNextPath must reject protocol-relative paths like //evil.com"
+        )
+
+    def test_safe_next_path_rejects_non_path_absolute(self):
+        """_safeNextPath guard must require path starts with '/'."""
+        src = self._login_js()
+        assert "charAt(0) !== '/'" in src or "startsWith('/')" in src, (
+            "_safeNextPath must reject non-path-absolute inputs (e.g. 'http://...')"
+        )

tests/test_1044_mermaid_csp_font.py

@@ -0,0 +1,51 @@
+"""
+Tests for issue #1044 — Mermaid CSP font violation.
+
+Mermaid's built-in themes inject an @import for Google Fonts (Manrope) at
+render time, which is blocked by the CSP's style-src directive. Fix: pass
+fontFamily:'inherit' in themeVariables so Mermaid never requests an external
+font URL.
+"""
+
+from pathlib import Path
+
+ROOT = Path(__file__).parent.parent
+
+
+def _ui_js() -> str:
+    return (ROOT / "static" / "ui.js").read_text(encoding="utf-8")
+
+
+class TestMermaidCSPFont:
+    def test_mermaid_init_has_font_family_inherit(self):
+        """themeVariables in mermaid.initialize() must set fontFamily to 'inherit'."""
+        src = _ui_js()
+        assert "fontFamily:'inherit'" in src, (
+            "mermaid.initialize() themeVariables must set fontFamily:'inherit' "
+            "to suppress the Google Fonts (Manrope) import that violates CSP"
+        )
+
+    def test_mermaid_init_no_google_fonts_url(self):
+        """ui.js must not contain a hardcoded fonts.googleapis.com URL."""
+        src = _ui_js()
+        assert "fonts.googleapis.com" not in src, (
+            "ui.js must not reference fonts.googleapis.com — use fontFamily:'inherit'"
+        )
+
+    def test_mermaid_font_family_inside_theme_variables_block(self):
+        """fontFamily:'inherit' must be inside the themeVariables block of mermaid.initialize()."""
+        src = _ui_js()
+        init_idx = src.find("mermaid.initialize(")
+        assert init_idx != -1, "mermaid.initialize() call not found in ui.js"
+        # Find the themeVariables block after the initialize call
+        tv_idx = src.find("themeVariables", init_idx)
+        assert tv_idx != -1, "themeVariables not found inside mermaid.initialize()"
+        font_idx = src.find("fontFamily:'inherit'", tv_idx)
+        assert font_idx != -1, (
+            "fontFamily:'inherit' must appear inside themeVariables in mermaid.initialize()"
+        )
+        # The closing brace of themeVariables should come after fontFamily
+        close_brace = src.find("})", tv_idx)
+        assert font_idx < close_brace, (
+            "fontFamily:'inherit' must be inside the themeVariables block (before })"
+        )

tests/test_1045_bfcache_layout_restore.py

@@ -0,0 +1,116 @@
+"""
+Tests for issue #1045 — bfcache layout broken on tab restore.
+
+When the browser restores a page from bfcache (event.persisted === true),
+the async boot IIFE does not re-run. The existing pageshow handler (added for
+#822) only cleared the session search field and re-rendered the session list.
+This left the rail, topbar, workspace panel, and resize handles in the stale
+bfcache DOM state, producing a broken layout.
+
+Fix: extend the pageshow handler to also call syncTopbar, syncWorkspacePanelState,
+_initResizePanels, and startGatewaySSE — all guarded so missing helpers degrade.
+"""
+
+from pathlib import Path
+
+ROOT = Path(__file__).parent.parent
+
+
+def _boot_js() -> str:
+    return (ROOT / "static" / "boot.js").read_text(encoding="utf-8")
+
+
+class TestBfcacheLayoutRestore:
+    def test_pageshow_calls_sync_topbar(self):
+        """pageshow handler must call syncTopbar() on bfcache restore."""
+        src = _boot_js()
+        # Find the pageshow listener block
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        assert ps_idx != -1, "pageshow listener not found in boot.js"
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "syncTopbar" in handler_body, (
+            "pageshow handler must call syncTopbar() to restore topbar state after bfcache"
+        )
+
+    def test_pageshow_calls_sync_workspace_panel_state(self):
+        """pageshow handler must call syncWorkspacePanelState()."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "syncWorkspacePanelState" in handler_body, (
+            "pageshow handler must call syncWorkspacePanelState() on bfcache restore"
+        )
+
+
+    def test_pageshow_calls_start_gateway_sse(self):
+        """pageshow handler must call startGatewaySSE() to reconnect the dead SSE connection."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "startGatewaySSE" in handler_body, (
+            "pageshow handler must restart gateway SSE (bfcache-persisted connections are dead)"
+        )
+
+    def test_pageshow_still_clears_session_search(self):
+        """pageshow handler must still clear #sessionSearch (original #822 fix preserved)."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "sessionSearch" in handler_body, (
+            "pageshow handler must still clear #sessionSearch (regression: #822 fix must be preserved)"
+        )
+
+    def test_pageshow_still_calls_render_session_list_from_cache(self):
+        """pageshow handler must still call renderSessionListFromCache()."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "renderSessionListFromCache" in handler_body, (
+            "pageshow handler must still call renderSessionListFromCache() (regression: #822 fix)"
+        )
+
+    def test_pageshow_does_not_call_init_resize_panels(self):
+        """pageshow handler must NOT call _initResizePanels() — bfcache
+        preserves event listeners so re-attaching them stacks duplicates."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        assert "_initResizePanels" not in handler_body, (
+            "pageshow handler must not call _initResizePanels() — it stacks "
+            "duplicate mousedown listeners on every bfcache restore"
+        )
+
+    def test_new_calls_are_guarded_with_typeof(self):
+        """New calls in the pageshow handler must be typeof-guarded for safe degradation."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        # Each of the new calls must be guarded
+        for fn in ("syncTopbar", "syncWorkspacePanelState", "startGatewaySSE",
+                   "closeModelDropdown", "closeReasoningDropdown", "closeWsDropdown", "closeProfileDropdown"):
+            assert f"typeof {fn} === 'function'" in handler_body, (
+                f"{fn}() call in pageshow handler must be guarded with typeof === 'function'"
+            )
+
+    def test_pageshow_closes_all_dropdowns(self):
+        """pageshow handler must close all known dropdowns to reset frozen bfcache popover state."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        for fn in ("closeModelDropdown", "closeReasoningDropdown", "closeWsDropdown", "closeProfileDropdown"):
+            assert fn in handler_body, (
+                f"pageshow handler must call {fn}() to dismiss any dropdown left open by bfcache"
+            )
+
+    def test_dropdowns_closed_before_layout_sync(self):
+        """Dropdown closes must come before layout sync calls (clean state first)."""
+        src = _boot_js()
+        ps_idx = src.find("window.addEventListener('pageshow'")
+        handler_body = src[ps_idx:ps_idx + 1600]
+        close_idx = handler_body.find("closeModelDropdown")
+        sync_idx = handler_body.find("syncTopbar")
+        assert close_idx != -1 and sync_idx != -1, "Both close and sync calls must be present"
+        assert close_idx < sync_idx, (
+            "Dropdown close calls must appear before layout sync calls in the pageshow handler"
+        )
+

tests/test_745_code_block_newlines.py

@@ -0,0 +1,104 @@
+"""
+Tests for #745: code blocks losing newlines when not preceded by double blank line.
+
+Root cause: the paragraph-splitter in renderMd() replaced \n with <br> inside
+<pre><code> blocks when they were not separated by a double newline from surrounding
+text. The fix stashes <pre> blocks (and pre-header divs, mermaid, katex) before
+the paragraph split and restores them afterwards.
+"""
+import re
+import subprocess
+import sys
+import os
+
+UI_JS = os.path.join(os.path.dirname(__file__), '..', 'static', 'ui.js')
+
+
+def get_ui_js():
+    return open(UI_JS, encoding='utf-8').read()
+
+
+class TestCodeBlockNewlinePreservation:
+
+    def test_pre_stash_present(self):
+        """The _pre_stash variable must exist in ui.js."""
+        src = get_ui_js()
+        assert '_pre_stash' in src, "_pre_stash not found in ui.js"
+
+    def test_pre_stash_token_E_used(self):
+        """Stash token \\x00E must be used for pre-block stashing."""
+        src = get_ui_js()
+        assert r'\x00E' in src, r"\x00E stash token not found in ui.js"
+
+    def test_stash_before_paragraph_split(self):
+        """_pre_stash must be populated BEFORE the parts=s.split line."""
+        src = get_ui_js()
+        pre_stash_pos = src.index('_pre_stash=[]')
+        split_pos = src.index('const parts=s.split(/\\n{2,}/)')
+        assert pre_stash_pos < split_pos, \
+            "_pre_stash must be initialised before the paragraph split"
+
+    def test_restore_after_paragraph_split(self):
+        """_pre_stash restore must happen AFTER the paragraph map/join line."""
+        src = get_ui_js()
+        restore_pos = src.index('_pre_stash[+i]')
+        split_pos = src.index("}).join('\\n');", src.index('const parts=s.split'))
+        assert restore_pos > split_pos, \
+            "_pre_stash must be restored after the paragraph split/join"
+
+    def test_paragraph_split_bypasses_stash_tokens(self):
+        """The paragraph map must bypass lines that start with \\x00E."""
+        src = get_ui_js()
+        # The map line must check for \x00E in its bypass condition
+        map_line = next(
+            l for l in src.splitlines()
+            if 'parts.map' in l and '<br>' in l
+        )
+        assert r'\x00E' in map_line, \
+            r"paragraph map must bypass \x00E stash tokens"
+
+    def test_pre_regex_covers_pre_header_div(self):
+        """The stash regex must match <div class=\"pre-header\"> before <pre>."""
+        src = get_ui_js()
+        # Find the replacement regex used to populate _pre_stash
+        stash_block_idx = src.index('_pre_stash=[]')
+        stash_block = src[stash_block_idx:stash_block_idx + 400]
+        assert 'pre-header' in stash_block, \
+            "pre-stash regex must match <div class=\"pre-header\"> wrappers"
+
+    def test_mermaid_covered_by_stash(self):
+        """The stash regex must also cover mermaid-block divs."""
+        src = get_ui_js()
+        stash_block_idx = src.index('_pre_stash=[]')
+        stash_block = src[stash_block_idx:stash_block_idx + 400]
+        assert 'mermaid-block' in stash_block, \
+            "pre-stash regex must cover mermaid-block divs"
+
+    def test_katex_covered_by_stash(self):
+        """The stash regex must also cover katex-block divs."""
+        src = get_ui_js()
+        stash_block_idx = src.index('_pre_stash=[]')
+        stash_block = src[stash_block_idx:stash_block_idx + 400]
+        assert 'katex-block' in stash_block, \
+            "pre-stash regex must cover katex-block divs"
+
+    def test_js_syntax_valid(self):
+        """ui.js must pass node --check after the fix."""
+        result = subprocess.run(
+            ['node', '--check', UI_JS],
+            capture_output=True, text=True
+        )
+        assert result.returncode == 0, \
+            f"node --check failed:\n{result.stderr}"
+
+    def test_stash_token_e_not_used_elsewhere(self):
+        """\\x00E must only appear in the pre-stash section (not reused)."""
+        src = get_ui_js()
+        occurrences = [
+            i for i in range(len(src))
+            if src[i:i+4] == r'\x00' and i + 4 < len(src) and src[i+4] == 'E'
+        ]
+        # Allow 2 occurrences: the push token and the restore regex
+        # (may be 3 if there's also a comment mentioning it)
+        assert len(occurrences) >= 2, \
+            r"Expected at least 2 uses of \x00E (push + restore)"

tests/test_779_html_preview.py

@@ -0,0 +1,94 @@
+"""Tests for inline HTML preview in workspace panel (issue #779)."""
+import pytest
+
+
+def _get_routes_content():
+    return open("api/routes.py", encoding="utf-8").read()
+
+
+def _get_workspace_js():
+    return open("static/workspace.js", encoding="utf-8").read()
+
+
+def _get_index_html():
+    return open("static/index.html", encoding="utf-8").read()
+
+
+def test_inline_preview_param_in_file_raw():
+    """?inline=1 must bypass Content-Disposition: attachment for text/html."""
+    content = _get_routes_content()
+    assert "inline_preview" in content, (
+        "_handle_file_raw must read the inline query parameter"
+    )
+    assert "html_inline_ok" in content, (
+        "_handle_file_raw must allow HTML inline when inline_preview=True"
+    )
+
+
+def test_iframe_uses_inline_param():
+    """workspace.js must pass &inline=1 when setting the preview iframe src."""
+    content = _get_workspace_js()
+    assert "inline=1" in content, (
+        "workspace.js must pass ?inline=1 to api/file/raw for the HTML preview iframe"
+    )
+
+
+def test_html_preview_iframe_exists_in_html():
+    """The previewHtmlIframe element must be present in index.html."""
+    content = _get_index_html()
+    assert "previewHtmlIframe" in content, (
+        "index.html must contain the previewHtmlIframe element"
+    )
+
+
+def test_html_exts_defined_in_workspace_js():
+    """HTML_EXTS set must include .html and .htm."""
+    content = _get_workspace_js()
+    assert "HTML_EXTS" in content, "workspace.js must define HTML_EXTS"
+    assert "'.html'" in content or '".html"' in content, "HTML_EXTS must include .html"
+    assert "'.htm'" in content or '".htm"' in content, "HTML_EXTS must include .htm"
+
+
+def test_sandbox_allows_scripts_only():
+    """iframe sandbox must not include allow-same-origin (XSS risk)."""
+    content = _get_index_html()
+    # Find the sandbox attribute value
+    import re
+    sandboxes = re.findall(r'sandbox="([^"]*)"', content)
+    preview_sandboxes = [s for s in sandboxes if "allow" in s]
+    for sb in preview_sandboxes:
+        assert "allow-same-origin" not in sb, (
+            "HTML preview iframe must not have allow-same-origin (would expose parent cookies)"
+        )
+
+
+def test_inline_html_response_sets_csp_sandbox():
+    """Defense-in-depth: ?inline=1 HTML responses must set Content-Security-Policy:
+    sandbox so the same origin isolation applies even when the URL is opened
+    directly in a top-level tab (not just inside the workspace panel iframe).
+
+    Without this, a user tricked into clicking a chat link like
+    /api/file/raw?path=evil.html&inline=1 would render the HTML in the WebUI's
+    origin without any sandbox, giving the page full access to cookies and
+    localStorage. The CSP sandbox directive (no allow-same-origin) downgrades
+    the document to a unique opaque origin server-side.
+    """
+    content = _get_routes_content()
+    # Find the html_inline_ok block in _handle_file_raw
+    idx = content.find("html_inline_ok")
+    assert idx != -1, "html_inline_ok block not found"
+    block = content[idx:idx + 2500]
+    assert "Content-Security-Policy" in block, (
+        "_handle_file_raw must set Content-Security-Policy header on inline HTML responses"
+    )
+    assert "sandbox" in block, (
+        "CSP must include the sandbox directive"
+    )
+    # Must NOT have allow-same-origin in the sandbox directive
+    csp_sections = [line for line in block.splitlines() if "sandbox" in line and "Policy" in line]
+    for line in csp_sections:
+        # The line setting the CSP header — make sure it doesn't grant same-origin
+        if "send_header" in line:
+            assert "allow-same-origin" not in line, (
+                "CSP sandbox must NOT include allow-same-origin — that would defeat the isolation"
+            )

tests/test_886_ordered_list_numbering.py

@@ -0,0 +1,92 @@
+"""
+Tests for #886: ordered list items always rendered as "1." regardless of position.
+
+Root cause: when LLMs output numbered lists with blank lines between items,
+the paragraph-splitter in renderMd() splits the markdown into one chunk per item,
+so the ordered-list regex wraps each item in its own <ol>. Each <ol> restarts
+at 1, producing "1. 1. 1." instead of "1. 2. 3.".
+
+Fix: emit value="N" on every <li> so the correct ordinal is preserved even when
+items end up in separate <ol> containers after the paragraph split.
+"""
+import os
+import re
+
+UI_JS = os.path.join(os.path.dirname(__file__), '..', 'static', 'ui.js')
+
+
+def get_ui_js():
+    return open(UI_JS, encoding='utf-8').read()
+
+
+class TestOrderedListNumbering:
+
+    def test_li_value_attr_present_in_ordered_list_block(self):
+        """The ordered-list renderer must emit value= on each <li>."""
+        src = get_ui_js()
+        # Locate the ordered-list replace block
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found in ui.js"
+        # Extract a window large enough to cover the whole closure (~400 chars)
+        ol_block = src[ol_idx:ol_idx + 500]
+        assert 'value=' in ol_block, (
+            "Ordered-list block must emit value= attribute on <li> elements to "
+            "preserve numbering when items are separated by blank lines (#886)"
+        )
+
+    def test_li_value_uses_parsed_number(self):
+        """The value= must be derived from parseInt of the captured digit, not hardcoded."""
+        src = get_ui_js()
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found in ui.js"
+        ol_block = src[ol_idx:ol_idx + 500]
+        assert 'parseInt' in ol_block, (
+            "Ordered-list block should use parseInt() to parse the list number (#886)"
+        )
+
+    def test_numMatch_variable_present(self):
+        """The numMatch variable (or equivalent digit capture) must exist in the OL block."""
+        src = get_ui_js()
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found in ui.js"
+        ol_block = src[ol_idx:ol_idx + 500]
+        # Either numMatch or a similar digit-capture variable
+        assert 'numMatch' in ol_block or re.search(r'match\(/.*\\d', ol_block), (
+            "Ordered-list block should capture the list item number with a regex match (#886)"
+        )
+
+    def test_valAttr_or_value_template_present(self):
+        """The <li> template must include the value attribute conditionally or unconditionally."""
+        src = get_ui_js()
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found in ui.js"
+        ol_block = src[ol_idx:ol_idx + 500]
+        # Either a valAttr variable or an inline value= in the template
+        has_val_attr = 'valAttr' in ol_block
+        has_inline_value = re.search(r'<li.*value=', ol_block)
+        assert has_val_attr or has_inline_value, (
+            "Ordered-list block must have value= on <li> (via valAttr var or inline) (#886)"
+        )
+
+    def test_ordered_list_comment_references_issue(self):
+        """A comment near the OL fix should reference the issue (#886) or the symptom."""
+        src = get_ui_js()
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found in ui.js"
+        # Look at the 300 chars BEFORE the replace line for an explanatory comment
+        context = src[max(0, ol_idx - 300):ol_idx]
+        has_comment = '#886' in context or '1. 1. 1.' in context or 'blank lines' in context.lower()
+        assert has_comment, (
+            "Expected a comment near the OL fix explaining the blank-line issue (#886)"
+        )
+
+    def test_list_without_blank_lines_unaffected(self):
+        """A compact list (no blank lines) should still produce one <ol> with sequential items."""
+        src = get_ui_js()
+        # Structural check: the regex still captures multi-line blocks (\\n? allows groups)
+        ol_idx = src.find('s=s.replace(/((?:^(?:  )?\\d+\\. .+\\n?)+)/gm')
+        assert ol_idx != -1, "Ordered-list replace block not found"
+        # The \\n? quantifier that allows grouping must still be present
+        assert '\\n?' in src[ol_idx:ol_idx + 80], (
+            "The \\\\n? in the ordered-list regex was removed — compact lists may break"
+        )

tests/test_approval_queue.py

@@ -0,0 +1,188 @@
+"""Tests for approval queue multi-entry support (issue #527).
+
+Previously _pending[sid] held one entry, so simultaneous approvals overwrote
+each other. This PR changes submit_pending() to append to a list and adds
+approval_id so /api/approval/respond can target a specific entry.
+"""
+import json
+import pathlib
+import re
+import sys
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
+sys.path.insert(0, str(REPO_ROOT))
+
+ROUTES_SRC = (REPO_ROOT / "api" / "routes.py").read_text(encoding="utf-8")
+MESSAGES_JS = (REPO_ROOT / "static" / "messages.js").read_text(encoding="utf-8")
+INDEX_HTML = (REPO_ROOT / "static" / "index.html").read_text(encoding="utf-8")
+
+
+# ---------------------------------------------------------------------------
+# Static-analysis: Python routes
+# ---------------------------------------------------------------------------
+
+def test_submit_pending_appends_to_list():
+    """submit_pending() must append to a list, not overwrite."""
+    # The new wrapper must contain queue.append
+    assert "queue.append(entry)" in ROUTES_SRC, \
+        "submit_pending() must append entry to a list queue, not overwrite _pending[sid]"
+
+
+def test_submit_pending_adds_approval_id():
+    """Each queued entry must get a unique approval_id."""
+    assert "approval_id" in ROUTES_SRC and "uuid.uuid4().hex" in ROUTES_SRC, \
+        "submit_pending() must assign a uuid4 approval_id to each queued entry"
+
+
+def test_handle_approval_pending_returns_count():
+    """_handle_approval_pending must return pending_count in its response."""
+    assert '"pending_count"' in ROUTES_SRC, \
+        "_handle_approval_pending must include pending_count in the JSON response"
+
+
+def test_handle_approval_respond_pops_by_approval_id():
+    """_handle_approval_respond must target entry by approval_id."""
+    assert 'approval_id = body.get("approval_id"' in ROUTES_SRC, \
+        "_handle_approval_respond must read approval_id from request body"
+    assert 'entry.get("approval_id") == approval_id' in ROUTES_SRC, \
+        "_handle_approval_respond must find and pop the matching entry by approval_id"
+
+
+def test_handle_approval_respond_fallback_to_oldest():
+    """When no approval_id is given, fall back to popping the oldest entry (FIFO)."""
+    # The fallback path: queue.pop(0) when approval_id is empty
+    assert "queue.pop(0)" in ROUTES_SRC, \
+        "_handle_approval_respond must fall back to popping the oldest entry when approval_id is absent"
+
+
+def test_backward_compat_legacy_dict_value():
+    """The respond handler must tolerate a legacy single-dict value in _pending."""
+    assert "Legacy single-dict value" in ROUTES_SRC or \
+           "# Legacy single-dict" in ROUTES_SRC or \
+           "elif queue:" in ROUTES_SRC, \
+        "respond handler must handle legacy single-dict _pending values for backward compatibility"
+
+
+# ---------------------------------------------------------------------------
+# Static-analysis: JavaScript frontend
+# ---------------------------------------------------------------------------
+
+def test_respond_sends_approval_id():
+    """respondApproval() must include approval_id in the POST body."""
+    assert "approval_id: approvalId" in MESSAGES_JS, \
+        "respondApproval() must send approval_id in the POST body to /api/approval/respond"
+
+
+def test_show_approval_card_accepts_count():
+    """showApprovalCard must accept a pendingCount parameter."""
+    assert re.search(r"function showApprovalCard\(pending,\s*pendingCount\)", MESSAGES_JS), \
+        "showApprovalCard() must accept a pendingCount argument"
+
+
+def test_show_approval_card_renders_counter():
+    """showApprovalCard must display a '1 of N pending' counter when N > 1."""
+    assert '"1 of " + pendingCount + " pending"' in MESSAGES_JS or \
+           "'1 of ' + pendingCount + ' pending'" in MESSAGES_JS, \
+        "showApprovalCard() must render '1 of N pending' counter for multiple queued approvals"
+
+
+def test_approval_current_id_tracked():
+    """_approvalCurrentId must be set and cleared around each approval."""
+    assert "_approvalCurrentId" in MESSAGES_JS, \
+        "_approvalCurrentId must track the approval_id of the currently displayed card"
+    assert "_approvalCurrentId = pending.approval_id" in MESSAGES_JS or \
+           "_approvalCurrentId = pending.approval_id || null" in MESSAGES_JS, \
+        "_approvalCurrentId must be assigned from pending.approval_id"
+    # Must be nulled on respond
+    assert "_approvalCurrentId = null" in MESSAGES_JS, \
+        "_approvalCurrentId must be cleared when respondApproval() is called"
+
+
+def test_polling_passes_count_to_show():
+    """The poll loop must pass pending_count to showApprovalCard."""
+    assert "showApprovalCard(data.pending, data.pending_count" in MESSAGES_JS, \
+        "Poll loop must pass data.pending_count to showApprovalCard"
+
+
+# ---------------------------------------------------------------------------
+# HTML: counter element present
+# ---------------------------------------------------------------------------
+
+def test_approval_counter_element_exists():
+    """index.html must contain an approvalCounter element."""
+    assert 'id="approvalCounter"' in INDEX_HTML, \
+        "index.html must contain an element with id='approvalCounter' for the '1 of N' display"
+
+
+# ---------------------------------------------------------------------------
+# Functional: multiple entries behave correctly (via routes module directly)
+# ---------------------------------------------------------------------------
+
+def test_multiple_approvals_both_surfaced():
+    """Two submit_pending calls must produce two queued entries, not one."""
+    import threading
+    from api import routes as r
+
+    # Reset state
+    sid = "test-multi-approval-sid"
+    with r._lock:
+        r._pending.pop(sid, None)
+
+    r.submit_pending(sid, {"command": "cmd1", "pattern_key": "p1", "pattern_keys": ["p1"], "description": "d1"})
+    r.submit_pending(sid, {"command": "cmd2", "pattern_key": "p2", "pattern_keys": ["p2"], "description": "d2"})
+
+    with r._lock:
+        queue = r._pending.get(sid)
+
+    assert isinstance(queue, list), "After two submit_pending calls, _pending[sid] must be a list"
+    assert len(queue) == 2, f"Expected 2 queued entries, got {len(queue)}"
+    assert queue[0]["command"] == "cmd1"
+    assert queue[1]["command"] == "cmd2"
+    assert queue[0].get("approval_id"), "First entry must have an approval_id"
+    assert queue[1].get("approval_id"), "Second entry must have an approval_id"
+    assert queue[0]["approval_id"] != queue[1]["approval_id"], "Each entry must have a unique approval_id"
+
+    # Cleanup
+    with r._lock:
+        r._pending.pop(sid, None)
+
+
+def test_respond_by_approval_id_pops_correct_entry():
+    """Responding with approval_id must remove only the targeted entry."""
+    from api import routes as r
+
+    sid = "test-respond-by-id-sid"
+    with r._lock:
+        r._pending.pop(sid, None)
+
+    r.submit_pending(sid, {"command": "cmd1", "pattern_key": "p1", "pattern_keys": ["p1"], "description": "d1"})
+    r.submit_pending(sid, {"command": "cmd2", "pattern_key": "p2", "pattern_keys": ["p2"], "description": "d2"})
+
+    with r._lock:
+        queue = r._pending.get(sid, [])
+        aid2 = queue[1]["approval_id"] if len(queue) > 1 else None
+
+    assert aid2, "Second entry must have an approval_id"
+
+    # Respond to the SECOND entry by its approval_id
+    # We call the handler internals directly (no HTTP)
+    with r._lock:
+        queue = r._pending.get(sid, [])
+        popped = None
+        for i, entry in enumerate(queue):
+            if entry.get("approval_id") == aid2:
+                popped = queue.pop(i)
+                break
+
+    assert popped is not None, "Should have found and popped entry by approval_id"
+    assert popped["command"] == "cmd2", "Popped the wrong entry"
+
+    with r._lock:
+        remaining = r._pending.get(sid, [])
+
+    assert len(remaining) == 1, "One entry should remain after popping the second"
+    assert remaining[0]["command"] == "cmd1", "The remaining entry should be cmd1"
+
+    # Cleanup
+    with r._lock:
+        r._pending.pop(sid, None)

tests/test_approval_unblock.py

@@ -0,0 +1,288 @@
+"""
+Tests for fix/approval-stuck-thinking:
+Verify that /api/approval/respond correctly unblocks gateway approval queues
+and that the approval module exports the symbols streaming.py and routes.py
+need to prevent the UI getting stuck in "Thinking…" during dangerous commands.
+"""
+
+import json
+import threading
+import uuid
+import urllib.request
+import urllib.error
+import urllib.parse
+
+import pytest
+
+# Import approval internals — shared module-level state within this process.
+# The HTTP tests use the test server (port 8788, separate process).
+# The unit tests operate directly on the module.
+try:
+    from tools.approval import (
+        register_gateway_notify,
+        unregister_gateway_notify,
+        resolve_gateway_approval,
+        _gateway_queues,
+        _gateway_notify_cbs,
+        _lock,
+        _ApprovalEntry,
+        submit_pending,
+    )
+    # has_pending and pop_pending were removed from tools.approval when the
+    # agent renamed has_pending -> has_blocking_approval (gateway queue check)
+    # and removed the polling-mode pop_pending. Routes now check _pending
+    # directly. These symbols are no longer part of the public API.
+    APPROVAL_AVAILABLE = True
+except ImportError:
+    APPROVAL_AVAILABLE = False
+
+pytestmark = pytest.mark.skipif(
+    not APPROVAL_AVAILABLE,
+    reason="tools.approval not available in this environment"
+)
+
+from tests._pytest_port import BASE
+
+
+def get(path):
+    url = BASE + path
+    with urllib.request.urlopen(url, timeout=10) as r:
+        return json.loads(r.read())
+
+
+def post(path, body=None):
+    url = BASE + path
+    data = json.dumps(body or {}).encode()
+    req = urllib.request.Request(url, data=data,
+          headers={"Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=10) as r:
+            return json.loads(r.read()), r.status
+    except urllib.error.HTTPError as e:
+        return json.loads(e.read()), e.code
+
+
+# ── Unit tests (in-process, no HTTP server needed) ──────────────────────────
+
+class TestGatewayApprovalUnblocking:
+    """Unit tests for the gateway queue unblocking mechanism."""
+
+    def test_resolve_gateway_approval_sets_event(self):
+        """resolve_gateway_approval() must set the entry's event and store the result."""
+        sid = f"unit-resolve-{uuid.uuid4().hex[:8]}"
+        data = {"command": "rm -rf /tmp/x", "description": "recursive delete"}
+        entry = _ApprovalEntry(data)
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+
+        resolved = resolve_gateway_approval(sid, "once", resolve_all=False)
+        assert resolved == 1
+        assert entry.event.is_set()
+        assert entry.result == "once"
+
+        # Queue should be cleaned up
+        with _lock:
+            assert sid not in _gateway_queues
+
+    def test_resolve_gateway_approval_deny(self):
+        """Deny choice is propagated correctly."""
+        sid = f"unit-deny-{uuid.uuid4().hex[:8]}"
+        entry = _ApprovalEntry({"command": "pkill -9 x", "description": "force kill"})
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+
+        resolve_gateway_approval(sid, "deny")
+        assert entry.result == "deny"
+
+    def test_resolve_gateway_approval_no_queue_is_harmless(self):
+        """resolve_gateway_approval with no queue entry returns 0, no crash."""
+        sid = f"unit-no-queue-{uuid.uuid4().hex[:8]}"
+        result = resolve_gateway_approval(sid, "once")
+        assert result == 0
+
+    def test_resolve_all_unblocks_multiple_entries(self):
+        """resolve_all=True unblocks every pending entry in the queue."""
+        sid = f"unit-resolve-all-{uuid.uuid4().hex[:8]}"
+        entries = [_ApprovalEntry({"command": f"cmd{i}"}) for i in range(3)]
+        with _lock:
+            _gateway_queues[sid] = list(entries)
+
+        resolved = resolve_gateway_approval(sid, "session", resolve_all=True)
+        assert resolved == 3
+        for e in entries:
+            assert e.event.is_set()
+            assert e.result == "session"
+
+    def test_register_and_fire_notify_cb(self):
+        """register_gateway_notify stores the cb; calling it delivers approval data."""
+        sid = f"unit-notify-{uuid.uuid4().hex[:8]}"
+        fired = []
+        register_gateway_notify(sid, lambda d: fired.append(d))
+
+        with _lock:
+            cb = _gateway_notify_cbs.get(sid)
+        assert cb is not None
+
+        data = {"command": "test", "description": "test"}
+        cb(data)
+        assert fired == [data]
+
+        unregister_gateway_notify(sid)
+
+    def test_unregister_clears_cb_and_signals_entries(self):
+        """unregister_gateway_notify removes cb and unblocks any queued entries."""
+        sid = f"unit-unreg-{uuid.uuid4().hex[:8]}"
+        register_gateway_notify(sid, lambda d: None)
+
+        entry = _ApprovalEntry({"command": "x"})
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+
+        unregister_gateway_notify(sid)
+
+        assert entry.event.is_set(), "unregister should signal blocked entries"
+        with _lock:
+            assert sid not in _gateway_notify_cbs
+            assert sid not in _gateway_queues
+
+    def test_streaming_approval_integration(self):
+        """
+        End-to-end unit simulation of the streaming.py fix:
+        1. streaming.py registers notify_cb
+        2. check_all_command_guards fires notify_cb (pushing approval SSE)
+        3. User responds — resolve_gateway_approval unblocks agent thread
+        4. Agent thread sees choice and continues
+        """
+        sid = f"unit-e2e-{uuid.uuid4().hex[:8]}"
+        approval_events_sent = []
+
+        # Step 1: streaming.py registers the notify callback
+        def _approval_notify_cb(approval_data):
+            approval_events_sent.append(approval_data)  # would be put('approval', ...)
+        register_gateway_notify(sid, _approval_notify_cb)
+
+        # Step 2: check_all_command_guards fires the callback and queues an entry
+        approval_data = {
+            "command": "rm -rf /tmp/test",
+            "pattern_key": "recursive delete",
+            "pattern_keys": ["recursive delete"],
+            "description": "recursive delete",
+        }
+        entry = _ApprovalEntry(approval_data)
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+        # notify_cb fires synchronously (gateway notifies user)
+        with _lock:
+            cb = _gateway_notify_cbs.get(sid)
+        cb(approval_data)
+
+        assert len(approval_events_sent) == 1, "approval SSE event should have been queued"
+
+        # Step 3: user responds via /api/approval/respond → resolve_gateway_approval
+        resolved = resolve_gateway_approval(sid, "once")
+        assert resolved == 1
+
+        # Step 4: agent thread is unblocked with the correct choice
+        assert entry.event.is_set()
+        assert entry.result == "once"
+
+        # Cleanup
+        unregister_gateway_notify(sid)
+
+
+# ── Symbol existence tests ───────────────────────────────────────────────────
+
+class TestApprovalModuleExports:
+    """Verify the module exports all symbols that streaming.py and routes.py need."""
+
+    def test_register_gateway_notify_exported(self):
+        import tools.approval as ap
+        assert hasattr(ap, "register_gateway_notify"), \
+            "tools.approval must export register_gateway_notify"
+
+    def test_unregister_gateway_notify_exported(self):
+        import tools.approval as ap
+        assert hasattr(ap, "unregister_gateway_notify"), \
+            "tools.approval must export unregister_gateway_notify"
+
+    def test_resolve_gateway_approval_exported(self):
+        import tools.approval as ap
+        assert hasattr(ap, "resolve_gateway_approval"), \
+            "tools.approval must export resolve_gateway_approval"
+
+    def test_approval_entry_exported(self):
+        import tools.approval as ap
+        assert hasattr(ap, "_ApprovalEntry"), \
+            "tools.approval must export _ApprovalEntry"
+
+
+# ── HTTP regression tests (test server, port 8788) ───────────────────────────
+
+class TestApprovalHTTPEndpoints:
+    """
+    Regression tests for /api/approval/respond against the live test server.
+    These verify that the HTTP layer behaves correctly — they don't rely on
+    in-process module state shared with the server subprocess.
+    """
+
+    def test_respond_returns_ok_no_pending(self):
+        """respond with no pending entry returns ok (no crash, no 500)."""
+        sid = f"http-no-pending-{uuid.uuid4().hex[:8]}"
+        result, status = post("/api/approval/respond", {
+            "session_id": sid,
+            "choice": "deny",
+        })
+        assert status == 200
+        assert result["ok"] is True
+
+    def test_respond_clears_injected_pending(self):
+        """Inject a pending entry, respond, verify it's cleared."""
+        sid = f"http-clear-{uuid.uuid4().hex[:8]}"
+        cmd = "rm -rf /tmp/testdir"
+
+        inject = get(f"/api/approval/inject_test?session_id={urllib.parse.quote(sid)}"
+                     f"&pattern_key=recursive+delete&command={urllib.parse.quote(cmd)}")
+        assert inject["ok"] is True
+
+        data = get(f"/api/approval/pending?session_id={urllib.parse.quote(sid)}")
+        assert data["pending"] is not None
+
+        result, status = post("/api/approval/respond", {
+            "session_id": sid,
+            "choice": "deny",
+        })
+        assert status == 200
+        assert result["ok"] is True
+
+        data2 = get(f"/api/approval/pending?session_id={urllib.parse.quote(sid)}")
+        assert data2["pending"] is None, "pending should be cleared after respond"
+
+    def test_respond_rejects_invalid_choice(self):
+        """respond with an unknown choice returns 400."""
+        result, status = post("/api/approval/respond", {
+            "session_id": "some-session",
+            "choice": "INVALID",
+        })
+        assert status == 400
+
+    def test_respond_requires_session_id(self):
+        """respond without session_id returns 400."""
+        result, status = post("/api/approval/respond", {"choice": "deny"})
+        assert status == 400
+
+    def test_respond_session_choice_clears_pending(self):
+        """Inject pending, respond with 'session', verify cleared."""
+        sid = f"http-session-{uuid.uuid4().hex[:8]}"
+        inject = get(f"/api/approval/inject_test?session_id={urllib.parse.quote(sid)}"
+                     f"&pattern_key=force+kill+processes&command=pkill+-9+something")
+        assert inject["ok"] is True
+
+        result, status = post("/api/approval/respond", {
+            "session_id": sid,
+            "choice": "session",
+        })
+        assert status == 200
+        assert result["choice"] == "session"
+
+        data = get(f"/api/approval/pending?session_id={urllib.parse.quote(sid)}")
+        assert data["pending"] is None

tests/test_auth_session_persistence.py

@@ -0,0 +1,94 @@
+"""Regression tests: auth sessions persist across process restarts.
+
+_sessions is an in-memory dict. Without persistence, any restart (launchd,
+systemd, container) invalidates all active browser sessions and floods clients
+with 401s until they clear cookies. The HMAC signing key already persists to
+STATE_DIR; this PR persists the session table using the same pattern.
+"""
+import importlib
+import json
+import os
+import sys
+import tempfile
+import time
+import unittest
+from pathlib import Path
+
+# Isolate state dir so tests never touch real sessions
+_TEST_STATE = Path(tempfile.mkdtemp())
+os.environ["HERMES_WEBUI_STATE_DIR"] = str(_TEST_STATE)
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+import api.auth as auth
+
+
+class TestSessionPersistence(unittest.TestCase):
+    """Sessions survive a simulated process restart (module reload)."""
+
+    def setUp(self) -> None:
+        auth._sessions.clear()
+        sessions_file = _TEST_STATE / '.sessions.json'
+        if sessions_file.exists():
+            sessions_file.unlink()
+
+    def _simulate_restart(self) -> None:
+        """Reload auth module to simulate a fresh process start."""
+        importlib.reload(auth)
+
+    def test_session_survives_restart(self) -> None:
+        """A session created before restart should still verify after reload."""
+        cookie = auth.create_session()
+        self.assertTrue(auth.verify_session(cookie))
+        self._simulate_restart()
+        self.assertTrue(auth.verify_session(cookie),
+                        "Session must survive process restart via persisted .sessions.json")
+
+    def test_invalidated_session_does_not_survive_restart(self) -> None:
+        """Invalidating a session must be reflected after reload."""
+        cookie = auth.create_session()
+        auth.invalidate_session(cookie)
+        self._simulate_restart()
+        self.assertFalse(auth.verify_session(cookie),
+                         "Invalidated session must not be reinstated after restart")
+
+    def test_expired_sessions_pruned_on_load(self) -> None:
+        """Sessions that expire between restarts must not be loaded."""
+        sessions_file = _TEST_STATE / '.sessions.json'
+        # Write a sessions file with one expired and one valid entry
+        now = time.time()
+        sessions_file.write_text(json.dumps({
+            "expired_token": now - 10,
+            "valid_token": now + 3600,
+        }))
+        self._simulate_restart()
+        self.assertNotIn("expired_token", auth._sessions)
+        self.assertIn("valid_token", auth._sessions)
+
+    def test_sessions_file_permissions(self) -> None:
+        """Sessions file must be owner-read-only (0600)."""
+        auth.create_session()
+        sessions_file = _TEST_STATE / '.sessions.json'
+        self.assertTrue(sessions_file.exists(), ".sessions.json was not created")
+        mode = oct(sessions_file.stat().st_mode & 0o777)
+        self.assertEqual(mode, oct(0o600),
+                         f".sessions.json permissions {mode} — expected 0o600")
+
+    def test_malformed_sessions_file_starts_fresh(self) -> None:
+        """A corrupt sessions file must not crash auth — start with empty dict."""
+        sessions_file = _TEST_STATE / '.sessions.json'
+        sessions_file.write_text("not valid json {{{{")
+        self._simulate_restart()
+        self.assertEqual(auth._sessions, {},
+                         "Corrupt sessions file must result in empty session dict")
+
+    def test_sessions_file_wrong_type_starts_fresh(self) -> None:
+        """A sessions file containing a non-dict must be ignored."""
+        sessions_file = _TEST_STATE / '.sessions.json'
+        sessions_file.write_text(json.dumps(["list", "not", "dict"]))
+        self._simulate_restart()
+        self.assertEqual(auth._sessions, {})
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_auth_sessions.py

@@ -0,0 +1,134 @@
+"""
+Tests for auth session lifecycle — session creation, verification, expiry,
+and lazy pruning of expired entries.
+"""
+import time
+import unittest
+from pathlib import Path
+import tempfile
+import os
+
+# Isolate state dir so we don't touch real sessions
+_TEST_STATE = Path(tempfile.mkdtemp())
+os.environ["HERMES_WEBUI_STATE_DIR"] = str(_TEST_STATE)
+
+import sys
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+import importlib
+
+# Force re-import of auth module so it picks up our TEST_STATE_DIR
+auth = importlib.import_module("api.auth")
+
+
+class TestSessionPruning(unittest.TestCase):
+    """Verify expired session cleanup works correctly."""
+
+    def setUp(self):
+        # Clear any leftover sessions from other tests
+        auth._sessions.clear()
+
+    def test_session_created_valid(self):
+        """A fresh session token should verify as valid."""
+        token = auth.create_session()
+        self.assertTrue(auth.verify_session(token))
+
+    def test_expired_session_pruned(self):
+        """Manually inserting an expired entry should be pruned on next verify_session call."""
+        # Insert sessions that have already expired
+        auth._sessions["fake_token"] = time.time() - 100
+        auth._sessions["another_fake"] = time.time() - 50
+        # Insert one valid session (far future)
+        auth._sessions["good_token"] = time.time() + 3600
+
+        # _sessions has 3 entries, 2 expired
+        self.assertEqual(len(auth._sessions), 3)
+
+        # Call verify_session — this triggers _prune_expired_sessions()
+        # Cookie format is token.signature, so we need a dot to pass the early check
+        auth.verify_session("fake_token.fake_sig")
+
+        # After verification, only the valid session should remain
+        self.assertEqual(len(auth._sessions), 1)
+        self.assertIn("good_token", auth._sessions)
+        self.assertNotIn("fake_token", auth._sessions)
+        self.assertNotIn("another_fake", auth._sessions)
+
+    def test_prune_does_not_remove_valid_sessions(self):
+        """_prune_expired_sessions should never remove sessions that are still active."""
+        auth._sessions["active_1"] = time.time() + 86400  # 24 hours from now
+        auth._sessions["active_2"] = time.time() + 7200    # 2 hours from now
+        auth._sessions["expired_1"] = time.time() - 10
+
+        auth._prune_expired_sessions()
+
+        self.assertEqual(len(auth._sessions), 2)
+        self.assertIn("active_1", auth._sessions)
+        self.assertIn("active_2", auth._sessions)
+        self.assertNotIn("expired_1", auth._sessions)
+
+    def test_verify_session_prunes_before_verification(self):
+        """verify_session should prune expired entries before checking the target token.
+
+        This ensures that _prune_expired_sessions() is called at the very top
+        of verify_session(), so cleanup happens on every auth check.
+        """
+        auth._sessions["expired_for_test"] = time.time() - 999
+
+        # verify_session with an invalid cookie triggers the full path:
+        # _prune_expired_sessions -> signature check -> return False
+        result = auth.verify_session("nonexistent.bad_sig")
+        self.assertFalse(result)
+
+        # The expired entry should have been cleaned up
+        self.assertNotIn("expired_for_test", auth._sessions)
+
+    def test_prune_handles_empty_dict(self):
+        """_prune_expired_sessions should be safe on an empty dict."""
+        auth._sessions.clear()
+        auth._prune_expired_sessions()
+        self.assertEqual(len(auth._sessions), 0)
+
+    def test_session_ttl_is_24_hours(self):
+        """Newly created sessions should have the expected 24-hour TTL."""
+        auth._sessions.clear()
+        token_hex = auth.create_session().split(".")[0]
+        # The _sessions dict stores token -> expiry_time
+        # We can check the expiry is approximately SESSION_TTL seconds from now
+        # by looking up the raw entry via the token
+        from api.auth import _sessions, SESSION_TTL
+        # find our entry
+        for t, exp in _sessions.items():
+            if t == token_hex:
+                # expiry should be within 5 seconds of now + SESSION_TTL
+                expected = time.time() + SESSION_TTL
+                self.assertAlmostEqual(exp, expected, delta=5)
+                break
+        else:
+            self.fail("Session token not found in _sessions")
+
+
+class TestSessionInvalidation(unittest.TestCase):
+    """Test session logout / invalidation."""
+
+    def setUp(self):
+        auth._sessions.clear()
+
+    def test_invalidate_session_removes_token(self):
+        """Calling invalidate_session should remove the token from _sessions."""
+        token = auth.create_session()
+        self.assertTrue(auth.verify_session(token))
+
+        auth.invalidate_session(token)
+        # Token should be gone
+        self.assertFalse(auth.verify_session(token))
+
+    def test_invalidate_unknown_token_is_safe(self):
+        """Invalidating a non-existent token should not raise."""
+        auth._sessions.clear()
+        auth.invalidate_session("nonexistent_token")
+        # Should not raise
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_background_tasks.py

@@ -0,0 +1,148 @@
+"""Regression tests for the /background task tracker.
+
+Covers two bugs caught in review of PR #932:
+
+1. `get_results()` was calling `_BACKGROUND_TASKS.pop(parent_sid, [])`, which
+   removed EVERY task (including still-running ones) on the first poll. Once
+   popped, `complete_background()` could no longer find the task to mark done,
+   so the final answer was silently lost.
+
+2. The `_handle_background` worker thread called `_run_agent_streaming` but
+   never invoked `complete_background()` after it returned. With no completion
+   hook, every background task stayed in `status="running"` forever —
+   `get_results()` filtered them out of its "done" list, and the user never
+   saw the result.
+
+These two bugs together made the `/background` command completely
+non-functional as originally shipped.  The fix in api/background.py +
+api/routes.py wires the completion hook and keeps running tasks in the
+tracker until they resolve.
+"""
+from __future__ import annotations
+
+import os
+import pathlib
+import sys
+import time
+import unittest
+from unittest.mock import patch
+
+
+# Ensure the repo root is importable without relying on CWD.
+REPO_ROOT = pathlib.Path(__file__).resolve().parent.parent
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+
+
+class TestGetResultsKeepsRunningTasks(unittest.TestCase):
+    """get_results() MUST NOT drop still-running tasks from _BACKGROUND_TASKS."""
+
+    def setUp(self):
+        import api.background as bg
+        bg._BACKGROUND_TASKS.clear()
+        self.bg = bg
+
+    def test_running_tasks_survive_get_results_call(self):
+        """A running task must remain in the tracker so complete_background()
+        can still find it after the first poll returns."""
+        parent = "parent-session-1"
+        self.bg.track_background(
+            parent_sid=parent, bg_sid="bg-a", stream_id="s-a",
+            task_id="task-a", prompt="long task",
+        )
+
+        # First poll: task is still running, no done results to return
+        results = self.bg.get_results(parent)
+        self.assertEqual(results, [], "no done tasks yet — nothing to return")
+
+        # The running task MUST still be tracked — otherwise the worker
+        # thread's complete_background call cannot find it.
+        remaining = self.bg.get_background_tasks(parent)
+        self.assertEqual(len(remaining), 1, (
+            "get_results dropped the still-running task — subsequent "
+            "complete_background() calls will silently no-op and the "
+            "result will be lost forever"
+        ))
+        self.assertEqual(remaining[0]["status"], "running")
+        self.assertEqual(remaining[0]["task_id"], "task-a")
+
+    def test_done_tasks_are_returned_and_removed(self):
+        """Done tasks are returned and popped; running tasks stay."""
+        parent = "parent-session-2"
+        self.bg.track_background(parent, "bg-done", "s-d", "task-done", "p1")
+        self.bg.track_background(parent, "bg-run", "s-r", "task-run", "p2")
+        self.bg.complete_background(parent, "task-done", "42")
+
+        results = self.bg.get_results(parent)
+        self.assertEqual(len(results), 1)
+        self.assertEqual(results[0]["task_id"], "task-done")
+        self.assertEqual(results[0]["answer"], "42")
+
+        # Done one is gone; running one is still tracked
+        remaining = self.bg.get_background_tasks(parent)
+        self.assertEqual(len(remaining), 1)
+        self.assertEqual(remaining[0]["task_id"], "task-run")
+        self.assertEqual(remaining[0]["status"], "running")
+
+    def test_complete_after_poll_still_reaches_tracker(self):
+        """Regression for the original bug: poll → complete → poll must surface
+        the result.  Before the fix, the first poll popped the running task and
+        complete_background()'s loop iterated over an empty list."""
+        parent = "parent-session-3"
+        self.bg.track_background(parent, "bg-x", "s-x", "task-x", "slow task")
+
+        # Frontend polls before the task finishes
+        first = self.bg.get_results(parent)
+        self.assertEqual(first, [])
+
+        # Worker thread finishes and calls complete_background
+        self.bg.complete_background(parent, "task-x", "answer!")
+
+        # Next poll must surface the answer
+        second = self.bg.get_results(parent)
+        self.assertEqual(len(second), 1)
+        self.assertEqual(second[0]["task_id"], "task-x")
+        self.assertEqual(second[0]["answer"], "answer!")
+
+    def test_empty_parent_is_cleaned_up(self):
+        """When all tasks are done and returned, the parent key is removed from the dict."""
+        parent = "parent-session-4"
+        self.bg.track_background(parent, "bg-1", "s-1", "task-1", "p")
+        self.bg.complete_background(parent, "task-1", "ok")
+        self.bg.get_results(parent)
+        self.assertNotIn(parent, self.bg._BACKGROUND_TASKS)
+
+
+class TestBackgroundCompletionHookWiring(unittest.TestCase):
+    """Static check: the _handle_background worker thread must call
+    complete_background() after _run_agent_streaming returns.  Without this,
+    running tasks stay forever-running and the user never sees the result.
+    """
+
+    def test_run_bg_and_notify_calls_complete_background(self):
+        """_handle_background must wrap _run_agent_streaming in a function
+        that subsequently invokes complete_background(parent_sid, task_id, answer)."""
+        routes_src = (REPO_ROOT / "api" / "routes.py").read_text(encoding="utf-8")
+        # Locate the _handle_background function
+        idx = routes_src.find("def _handle_background(")
+        self.assertGreater(idx, -1, "_handle_background() not found in routes.py")
+        # Take a generous window around the function body
+        end = routes_src.find("\ndef ", idx + 1)
+        body = routes_src[idx:end if end > 0 else idx + 3000]
+
+        self.assertIn("complete_background", body, (
+            "_handle_background worker must call complete_background() after "
+            "_run_agent_streaming returns — otherwise the tracker never "
+            "transitions the task to status='done' and /api/background/status "
+            "returns nothing forever. See api/background.py:complete_background."
+        ))
+        # Must extract the last assistant message content from the bg session
+        self.assertIn("_run_agent_streaming", body)
+        self.assertIn("Session.load", body, (
+            "_run_bg_and_notify must reload the bg session to extract the "
+            "final assistant reply so complete_background gets an actual answer"
+        ))
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_batch_fixes.py

@@ -0,0 +1,235 @@
+"""Tests for the batch of fixes from PRs #506-#521 (v0.50.47).
+
+Covers:
+  - /root workspace unblocking (#510/#521)
+  - Attached-files split guard (#521)
+  - custom_providers model visibility (#515/#519)
+  - Cron skill cache invalidation (#507/#508)
+  - System (auto) theme (#504/#506/#509/#514)
+"""
+
+import pathlib
+import re
+
+REPO = pathlib.Path(__file__).parent.parent
+
+
+def read(rel):
+    return (REPO / rel).read_text()
+
+
+# ── Group A: /root workspace ──────────────────────────────────────────────────
+
+class TestRootWorkspaceUnblocked:
+
+    def test_root_not_in_blocked_system_roots(self):
+        src = read("api/workspace.py")
+        assert "Path('/root')" not in src, (
+            "/root must not be in _BLOCKED_SYSTEM_ROOTS — "
+            "breaks deployments where Hermes runs as root"
+        )
+
+    def test_etc_still_blocked(self):
+        """Sanity: other dangerous paths remain blocked."""
+        src = read("api/workspace.py")
+        assert "Path('/etc')" in src
+        assert "Path('/proc')" in src
+
+    def test_split_guard_present(self):
+        src = read("api/streaming.py")
+        assert "'\\n\\n[Attached files:' in msg_text" in src, (
+            "base_text split must guard against missing '[Attached files:' "
+            "to avoid empty-string on plain messages"
+        )
+
+
+# ── Group B: custom_providers visibility ─────────────────────────────────────
+
+class TestCustomProvidersVisibility:
+
+    def test_has_custom_providers_variable_present(self):
+        src = read("api/config.py")
+        assert "_has_custom_providers" in src, (
+            "_has_custom_providers variable must exist in get_available_models()"
+        )
+
+    def test_discard_custom_conditional_on_no_custom_providers(self):
+        src = read("api/config.py")
+        assert "not _has_custom_providers" in src, (
+            "detected_providers.discard('custom') must be gated on "
+            "'not _has_custom_providers'"
+        )
+
+    def test_custom_providers_isinstance_check(self):
+        src = read("api/config.py")
+        assert "isinstance(_custom_providers_cfg, list)" in src, (
+            "_has_custom_providers must check isinstance(..., list)"
+        )
+
+
+# ── Group C: cron skill cache ─────────────────────────────────────────────────
+
+class TestCronSkillCacheInvalidation:
+
+    def _panels_src(self):
+        return read("static/panels.js")
+
+    def test_cache_busted_on_form_open(self):
+        src = self._panels_src()
+        # toggleCronForm should set cache to null unconditionally
+        # openCronCreate() opens the task create form (renamed from toggleCronForm
+        # in the main-view refactor). It must null the skills cache before fetching.
+        m = re.search(
+            r'function openCronCreate\(\)\{.*?_cronSkillsCache\s*=\s*null',
+            src, re.DOTALL
+        )
+        assert m, (
+            "openCronCreate must unconditionally null _cronSkillsCache "
+            "before fetching skills"
+        )
+
+    def test_cache_not_guarded_by_if_on_open(self):
+        src = self._panels_src()
+        # openCronCreate must not gate the fetch behind an if(!_cronSkillsCache) guard.
+        m = re.search(
+            r'function openCronCreate\(\)\{.*?\}',
+            src, re.DOTALL
+        )
+        assert m, "openCronCreate definition not found"
+        assert "if(!_cronSkillsCache)" not in m.group(0), (
+            "openCronCreate should not use 'if(!_cronSkillsCache)' guard — "
+            "cache must always be busted on open"
+        )
+
+    def test_cache_busted_on_skill_save(self):
+        src = self._panels_src()
+        # saveSkillForm() is the handler invoked on skill save (renamed from
+        # submitSkillSave in the main-view refactor; the old name still aliases it).
+        m = re.search(
+            r'async function saveSkillForm\(\).*?_skillsData\s*=\s*null.*?_cronSkillsCache\s*=\s*null',
+            src, re.DOTALL
+        )
+        assert m, (
+            "_cronSkillsCache must be set to null in saveSkillForm() "
+            "right after _skillsData = null"
+        )
+
+
+# ── Group D: System (auto) theme ──────────────────────────────────────────────
+
+class TestSystemTheme:
+
+    def test_apply_theme_helper_in_boot_js(self):
+        src = read("static/boot.js")
+        assert "function _applyTheme(" in src, (
+            "_applyTheme helper function must be defined in boot.js"
+        )
+
+    def test_apply_theme_resolves_system(self):
+        src = read("static/boot.js")
+        assert "normalized.theme==='system'" in src or "=== 'system'" in src, (
+            "_applyTheme must branch on 'system' to resolve via matchMedia"
+        )
+
+    def test_apply_theme_uses_matchmedia(self):
+        src = read("static/boot.js")
+        assert "prefers-color-scheme" in src, (
+            "_applyTheme must use matchMedia('(prefers-color-scheme:dark)')"
+        )
+
+    def test_load_settings_calls_apply_theme(self):
+        src = read("static/boot.js")
+        assert "_applyTheme(appearance.theme)" in src, (
+            "loadSettings must call _applyTheme() instead of direct data-theme assignment"
+        )
+
+    def test_system_option_in_theme_picker(self):
+        html = read("static/index.html")
+        assert "_pickTheme('system')" in html, (
+            "Theme picker must include a system theme button"
+        )
+        assert ">System<" in html, (
+            "Theme picker must show 'System' label"
+        )
+
+    def test_theme_picker_uses_pick_theme(self):
+        html = read("static/index.html")
+        assert "_pickTheme(" in html, (
+            "Theme buttons must call _pickTheme()"
+        )
+
+    def test_flicker_script_resolves_system(self):
+        html = read("static/index.html")
+        # The head flicker-prevention IIFE must handle 'system'
+        assert "==='system'" in html or "=== 'system'" in html, (
+            "Flicker-prevention head script must resolve 'system' before setting data-theme"
+        )
+        assert "legacy={slate:['dark','slate']" in html, (
+            "Flicker-prevention head script must normalize legacy theme names on first paint"
+        )
+
+    def test_system_in_commands_themes_list(self):
+        src = read("static/commands.js")
+        assert "'system'" in src, (
+            "/theme command must include 'system' in the valid themes array"
+        )
+
+    def test_commands_uses_apply_theme(self):
+        src = read("static/commands.js")
+        assert "_applyTheme(appearance.theme)" in src, (
+            "cmdTheme must call _applyTheme() with the normalized canonical theme"
+        )
+
+    def test_commands_accept_legacy_theme_aliases(self):
+        src = read("static/commands.js")
+        assert "const legacyThemes=Object.keys(_LEGACY_THEME_MAP||{});" in src, (
+            "cmdTheme must accept legacy theme aliases and map them onto canonical appearance values"
+        )
+
+    def test_panels_reverts_via_apply_theme(self):
+        src = read("static/panels.js")
+        assert "_applyTheme(_settingsThemeOnOpen)" in src or \
+               "_applyTheme(" in src, (
+            "_revertSettingsPreview must call _applyTheme() so 'system' "
+            "is correctly re-activated on settings discard"
+        )
+
+    def test_panels_saves_system_string_not_resolved(self):
+        src = read("static/panels.js")
+        assert "localStorage.getItem('hermes-theme')" in src, (
+            "_settingsThemeOnOpen must read from localStorage to preserve "
+            "the 'system' string, not the resolved 'dark'/'light'"
+        )
+
+    def test_i18n_cmd_theme_includes_system_english(self):
+        src = read("static/i18n.js")
+        assert "system/dark/light" in src, (
+            "English cmd_theme i18n key must include 'system' in the theme list"
+        )
+
+    def test_i18n_cmd_theme_all_locales(self):
+        src = read("static/i18n.js")
+        count = src.count("system/dark/light")
+        assert count >= 5, (
+            f"cmd_theme description should mention 'system' in all 5 locales; "
+            f"found {count}"
+        )
+
+    def test_theme_listener_cleanup_uses_stable_handler(self):
+        src = read("static/boot.js")
+        assert "_systemThemeMq&&_onSystemThemeChange" in src, (
+            "_applyTheme must track the active OS-theme listener so it can be removed cleanly"
+        )
+        assert "removeEventListener('change',_onSystemThemeChange)" in src, (
+            "_applyTheme must remove the previous OS-theme listener before adding a new one"
+        )
+
+    def test_panels_hydrates_appearance_before_models_fetch(self):
+        src = read("static/panels.js")
+        skin_idx = src.index("const skinVal=(settings.skin||'default').toLowerCase();")
+        # models is now declared as let models=null before the try block
+        models_idx = src.index("models=await api('/api/models');")
+        assert skin_idx < models_idx, (
+            "loadSettingsPanel must hydrate theme/skin before awaiting /api/models, "
+            "otherwise a slow model fetch can clobber an in-progress skin selection"
+        )

tests/test_bootstrap_dotenv.py

@@ -0,0 +1,187 @@
+"""
+Tests for bootstrap.py .env loading fix (issue #730).
+
+bootstrap.py is the primary documented entry point ("python3 bootstrap.py").
+Previously it did not load REPO_ROOT/.env, so HERMES_WEBUI_HOST, HERMES_WEBUI_PORT
+etc. were silently ignored when launching without start.sh.
+
+Covers:
+  1. _load_repo_dotenv() sets env vars from a repo .env file
+  2. _load_repo_dotenv() ignores commented lines and blank lines
+  3. _load_repo_dotenv() strips quotes from values
+  4. _load_repo_dotenv() is a no-op when .env does not exist
+  5. _load_repo_dotenv() prints a warning (not crash) on unreadable .env
+  6. _load_repo_dotenv() overwrites existing env vars (shell source semantics)
+  7. _load_repo_dotenv() handles 'export FOO=bar' prefix
+  8. _load_repo_dotenv() preserves values containing '='
+  9. Variables are set unconditionally (not setdefault)
+  10. Structural: loader is called before DEFAULT_HOST/DEFAULT_PORT
+"""
+import os
+import sys
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+REPO_ROOT = Path(__file__).parent.parent
+
+
+class TestLoadRepoDotenv:
+
+    def setup_method(self):
+        self._saved_env = os.environ.copy()
+
+    def teardown_method(self):
+        os.environ.clear()
+        os.environ.update(self._saved_env)
+
+    def _run(self, tmp_path, env_content: str):
+        """Write .env to tmp_path and run _load_repo_dotenv() with that root."""
+        import bootstrap as bs
+        (tmp_path / ".env").write_text(env_content, encoding="utf-8")
+        orig_root = bs.REPO_ROOT
+        try:
+            bs.REPO_ROOT = tmp_path
+            bs._load_repo_dotenv()
+        finally:
+            bs.REPO_ROOT = orig_root
+
+    def test_sets_env_var_from_dotenv(self, tmp_path):
+        """Basic key=value is loaded into os.environ."""
+        self._run(tmp_path, "HERMES_WEBUI_HOST=0.0.0.0\n")
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0"
+
+    def test_sets_port_from_dotenv(self, tmp_path):
+        """HERMES_WEBUI_PORT is loaded as a string (caller does int() conversion)."""
+        self._run(tmp_path, "HERMES_WEBUI_PORT=18787\n")
+        assert os.environ.get("HERMES_WEBUI_PORT") == "18787"
+
+    def test_ignores_comment_lines(self, tmp_path):
+        """Lines starting with # are not loaded."""
+        os.environ.pop("HERMES_WEBUI_HOST", None)
+        self._run(tmp_path, "# HERMES_WEBUI_HOST=should-be-ignored\n")
+        assert os.environ.get("HERMES_WEBUI_HOST") is None
+
+    def test_ignores_blank_lines(self, tmp_path):
+        """Blank lines are silently skipped without error."""
+        self._run(tmp_path, "\n\nHERMES_WEBUI_PORT=9000\n\n")
+        assert os.environ.get("HERMES_WEBUI_PORT") == "9000"
+
+    def test_strips_double_quoted_values(self, tmp_path):
+        """Values wrapped in double quotes are stripped."""
+        self._run(tmp_path, 'HERMES_WEBUI_HOST="0.0.0.0"\n')
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0"
+
+    def test_strips_single_quoted_values(self, tmp_path):
+        """Values wrapped in single quotes are stripped."""
+        self._run(tmp_path, "HERMES_WEBUI_HOST='0.0.0.0'\n")
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0"
+
+    def test_noop_when_no_dotenv(self, tmp_path):
+        """No .env file — function returns silently without error."""
+        import bootstrap as bs
+        orig = bs.REPO_ROOT
+        try:
+            bs.REPO_ROOT = tmp_path  # tmp_path has no .env
+            bs._load_repo_dotenv()  # must not raise
+        finally:
+            bs.REPO_ROOT = orig
+
+    def test_noop_when_dotenv_unreadable(self, tmp_path, capsys):
+        """Unreadable .env prints a warning to stderr — does not crash."""
+        import bootstrap as bs
+        env_path = tmp_path / ".env"
+        env_path.write_text("HERMES_WEBUI_PORT=9999\n")
+        orig = bs.REPO_ROOT
+        try:
+            bs.REPO_ROOT = tmp_path
+            with patch("pathlib.Path.read_text", side_effect=PermissionError("no access")):
+                bs._load_repo_dotenv()  # must not raise
+        finally:
+            bs.REPO_ROOT = orig
+        captured = capsys.readouterr()
+        assert "bootstrap" in captured.err.lower() or "warning" in captured.err.lower() or \
+               "could not load" in captured.err.lower(), (
+            "_load_repo_dotenv() should print a warning to stderr on read failure"
+        )
+
+    def test_overwrites_existing_env_var(self, tmp_path):
+        """Unconditional overwrite matches shell source semantics."""
+        os.environ["HERMES_WEBUI_HOST"] = "127.0.0.1"
+        self._run(tmp_path, "HERMES_WEBUI_HOST=0.0.0.0\n")
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0"
+
+    def test_does_not_set_empty_values(self, tmp_path):
+        """A key whose value is empty after stripping is not set to a non-empty string."""
+        os.environ.pop("HERMES_EMPTY_KEY", None)
+        self._run(tmp_path, 'HERMES_EMPTY_KEY=""\n')
+        # The current implementation sets key to "" (empty string) — verify it is
+        # not set to a non-empty string, which would be clearly wrong.
+        val = os.environ.get("HERMES_EMPTY_KEY")
+        assert val != "something-wrong", "Empty-value key must not be set to a non-empty string"
+        # Specifically: empty string or absent are both acceptable behaviours.
+        assert val in (None, ""), f"Unexpected value for empty-quoted key: {val!r}"
+
+    def test_multiple_keys_all_loaded(self, tmp_path):
+        """Multiple key=value pairs in one file are all loaded."""
+        content = "HERMES_WEBUI_HOST=0.0.0.0\nHERMES_WEBUI_PORT=18787\n"
+        self._run(tmp_path, content)
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0"
+        assert os.environ.get("HERMES_WEBUI_PORT") == "18787"
+
+    def test_value_with_equals_sign_preserved(self, tmp_path):
+        """Values containing '=' (e.g. base64) are preserved correctly."""
+        self._run(tmp_path, "MY_KEY=abc=def==\n")
+        assert os.environ.get("MY_KEY") == "abc=def=="
+
+    def test_export_prefix_stripped(self, tmp_path):
+        """'export FOO=bar' lines are parsed correctly — export prefix is stripped."""
+        self._run(tmp_path, "export HERMES_WEBUI_HOST=0.0.0.0\n")
+        assert os.environ.get("HERMES_WEBUI_HOST") == "0.0.0.0", (
+            "'export KEY=value' lines must set KEY, not 'export KEY'"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Structural tests — confirm the fix is in place
+# ---------------------------------------------------------------------------
+
+class TestBootstrapStructure:
+
+    def test_load_repo_dotenv_function_exists(self):
+        """bootstrap.py must export _load_repo_dotenv()."""
+        import bootstrap as bs
+        assert callable(getattr(bs, "_load_repo_dotenv", None)), (
+            "bootstrap.py must define _load_repo_dotenv() so that "
+            "python3 bootstrap.py loads REPO_ROOT/.env before reading env defaults"
+        )
+
+    def test_dotenv_loaded_before_default_host_port(self):
+        """_load_repo_dotenv() call must appear before DEFAULT_HOST/DEFAULT_PORT in source."""
+        src = (REPO_ROOT / "bootstrap.py").read_text(encoding="utf-8")
+        load_pos = src.find("_load_repo_dotenv()")
+        host_pos = src.find("DEFAULT_HOST")
+        port_pos = src.find("DEFAULT_PORT")
+        assert load_pos != -1, "_load_repo_dotenv() call not found in bootstrap.py"
+        assert load_pos < host_pos, (
+            "_load_repo_dotenv() must be called before DEFAULT_HOST assignment "
+            "so that HERMES_WEBUI_HOST from .env is picked up"
+        )
+        assert load_pos < port_pos, (
+            "_load_repo_dotenv() must be called before DEFAULT_PORT assignment "
+            "so that HERMES_WEBUI_PORT from .env is picked up"
+        )
+
+    def test_start_sh_and_bootstrap_equivalent_env_loading(self):
+        """start.sh sources .env before bootstrap.py; bootstrap.py must now do the same."""
+        start_sh = (REPO_ROOT / "start.sh").read_text(encoding="utf-8")
+        bootstrap_src = (REPO_ROOT / "bootstrap.py").read_text(encoding="utf-8")
+        # start.sh sources .env
+        assert "source" in start_sh and ".env" in start_sh, (
+            "start.sh should still source .env (regression guard)"
+        )
+        # bootstrap.py now loads it too
+        assert "_load_repo_dotenv" in bootstrap_src, (
+            "bootstrap.py must load .env so direct invocation matches start.sh behaviour"
+        )

tests/test_bugbatch_apr2026.py

@@ -0,0 +1,189 @@
+"""
+Bug batch fixes — April 2026.
+
+Covers:
+- #594: .app-dialog and .file-rename-input have light theme overrides in style.css
+- #576: workspace panel localStorage restore is gated on session.workspace presence (boot.js)
+- #585: get_available_models() calls reload_config() before reading config cache
+- #567: docker-compose.yml comment mentions macOS UID mismatch
+- #590: _transcribeBlob already calls setComposerStatus('Transcribing…') — confirmed present
+"""
+import pathlib
+import re
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent
+STYLE_CSS = (REPO_ROOT / "static" / "style.css").read_text(encoding="utf-8")
+BOOT_JS   = (REPO_ROOT / "static" / "boot.js").read_text(encoding="utf-8")
+COMPOSE   = (REPO_ROOT / "docker-compose.yml").read_text(encoding="utf-8")
+
+
+# ── #594: light theme dialog overrides ───────────────────────────────────────
+
+def test_594_app_dialog_has_light_mode_override():
+    """style.css must have a light mode rule targeting .app-dialog background."""
+    assert ':root:not(.dark) .app-dialog{' in STYLE_CSS, (
+        "Missing light mode override for .app-dialog — dialogs appear dark on light theme"
+    )
+
+
+def test_594_app_dialog_input_has_light_mode_override():
+    """style.css must have a light mode rule for .app-dialog-input."""
+    assert ":root:not(.dark) .app-dialog-input{" in STYLE_CSS, (
+        "Missing light mode override for .app-dialog-input"
+    )
+
+
+def test_594_app_dialog_btn_has_light_mode_override():
+    """style.css must have a light mode rule for .app-dialog-btn."""
+    assert ":root:not(.dark) .app-dialog-btn{" in STYLE_CSS, (
+        "Missing light mode override for .app-dialog-btn"
+    )
+
+
+def test_594_app_dialog_close_has_light_mode_override():
+    """style.css must have a light mode rule for .app-dialog-close."""
+    assert ":root:not(.dark) .app-dialog-close{" in STYLE_CSS, (
+        "Missing light mode override for .app-dialog-close"
+    )
+
+
+def test_594_file_rename_input_has_light_mode_override():
+    """style.css must have a light mode rule for .file-rename-input."""
+    assert ":root:not(.dark) .file-rename-input{" in STYLE_CSS, (
+        "Missing light mode override for .file-rename-input"
+    )
+
+
+# ── dark-mode user bubble semantics ──────────────────────────────────────────
+
+def test_dark_user_bubbles_use_dark_tinted_surface():
+    """Dark mode should keep user bubbles dark, with skin only tinting the bubble."""
+    assert "--user-bubble-bg: var(--accent-bg-strong);" in STYLE_CSS, (
+        "Dark mode user bubbles should use the dark accent tint, not the full bright accent fill"
+    )
+    assert "--user-bubble-border: var(--accent-bg-strong);" in STYLE_CSS, (
+        "Dark mode user bubble borders should match the quieter thinking-card border intensity"
+    )
+    assert "--user-bubble-text: var(--text);" in STYLE_CSS, (
+        "Dark mode user bubble text should inherit the theme text color"
+    )
+
+
+def test_dark_user_bubbles_do_not_need_per_skin_text_hacks():
+    """Dark-mode user bubble contrast should not rely on per-skin text overrides."""
+    assert re.search(r':root\.dark\[data-skin="[^"]+"\]\s*\{\s*--user-bubble-text:', STYLE_CSS) is None, (
+        "Dark-mode user bubble contrast should come from shared theme tokens, not per-skin text hacks"
+    )
+
+
+def test_user_bubbles_define_selection_tokens_for_both_modes():
+    """User bubbles need dedicated selection colors so selected text remains readable."""
+    assert "--user-selection-bg: rgba(0,0,0,.22);" in STYLE_CSS, (
+        "Light-mode user bubbles should define a darker selection fill for contrast"
+    )
+    assert "--user-selection-bg: rgba(255,255,255,.18);" in STYLE_CSS, (
+        "Dark-mode user bubbles should define a lighter selection fill for contrast"
+    )
+    assert "--user-selection-text: #fff;" in STYLE_CSS, (
+        "Light-mode user bubble selection should preserve readable text color"
+    )
+
+
+def test_user_bubble_selection_is_scoped_to_user_message_body():
+    """Selection override must apply only to user bubbles, including nested markdown nodes."""
+    assert '.msg-row[data-role="user"] .msg-body::selection,' in STYLE_CSS, (
+        "Missing selection override on the user message bubble"
+    )
+    assert '.msg-row[data-role="user"] .msg-body *::selection {' in STYLE_CSS, (
+        "Nested elements inside user messages must inherit the same selection colors"
+    )
+
+
+# ── #576: workspace panel snap fix ───────────────────────────────────────────
+
+def test_576_panel_restore_gated_on_workspace():
+    """boot.js: localStorage panel restore must be gated on session.workspace."""
+    # The guard must appear: session.workspace check before _workspacePanelMode='browse'
+    # Panel pref key takes priority over runtime key (toolbar close must not clear preference)
+    assert "S.session&&S.session.workspace&&panelPref" in BOOT_JS, (
+        "Workspace panel localStorage restore must be gated on S.session.workspace "
+        "to prevent snap-open-then-closed on sessions without a workspace (#576)"
+    )
+    assert "'hermes-webui-workspace-panel-pref'" in BOOT_JS, (
+        "Panel restore must check the preference key so toolbar close does not clear it"
+    )
+
+
+def test_576_restore_happens_after_load_session():
+    """boot.js: loadSession() must come before the panel restore guard."""
+    load_pos    = BOOT_JS.find("await loadSession(saved)")
+    restore_pos = BOOT_JS.find("panelPref")
+    assert load_pos != -1, "loadSession call not found in boot.js"
+    assert restore_pos != -1, "workspace panel restore guard not found"
+    assert load_pos < restore_pos, (
+        "loadSession() must run before the panel restore guard "
+        "so S.session.workspace is known at restore time"
+    )
+
+
+# ── #585: get_available_models reloads config ─────────────────────────────────
+
+def test_585_get_available_models_calls_reload_config():
+    """api/config.py: get_available_models() must do a mtime-based reload check."""
+    config_src = (REPO_ROOT / "api" / "config.py").read_text(encoding="utf-8")
+    fn_start = config_src.find("def get_available_models()")
+    assert fn_start != -1, "get_available_models not found"
+    fn_body_end = config_src.find('"""', config_src.find('"""', fn_start + 30) + 3) + 3
+    # Must check mtime before reading config
+    mtime_pos    = config_src.find("_current_mtime", fn_body_end)
+    active_prov_pos = config_src.find("active_provider = None", fn_body_end)
+    assert mtime_pos != -1, (
+        "get_available_models() must check config file mtime before reading cache (#585)"
+    )
+    assert mtime_pos < active_prov_pos, (
+        "mtime check must come before active_provider = None in get_available_models()"
+    )
+
+
+# ── #567: docker-compose UID note ─────────────────────────────────────────────
+
+def test_567_compose_mentions_macos_uid():
+    """docker-compose.yml must mention macOS UID / id -u to help macOS users."""
+    assert "macOS" in COMPOSE or "macos" in COMPOSE.lower(), (
+        "docker-compose.yml should mention macOS UID issue (#567)"
+    )
+    assert "id -u" in COMPOSE, (
+        "docker-compose.yml should tell users to run 'id -u' to find their UID (#567)"
+    )
+
+
+# ── #590: transcription spinner already present ───────────────────────────────
+
+def test_590_transcribing_status_shown_before_fetch():
+    """boot.js: setComposerStatus('Transcribing…') must fire before the fetch call."""
+    transcribe_fn_start = BOOT_JS.find("async function _transcribeBlob(")
+    assert transcribe_fn_start != -1, "_transcribeBlob not found in boot.js"
+    fn_body = BOOT_JS[transcribe_fn_start:transcribe_fn_start + 600]
+    status_pos = fn_body.find("setComposerStatus('Transcribing")
+    fetch_pos  = fn_body.find("await fetch(")
+    assert status_pos != -1, (
+        "setComposerStatus('Transcribing…') must be called before the fetch in _transcribeBlob"
+    )
+    assert fetch_pos != -1, "await fetch not found in _transcribeBlob"
+    assert status_pos < fetch_pos, (
+        "setComposerStatus('Transcribing…') must appear before 'await fetch' "
+        "so the UI shows a spinner immediately on stop (#590)"
+    )
+
+
+def test_590_recording_stops_before_transcribe():
+    """boot.js: _setRecording(false) must fire in onstop before _transcribeBlob."""
+    onstop_start = BOOT_JS.find("mediaRecorder.onstop")
+    assert onstop_start != -1, "mediaRecorder.onstop not found"
+    onstop_body = BOOT_JS[onstop_start:onstop_start + 400]
+    rec_pos = onstop_body.find("_setRecording(false)")
+    blob_pos = onstop_body.find("_transcribeBlob(")
+    assert rec_pos != -1 and blob_pos != -1
+    assert rec_pos < blob_pos, (
+        "_setRecording(false) must come before _transcribeBlob so mic icon clears immediately"
+    )

tests/test_byok_model_dropdown.py

@@ -0,0 +1,381 @@
+"""Tests for #815 — BYOK/custom provider models missing from WebUI model dropdown.
+
+Root causes fixed:
+  1. active_provider alias not normalized in get_available_models()
+     ('z.ai' -> 'zai', 'x.ai' -> 'xai', 'google' -> 'gemini', etc.)
+     causing the provider to fall to the 'else/unknown' branch with no models.
+
+  2. /api/models/live didn't normalize the provider query param, so
+     provider_model_ids() received the un-aliased form and returned [].
+
+  3. /api/models/live returned empty for provider='custom' even when
+     custom_providers entries exist in config.yaml — the live enrichment
+     step never added those models.
+"""
+import pathlib
+import re
+import sys
+import unittest.mock as mock
+
+import pytest
+
+REPO = pathlib.Path(__file__).parent.parent
+sys.path.insert(0, str(REPO))
+sys.path.insert(0, str(REPO.parent / ".hermes" / "hermes-agent"))
+
+
+def read(rel):
+    return (REPO / rel).read_text(encoding="utf-8")
+
+
+@pytest.fixture(autouse=True)
+def _isolate_models_cache():
+    """Invalidate the TTL model cache before AND after every test.
+
+    ``get_available_models()`` caches its result keyed on config.yaml mtime.
+    Tests in this file repoint ``_get_config_path`` to a tmp_path, populate
+    the cache there, then let monkeypatch restore the original path.  The
+    cache, keyed on the tmp_path's mtime, then poisons downstream tests
+    (e.g. test_model_resolver) which see stale data and never hit their
+    mocks.  Clearing the cache around each test breaks that linkage.
+    """
+    import api.config as c
+    try:
+        c.invalidate_models_cache()
+    except Exception:
+        pass
+    yield
+    try:
+        c.invalidate_models_cache()
+    except Exception:
+        pass
+
+
+# ── api/config.py — active_provider normalization ─────────────────────────────
+
+class TestActiveProviderNormalization:
+    """get_available_models() must normalize active_provider aliases before lookup."""
+
+    def _run(self, tmp_path, provider_str, monkeypatch):
+        """Return get_available_models() output for a given provider string."""
+        import api.config as c
+
+        cfgfile = tmp_path / "config.yaml"
+        cfgfile.write_text(
+            f"model:\n  provider: {provider_str}\n  default: test-model\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(c, "_get_config_path", lambda: cfgfile)
+        c.reload_config()
+        # Patch list_available_providers to avoid real network calls
+        fake_prov = mock.MagicMock()
+        fake_prov.return_value = []
+        try:
+            import hermes_cli.models as hm
+            monkeypatch.setattr(hm, "list_available_providers", fake_prov)
+        except Exception:
+            pass
+        result = c.get_available_models()
+        c.reload_config()
+        return result
+
+    def test_z_dot_ai_normalized_to_zai(self, tmp_path, monkeypatch):
+        result = self._run(tmp_path, "z.ai", monkeypatch)
+        # active_provider returned to browser must be canonical 'zai' or
+        # at minimum must not be 'z.ai' (which would miss the _PROVIDER_MODELS lookup)
+        ap = result.get("active_provider", "")
+        assert ap in ("zai", ""), f"active_provider should be 'zai', got {ap!r}"
+
+    def test_x_dot_ai_normalized_to_xai(self, tmp_path, monkeypatch):
+        result = self._run(tmp_path, "x.ai", monkeypatch)
+        ap = result.get("active_provider", "")
+        assert ap in ("xai", ""), f"active_provider should be 'xai', got {ap!r}"
+
+    def test_google_normalized_to_gemini(self, tmp_path, monkeypatch):
+        result = self._run(tmp_path, "google", monkeypatch)
+        ap = result.get("active_provider", "")
+        assert ap in ("gemini", ""), f"active_provider should be 'gemini', got {ap!r}"
+
+    def test_normalization_code_present(self):
+        """Source-level check: config.py must call _PROVIDER_ALIASES for active_provider."""
+        src = read("api/config.py")
+        # Must alias-normalize active_provider before the group-builder runs
+        assert "_PROVIDER_ALIASES" in src, (
+            "api/config.py must import _PROVIDER_ALIASES to normalize active_provider"
+        )
+        # The normalization must happen before the group builder loop
+        alias_pos = src.index("_PROVIDER_ALIASES")
+        group_builder_pos = src.index("for pid in sorted(detected_providers)")
+        assert alias_pos < group_builder_pos, (
+            "active_provider normalization must occur before the group-builder loop"
+        )
+
+
+# ── api/routes.py — /api/models/live provider normalization ───────────────────
+
+class TestLiveModelsProviderNormalization:
+    """_handle_live_models must normalize the provider query param."""
+
+    def test_live_models_normalizes_provider_alias(self):
+        src = read("api/routes.py")
+        # Find _handle_live_models function
+        m = re.search(
+            r"def _handle_live_models\(.*?\ndef ",
+            src,
+            re.DOTALL,
+        )
+        assert m, "_handle_live_models not found"
+        fn = m.group(0)
+        assert "_resolve_provider_alias" in fn, (
+            "_handle_live_models must normalize provider via "
+            "api.config._resolve_provider_alias so 'z.ai' -> 'zai' "
+            "before calling provider_model_ids()"
+        )
+
+    def test_live_models_normalization_before_provider_model_ids(self):
+        """Normalization call must appear before the provider_model_ids call site."""
+        src = read("api/routes.py")
+        alias_match = re.search(
+            r"provider\s*=\s*_resolve_provider_alias\(provider\)",
+            src,
+        )
+        pmi_call_match = re.search(
+            r"ids\s*=\s*_pmi\(provider\)",
+            src,
+        )
+        assert alias_match, "_resolve_provider_alias call not found in routes.py"
+        assert pmi_call_match, "ids = _pmi(provider) call not found"
+        assert alias_match.start() < pmi_call_match.start(), (
+            "alias normalization must occur before ids = _pmi(provider)"
+        )
+
+    def test_alias_resolver_works_without_hermes_cli(self):
+        """Normalization must work even when hermes_cli is not importable —
+        CI and installs without the agent cloned alongside the WebUI.
+        The WebUI ships its own _PROVIDER_ALIASES table; the agent's table
+        is merged only when available."""
+        import api.config as c
+        # Core CLI aliases from #815's bug report
+        assert c._resolve_provider_alias('z.ai') == 'zai'
+        assert c._resolve_provider_alias('x.ai') == 'xai'
+        assert c._resolve_provider_alias('google') == 'gemini'
+        assert c._resolve_provider_alias('grok') == 'xai'
+        # Case / whitespace insensitive
+        assert c._resolve_provider_alias('  Z.AI  ') == 'zai'
+        # Canonical names pass through unchanged
+        assert c._resolve_provider_alias('openrouter') == 'openrouter'
+        assert c._resolve_provider_alias('anthropic') == 'anthropic'
+        assert c._resolve_provider_alias('custom') == 'custom'
+        # Empty / None pass through
+        assert c._resolve_provider_alias('') == ''
+        assert c._resolve_provider_alias(None) is None
+
+
+# ── api/routes.py — /api/models/live custom_providers fallback ────────────────
+
+class TestLiveModelsCustomProviderFallback:
+    """When provider='custom' and provider_model_ids() returns [],
+    /api/models/live must fall back to custom_providers entries from config.yaml."""
+
+    def test_custom_fallback_code_present(self):
+        src = read("api/routes.py")
+        m = re.search(
+            r"def _handle_live_models\(.*?\ndef ",
+            src,
+            re.DOTALL,
+        )
+        assert m, "_handle_live_models not found"
+        fn = m.group(0)
+        assert "custom_providers" in fn, (
+            "_handle_live_models must read custom_providers from config "
+            "as fallback when provider='custom' and provider_model_ids() returns []"
+        )
+        assert 'provider == "custom"' in fn or "provider=='custom'" in fn, (
+            "_handle_live_models must check provider == 'custom' before fallback"
+        )
+
+    def test_custom_fallback_returns_configured_models(self, tmp_path, monkeypatch):
+        """End-to-end: /api/models/live?provider=custom returns custom_providers models."""
+        import api.config as c
+        import api.routes as r
+
+        cfgfile = tmp_path / "config.yaml"
+        cfgfile.write_text(
+            "model:\n  provider: custom\n  default: my-byok-model\n"
+            "custom_providers:\n"
+            "  - model: my-byok-model\n"
+            "    api_base: https://my-llm.example.com/v1\n"
+            "    api_key: sk-test\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(c, "_get_config_path", lambda: cfgfile)
+        c.reload_config()
+
+        # Mock handler and parsed URL
+        handler = mock.MagicMock()
+        responses = []
+        def fake_j(h, data, **kw):
+            responses.append(data)
+            return True
+        monkeypatch.setattr(r, "j", fake_j)
+
+        from urllib.parse import urlparse
+        parsed = mock.MagicMock()
+        parsed.query = "provider=custom"
+
+        # Mock provider_model_ids to return [] (simulating no live endpoint)
+        try:
+            import hermes_cli.models as hm
+            monkeypatch.setattr(hm, "provider_model_ids", lambda p: [])
+        except Exception:
+            pass
+
+        r._handle_live_models(handler, parsed)
+
+        assert responses, "handler must produce a response"
+        resp = responses[-1]
+        assert "models" in resp
+        model_ids = [m["id"] for m in resp.get("models", [])]
+        assert "my-byok-model" in model_ids, (
+            f"custom_providers model 'my-byok-model' must appear in live response; "
+            f"got {model_ids}"
+        )
+
+
+# ── Regression: known-good providers still work ───────────────────────────────
+
+class TestKnownProvidersUnaffected:
+    """Normalization must not break providers whose names are already canonical."""
+
+    def test_openrouter_unaffected(self):
+        src = read("api/config.py")
+        # _PROVIDER_ALIASES lookup: 'openrouter' -> 'openrouter' (no change)
+        assert "openrouter" in src, "openrouter must still exist in config"
+
+    def test_anthropic_unaffected(self):
+        src = read("api/config.py")
+        assert "anthropic" in src
+
+    def test_custom_unaffected(self):
+        """'custom' is not in _PROVIDER_ALIASES so normalization is a no-op."""
+        try:
+            from hermes_cli.models import _PROVIDER_ALIASES
+            assert "custom" not in _PROVIDER_ALIASES, (
+                "'custom' must not be aliased to anything — it's a special sentinel"
+            )
+        except ImportError:
+            pass  # hermes-agent not available in this env — skip
+
+
+# ── Source-level: active_provider returned to browser is canonical ─────────────
+
+class TestProviderIdInGroupResponse:
+    """get_available_models() must include provider_id on every group so the JS
+    _fetchLiveModels can match optgroups exactly rather than by substring."""
+
+    def test_groups_include_provider_id(self, tmp_path, monkeypatch):
+        import api.config as c
+
+        cfgfile = tmp_path / "config.yaml"
+        cfgfile.write_text(
+            "model:\n  provider: zai\n  default: glm-5\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(c, "_get_config_path", lambda: cfgfile)
+        c.reload_config()
+        try:
+            import hermes_cli.models as hm
+            monkeypatch.setattr(hm, "list_available_providers", lambda: [
+                {"id": "zai", "authenticated": True}
+            ])
+            import hermes_cli.auth as ha
+            monkeypatch.setattr(ha, "get_auth_status", lambda p: {"key_source": "env"})
+        except Exception:
+            pass
+        result = c.get_available_models()
+        c.reload_config()
+        for g in result.get("groups", []):
+            assert "provider_id" in g, (
+                f"group {g.get('provider')!r} missing provider_id — "
+                "JS _fetchLiveModels needs it to match optgroups exactly"
+            )
+
+    def test_provider_id_in_static_ui_js_optgroup(self):
+        src = read("static/ui.js")
+        assert "og.dataset.provider" in src, (
+            "populateModelDropdown must set og.dataset.provider from g.provider_id "
+            "so _fetchLiveModels can match by exact provider_id"
+        )
+
+    def test_fetch_live_models_prefers_data_provider_match(self):
+        src = read("static/ui.js")
+        # Live model optgroup matching was extracted to _addLiveModelsToSelect (#872)
+        m = re.search(r'function _addLiveModelsToSelect\b.*?\n\}', src, re.DOTALL)
+        if not m:
+            m = re.search(r'function _fetchLiveModels\b.*?\n\}', src, re.DOTALL)
+        assert m, "_addLiveModelsToSelect or _fetchLiveModels not found"
+        fn = m.group(0)
+        assert 'og.dataset.provider' in fn, (
+            "_addLiveModelsToSelect must check og.dataset.provider===provider before "
+            "falling back to label substring match"
+        )
+        # The data-provider check must come before the label.includes check
+        dp_pos = fn.index('og.dataset.provider')
+        label_pos = fn.index('og.label')
+        assert dp_pos < label_pos, (
+            "data-provider exact match must be attempted before label substring match"
+        )
+
+
+# ── Opus-identified edge case: 'ollama' normalizes to 'custom' ────────────────
+
+class TestOllamaAliasEdgeCase:
+    """Opus review found: 'ollama' -> 'custom' via _PROVIDER_ALIASES.
+    This is better behaviour (custom_providers fallback catches it) but worth
+    documenting and not regressing."""
+
+    def test_ollama_not_in_provider_aliases_as_ollama(self):
+        """'ollama' maps to 'custom' in _PROVIDER_ALIASES — verify this is the
+        intended behavior post-normalization (not a silent breakage)."""
+        try:
+            from hermes_cli.models import _PROVIDER_ALIASES
+            # 'ollama' -> 'custom' means ollama users hit the custom_providers path
+            # This is fine — ollama models appear via base_url auto-detection (step 3)
+            # in get_available_models, not via _PROVIDER_MODELS lookup.
+            ollama_target = _PROVIDER_ALIASES.get("ollama", "ollama")
+            # Acceptable outcomes: either unchanged (not in aliases) or 'custom'/'ollama-cloud'
+            assert ollama_target in ("ollama", "custom", "ollama-cloud"), (
+                f"Unexpected ollama alias: {ollama_target}"
+            )
+        except ImportError:
+            pass  # hermes-agent not available
+
+
+class TestGetAvailableModelsReturnsCanonicalProvider:
+    """get_available_models() must return normalized active_provider in its response
+    so the browser sends the right value to /api/models/live."""
+
+    def test_active_provider_in_response_is_normalized(self, tmp_path, monkeypatch):
+        import api.config as c
+
+        cfgfile = tmp_path / "config.yaml"
+        cfgfile.write_text(
+            "model:\n  provider: z.ai\n  default: glm-5\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(c, "_get_config_path", lambda: cfgfile)
+        c.reload_config()
+        try:
+            import hermes_cli.models as hm
+            monkeypatch.setattr(hm, "list_available_providers", lambda: [])
+        except Exception:
+            pass
+        result = c.get_available_models()
+        c.reload_config()
+        ap = result.get("active_provider", "")
+        # The browser will pass this value to /api/models/live?provider=<ap>
+        # It must be 'zai' so optgroup matching works in _fetchLiveModels
+        assert ap != "z.ai", (
+            "active_provider 'z.ai' must be normalized to 'zai' before being "
+            "returned to the browser (browser passes it back to /api/models/live)"
+        )

tests/test_cancel_interrupt.py

@@ -0,0 +1,122 @@
+"""
+Unit tests for cancel/interrupt functionality.
+Tests the integration between cancel_stream() and agent.interrupt().
+"""
+import pytest
+import queue
+import threading
+from unittest.mock import Mock
+
+from api.streaming import cancel_stream
+from api.config import AGENT_INSTANCES, STREAMS, CANCEL_FLAGS
+
+
+class TestCancelInterrupt:
+    """Test suite for cancel/interrupt functionality"""
+
+    def setup_method(self):
+        """Clean up before each test"""
+        AGENT_INSTANCES.clear()
+        STREAMS.clear()
+        CANCEL_FLAGS.clear()
+
+    def teardown_method(self):
+        """Clean up after each test"""
+        AGENT_INSTANCES.clear()
+        STREAMS.clear()
+        CANCEL_FLAGS.clear()
+
+    def test_cancel_calls_agent_interrupt(self):
+        """Verify that cancel_stream() calls agent.interrupt() when agent exists"""
+        # Setup
+        stream_id = "test_stream_123"
+        mock_agent = Mock()
+        mock_agent.interrupt = Mock()
+
+        STREAMS[stream_id] = queue.Queue()
+        CANCEL_FLAGS[stream_id] = threading.Event()
+        AGENT_INSTANCES[stream_id] = mock_agent
+
+        # Execute
+        result = cancel_stream(stream_id)
+
+        # Assert
+        assert result is True
+        mock_agent.interrupt.assert_called_once_with("Cancelled by user")
+        # CANCEL_FLAGS is eagerly popped after cancel (#776 fix) so the flag
+        # is no longer in the dict — verify the pop happened instead
+        assert stream_id not in CANCEL_FLAGS, \
+            "cancel_stream() should eagerly pop CANCEL_FLAGS after signalling"
+
+    def test_cancel_handles_interrupt_exception(self):
+        """Verify that cancel_stream() handles interrupt() exceptions gracefully"""
+        stream_id = "test_stream_456"
+        mock_agent = Mock()
+        mock_agent.interrupt = Mock(side_effect=RuntimeError("Agent error"))
+
+        STREAMS[stream_id] = queue.Queue()
+        CANCEL_FLAGS[stream_id] = threading.Event()
+        AGENT_INSTANCES[stream_id] = mock_agent
+
+        # Should not raise exception
+        result = cancel_stream(stream_id)
+
+        # Assert
+        assert result is True
+        mock_agent.interrupt.assert_called_once()
+        assert stream_id not in CANCEL_FLAGS, \
+            "cancel_stream() should eagerly pop CANCEL_FLAGS even on interrupt exception"
+
+    def test_cancel_before_agent_ready(self):
+        """Test cancel when agent not yet stored in AGENT_INSTANCES (race condition)"""
+        stream_id = "test_stream_789"
+
+        STREAMS[stream_id] = queue.Queue()
+        CANCEL_FLAGS[stream_id] = threading.Event()
+        # Note: AGENT_INSTANCES[stream_id] not set (simulating race condition)
+
+        # Should succeed even without agent
+        result = cancel_stream(stream_id)
+
+        # Assert
+        assert result is True
+        # CANCEL_FLAGS is eagerly popped; the agent thread checks the event
+        # object it already has a reference to — pop doesn't clear the event
+        assert stream_id not in CANCEL_FLAGS, \
+            "cancel_stream() should eagerly pop CANCEL_FLAGS even without an agent"
+        # Agent will check this flag (it holds a reference to the event object)
+
+    def test_cancel_nonexistent_stream(self):
+        """Test cancel for a stream that doesn't exist"""
+        result = cancel_stream("nonexistent_stream")
+        assert result is False
+
+    def test_cancel_sets_cancel_event(self):
+        """Verify that cancel_stream() sets the cancel_event flag"""
+        stream_id = "test_stream_event"
+
+        STREAMS[stream_id] = queue.Queue()
+        cancel_event = threading.Event()
+        CANCEL_FLAGS[stream_id] = cancel_event
+
+        result = cancel_stream(stream_id)
+
+        assert result is True
+        assert cancel_event.is_set()
+
+    def test_cancel_puts_sentinel_in_queue(self):
+        """Verify that cancel_stream() puts cancel sentinel in queue"""
+        stream_id = "test_stream_queue"
+        q = queue.Queue()
+
+        STREAMS[stream_id] = q
+        CANCEL_FLAGS[stream_id] = threading.Event()
+
+        result = cancel_stream(stream_id)
+
+        assert result is True
+        # Check that cancel message was queued
+        assert not q.empty()
+        event_type, data = q.get_nowait()
+        assert event_type == 'cancel'
+        assert data['message'] == 'Cancelled by user'

tests/test_chinese_locale.py

@@ -0,0 +1,111 @@
+from collections import Counter
+from pathlib import Path
+import re
+
+
+REPO = Path(__file__).resolve().parent.parent
+
+
+def read(path: Path) -> str:
+    return path.read_text(encoding="utf-8")
+
+
+def extract_locale_block(src: str, locale_key: str) -> str:
+    start_match = re.search(rf"\b{re.escape(locale_key)}\s*:\s*\{{", src)
+    assert start_match, f"{locale_key} locale block not found"
+
+    start = start_match.end() - 1  # "{"
+    depth = 0
+    in_single = False
+    in_double = False
+    in_backtick = False
+    escape = False
+
+    for i in range(start, len(src)):
+        ch = src[i]
+
+        if escape:
+            escape = False
+            continue
+
+        if in_single:
+            if ch == "\\":
+                escape = True
+            elif ch == "'":
+                in_single = False
+            continue
+
+        if in_double:
+            if ch == "\\":
+                escape = True
+            elif ch == '"':
+                in_double = False
+            continue
+
+        if in_backtick:
+            if ch == "\\":
+                escape = True
+            elif ch == "`":
+                in_backtick = False
+            continue
+
+        if ch == "'":
+            in_single = True
+            continue
+        if ch == '"':
+            in_double = True
+            continue
+        if ch == "`":
+            in_backtick = True
+            continue
+
+        if ch == "{":
+            depth += 1
+            continue
+        if ch == "}":
+            depth -= 1
+            if depth == 0:
+                return src[start + 1 : i]
+
+    raise AssertionError(f"{locale_key} locale block braces are not balanced")
+
+
+def test_chinese_locale_block_exists():
+    src = read(REPO / "static" / "i18n.js")
+    assert "\n  zh: {" in src
+    assert "_lang: 'zh'" in src
+    assert "_speech: 'zh-CN'" in src
+
+
+def test_chinese_locale_includes_representative_translations():
+    src = read(REPO / "static" / "i18n.js")
+    expected = [
+        "settings_title: '\\u8bbe\\u7f6e'",
+        "login_title: '\\u767b\\u5f55'",
+        "approval_heading: '需要审批'",
+        "tab_tasks: '任务'",
+        "tab_profiles: '配置'",
+        "session_time_just_now: '刚刚'",
+        "onboarding_title: '欢迎使用 Hermes Web UI'",
+        "onboarding_complete: '引导完成'",
+    ]
+    for entry in expected:
+        assert entry in src
+
+
+def test_chinese_locale_covers_english_keys():
+    src = read(REPO / "static" / "i18n.js")
+    key_pattern = re.compile(r"^\s{4}([a-zA-Z0-9_]+):", re.MULTILINE)
+    en_keys = set(key_pattern.findall(extract_locale_block(src, "en")))
+    zh_keys = set(key_pattern.findall(extract_locale_block(src, "zh")))
+
+    missing = sorted(en_keys - zh_keys)
+    assert not missing, f"Chinese locale missing keys: {missing}"
+
+
+def test_chinese_locale_has_no_duplicate_keys():
+    src = read(REPO / "static" / "i18n.js")
+    key_pattern = re.compile(r"^\s{4}([a-zA-Z0-9_]+):", re.MULTILINE)
+    keys = key_pattern.findall(extract_locale_block(src, "zh"))
+    duplicates = sorted(k for k, count in Counter(keys).items() if count > 1)
+    assert not duplicates, f"Chinese locale has duplicate keys: {duplicates}"

tests/test_clarify_unblock.py

@@ -0,0 +1,165 @@
+"""Tests for clarify prompt unblocking and HTTP endpoints."""
+
+import json
+import threading
+import uuid
+import urllib.request
+import urllib.error
+import urllib.parse
+
+import pytest
+
+try:
+    from api.clarify import (
+        register_gateway_notify,
+        unregister_gateway_notify,
+        resolve_clarify,
+        clear_pending,
+        _gateway_queues,
+        _gateway_notify_cbs,
+        _lock,
+        _ClarifyEntry,
+        submit_pending,
+    )
+    CLARIFY_AVAILABLE = True
+except ImportError:
+    CLARIFY_AVAILABLE = False
+
+pytestmark = pytest.mark.skipif(
+    not CLARIFY_AVAILABLE,
+    reason="api.clarify not available in this environment",
+)
+
+from tests._pytest_port import BASE
+
+
+def get(path):
+    url = BASE + path
+    with urllib.request.urlopen(url, timeout=10) as r:
+        return json.loads(r.read())
+
+
+def post(path, body=None):
+    url = BASE + path
+    data = json.dumps(body or {}).encode()
+    req = urllib.request.Request(url, data=data, headers={"Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=10) as r:
+            return json.loads(r.read()), r.status
+    except urllib.error.HTTPError as e:
+        return json.loads(e.read()), e.code
+
+
+class TestClarifyUnblocking:
+    """Unit tests for clarify queue resolution."""
+
+    def test_resolve_clarify_sets_event(self):
+        sid = f"unit-clarify-{uuid.uuid4().hex[:8]}"
+        entry = _ClarifyEntry({"question": "Pick one", "choices_offered": ["a", "b"]})
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+
+        resolved = resolve_clarify(sid, "a", resolve_all=False)
+        assert resolved == 1
+        assert entry.event.is_set()
+        assert entry.result == "a"
+
+    def test_register_and_fire_notify_cb(self):
+        sid = f"unit-notify-{uuid.uuid4().hex[:8]}"
+        fired = []
+        register_gateway_notify(sid, lambda d: fired.append(d))
+
+        with _lock:
+            cb = _gateway_notify_cbs.get(sid)
+        assert cb is not None
+
+        data = {"question": "What now?", "choices_offered": ["x", "y"]}
+        cb(data)
+        assert fired == [data]
+
+        unregister_gateway_notify(sid)
+
+    def test_clear_pending_unblocks_waiters(self):
+        sid = f"unit-clear-{uuid.uuid4().hex[:8]}"
+        entry = _ClarifyEntry({"question": "Wait", "choices_offered": []})
+        with _lock:
+            _gateway_queues.setdefault(sid, []).append(entry)
+
+        cleared = clear_pending(sid)
+        assert cleared == 1
+        assert entry.event.is_set()
+        with _lock:
+            assert sid not in _gateway_queues
+
+    def test_submit_pending_registers_entry(self):
+        sid = f"unit-submit-{uuid.uuid4().hex[:8]}"
+        data = {"question": "Pick", "choices_offered": ["one", "two"], "session_id": sid}
+        entry = submit_pending(sid, data)
+        assert entry.data == data
+        with _lock:
+            assert sid in _gateway_queues
+
+        clear_pending(sid)
+
+
+class TestClarifyModuleExports:
+    def test_register_gateway_notify_exported(self):
+        import api.clarify as ap
+        assert hasattr(ap, "register_gateway_notify")
+
+    def test_unregister_gateway_notify_exported(self):
+        import api.clarify as ap
+        assert hasattr(ap, "unregister_gateway_notify")
+
+    def test_resolve_clarify_exported(self):
+        import api.clarify as ap
+        assert hasattr(ap, "resolve_clarify")
+
+    def test_clarify_entry_exported(self):
+        import api.clarify as ap
+        assert hasattr(ap, "_ClarifyEntry")
+
+
+class TestClarifyHTTPEndpoints:
+    """Regression tests for /api/clarify/respond against the live test server."""
+
+    def test_respond_returns_ok_no_pending(self):
+        sid = f"http-no-pending-{uuid.uuid4().hex[:8]}"
+        result, status = post("/api/clarify/respond", {
+            "session_id": sid,
+            "response": "Use option A",
+        })
+        assert status == 200
+        assert result["ok"] is True
+
+    def test_respond_requires_session_id(self):
+        result, status = post("/api/clarify/respond", {"response": "Hello"})
+        assert status == 400
+
+    def test_respond_requires_response(self):
+        sid = f"http-no-response-{uuid.uuid4().hex[:8]}"
+        result, status = post("/api/clarify/respond", {"session_id": sid})
+        assert status == 400
+
+    def test_respond_clears_injected_pending(self):
+        sid = f"http-clear-{uuid.uuid4().hex[:8]}"
+        question = urllib.parse.quote("Pick the better option")
+        choices = urllib.parse.quote("A")
+        inject = get(
+            f"/api/clarify/inject_test?session_id={urllib.parse.quote(sid)}"
+            f"&question={question}&choices={choices}"
+        )
+        assert inject["ok"] is True
+
+        data = get(f"/api/clarify/pending?session_id={urllib.parse.quote(sid)}")
+        assert data["pending"] is not None
+
+        result, status = post("/api/clarify/respond", {
+            "session_id": sid,
+            "response": "B",
+        })
+        assert status == 200
+        assert result["ok"] is True
+
+        data2 = get(f"/api/clarify/pending?session_id={urllib.parse.quote(sid)}")
+        assert data2["pending"] is None

tests/test_cmd_dropdown_scroll_838.py

@@ -0,0 +1,69 @@
+"""Tests for #838 — slash command dropdown keyboard navigation keeps the
+selected item in view."""
+import os
+import re
+
+
+_SRC = os.path.join(os.path.dirname(__file__), "..")
+
+
+def _read(name):
+    return open(os.path.join(_SRC, name), encoding="utf-8").read()
+
+
+class TestNavigateCmdDropdownScroll:
+    """navigateCmdDropdown must scroll the newly selected item into view so
+    keyboard navigation on a long list doesn't leave the highlight below the
+    visible area of the dropdown."""
+
+    def test_navigate_calls_scroll_into_view(self):
+        js = _read("static/commands.js")
+        m = re.search(r'function navigateCmdDropdown\(.*?\n\}', js, re.DOTALL)
+        assert m, "navigateCmdDropdown not found"
+        fn = m.group(0)
+        assert 'scrollIntoView' in fn, (
+            "navigateCmdDropdown must call scrollIntoView on the newly "
+            "selected item so ↓/↑ keeps the highlight visible (#838)"
+        )
+
+    def test_scroll_uses_nearest_block_alignment(self):
+        """`{block:'nearest'}` is the correct option: scrolls only when
+        needed, minimum distance — won't jump the list around on every
+        arrow-key press when the item is already in view."""
+        js = _read("static/commands.js")
+        m = re.search(r'function navigateCmdDropdown\(.*?\n\}', js, re.DOTALL)
+        assert m
+        fn = m.group(0)
+        assert "block:'nearest'" in fn or 'block: "nearest"' in fn, (
+            "scrollIntoView should use {block:'nearest'} to scroll the "
+            "minimum amount needed"
+        )
+
+    def test_scroll_after_selected_class_update(self):
+        """The scroll call must come AFTER adding the .selected class so
+        the correct item is targeted."""
+        js = _read("static/commands.js")
+        m = re.search(r'function navigateCmdDropdown\(.*?\n\}', js, re.DOTALL)
+        assert m
+        fn = m.group(0)
+        selected_pos = fn.find("classList.add('selected')")
+        scroll_pos = fn.find("scrollIntoView")
+        assert selected_pos != -1 and scroll_pos != -1
+        assert selected_pos < scroll_pos, (
+            "scrollIntoView must run after classList.add('selected') so it "
+            "scrolls the newly-highlighted item into view"
+        )
+
+    def test_cmd_dropdown_is_scroll_container(self):
+        """Regression guard: the .cmd-dropdown must have overflow-y:auto
+        (or similar) so scrollIntoView finds it as the scroll ancestor
+        rather than bubbling up to the viewport."""
+        css = _read("static/style.css")
+        m = re.search(r'\.cmd-dropdown\s*\{[^}]+\}', css)
+        assert m, ".cmd-dropdown rule not found"
+        block = m.group(0)
+        assert 'overflow-y:auto' in block or 'overflow-y: auto' in block or \
+               'overflow:auto' in block or 'overflow: auto' in block, (
+            ".cmd-dropdown must have overflow-y:auto so scrollIntoView "
+            "scrolls within the dropdown, not the whole page"
+        )

tests/test_commands_endpoint.py

@@ -0,0 +1,84 @@
+"""Tests for GET /api/commands -- exposes hermes-agent COMMAND_REGISTRY."""
+import json
+import urllib.request
+
+import pytest
+
+from tests.conftest import TEST_BASE, requires_agent_modules
+
+
+def _get(path):
+    """GET helper -- returns parsed JSON or raises HTTPError."""
+    with urllib.request.urlopen(TEST_BASE + path, timeout=10) as r:
+        return json.loads(r.read())
+
+
+@requires_agent_modules
+def test_commands_endpoint_returns_list():
+    """GET /api/commands returns a JSON object with a 'commands' list."""
+    body = _get('/api/commands')
+    assert 'commands' in body
+    assert isinstance(body['commands'], list)
+    assert len(body['commands']) > 0
+
+
+@requires_agent_modules
+def test_commands_endpoint_includes_help():
+    """The 'help' command must always be present (it's not cli_only)."""
+    body = _get('/api/commands')
+    names = {c['name'] for c in body['commands']}
+    assert 'help' in names
+
+
+@requires_agent_modules
+def test_commands_endpoint_command_shape():
+    """Each command entry has the required fields."""
+    body = _get('/api/commands')
+    cmd = next(c for c in body['commands'] if c['name'] == 'help')
+    required = {
+        'name', 'description', 'category', 'aliases',
+        'args_hint', 'subcommands', 'cli_only', 'gateway_only',
+    }
+    assert set(cmd.keys()) >= required
+    assert isinstance(cmd['aliases'], list)
+    assert isinstance(cmd['subcommands'], list)
+    assert isinstance(cmd['cli_only'], bool)
+    assert isinstance(cmd['gateway_only'], bool)
+
+
+@requires_agent_modules
+def test_commands_endpoint_excludes_gateway_only_and_never_expose():
+    """gateway_only commands and the _NEVER_EXPOSE set are filtered out."""
+    body = _get('/api/commands')
+    names = {c['name'] for c in body['commands']}
+    # /sethome, /restart, /update are gateway_only; /commands is in _NEVER_EXPOSE
+    for name in ('sethome', 'restart', 'update', 'commands'):
+        assert name not in names, f"{name} must be excluded from /api/commands"
+
+
+@requires_agent_modules
+def test_commands_endpoint_keeps_new_with_reset_alias():
+    """The 'new' command stays exposed and carries its 'reset' alias."""
+    body = _get('/api/commands')
+    new_cmd = next(c for c in body['commands'] if c['name'] == 'new')
+    assert 'reset' in new_cmd['aliases']
+
+
+def test_list_commands_returns_empty_for_empty_registry():
+    """list_commands(_registry=[]) returns [] -- the same path as when
+    hermes_cli is missing (the empty-or-missing case)."""
+    from api.commands import list_commands
+    assert list_commands(_registry=[]) == []
+
+
+def test_list_commands_degrades_when_agent_missing(monkeypatch):
+    """If hermes_cli.commands is not importable, list_commands() returns []
+    via the ImportError path. Verified by stubbing sys.modules; test cleanup
+    is handled by monkeypatch + the fact that we don't reload api.commands."""
+    import sys
+    monkeypatch.setitem(sys.modules, 'hermes_cli.commands', None)
+    # NOTE: we do NOT reload api.commands. The lazy import inside
+    # list_commands() will re-attempt the import on each call and hit
+    # the stubbed-None module, raising ImportError, taking the fallback path.
+    from api.commands import list_commands
+    assert list_commands() == []

tests/test_credential_pool_providers.py

@@ -0,0 +1,595 @@
+"""Regression tests for credential_pool provider detection in /api/models."""
+
+import json
+import sys
+import types
+
+import api.config as config
+import api.profiles as profiles
+
+_AMBIENT_SOURCES = {"gh_cli", "gh auth token"}
+
+
+def _install_fake_hermes_cli(monkeypatch, *, with_load_pool: bool = False, pool_data: dict | None = None):
+    """Stub hermes_cli modules so tests are deterministic and offline.
+
+    When *with_load_pool* is True, also stubs hermes_cli.credential_pool with a
+    suppression-aware load_pool() implementation that mirrors upstream behaviour:
+    entries whose source/label/key_source signals ambient gh-cli auth are filtered out.
+    """
+    fake_pkg = types.ModuleType("hermes_cli")
+    fake_pkg.__path__ = []
+
+    fake_models = types.ModuleType("hermes_cli.models")
+    fake_models.list_available_providers = lambda: []
+    fake_models.provider_model_ids = lambda pid: (
+        ["gpt-oss:20b", "qwen3:30b-a3b"] if pid == "ollama-cloud" else []
+    )
+
+    fake_auth = types.ModuleType("hermes_cli.auth")
+    fake_auth.get_auth_status = lambda _pid: {}
+
+    monkeypatch.setitem(sys.modules, "hermes_cli", fake_pkg)
+    monkeypatch.setitem(sys.modules, "hermes_cli.models", fake_models)
+    monkeypatch.setitem(sys.modules, "hermes_cli.auth", fake_auth)
+
+    # Always remove the real agent.credential_pool so get_available_models() takes
+    # the ImportError fallback path and reads from the monkeypatched auth store,
+    # not the live ~/.hermes/auth.json via the real venv module.
+    monkeypatch.delitem(sys.modules, "agent.credential_pool", raising=False)
+    monkeypatch.delitem(sys.modules, "agent", raising=False)
+
+    if with_load_pool:
+        _pool_data = pool_data or {}
+
+        class _FakeEntry:
+            """Minimal PooledCredential stand-in with attribute access (matching the real class)."""
+            def __init__(self, d):
+                self.source = d.get("source", "manual")
+                self.label = d.get("label", "")
+                self.key_source = d.get("key_source", "")
+                self.id = d.get("id", "")
+
+        class _FakePool:
+            def __init__(self, entries_list):
+                self._entries = entries_list
+
+            def entries(self):
+                return self._entries
+
+        def _fake_load_pool(pid):
+            # Return ALL entries without filtering — mirrors the real load_pool()
+            # which does NOT suppress ambient gh-cli tokens on its own.
+            # Ambient-source filtering is the webui's responsibility.
+            raw = _pool_data.get(pid, [])
+            return _FakePool([_FakeEntry(e) for e in raw])
+
+        fake_cp = types.ModuleType("agent.credential_pool")
+        fake_cp.load_pool = _fake_load_pool
+        monkeypatch.setitem(sys.modules, "agent.credential_pool", fake_cp)
+
+
+def _call_get_available_models(monkeypatch, tmp_path, auth_payload, *, with_load_pool: bool = False):
+    """Call get_available_models() with auth.json pinned to a temp Hermes home."""
+    _install_fake_hermes_cli(
+        monkeypatch,
+        with_load_pool=with_load_pool,
+        pool_data=auth_payload.get("credential_pool", {}),
+    )
+
+    (tmp_path / "auth.json").write_text(json.dumps(auth_payload), encoding="utf-8")
+    monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: tmp_path)
+
+    old_cfg = dict(config.cfg)
+    old_mtime = config._cfg_mtime
+    config.cfg.clear()
+    config.cfg["model"] = {}
+    try:
+        # Pin mtime to avoid reload_config() clobbering our in-memory cfg patch.
+        config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+    except Exception:
+        config._cfg_mtime = 0.0
+
+    config.invalidate_models_cache()
+    try:
+        return config.get_available_models()
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+        config._cfg_mtime = old_mtime
+        config.invalidate_models_cache()
+
+
+def _group_by_provider(result):
+    return {g["provider"]: g["models"] for g in result.get("groups", [])}
+
+
+def test_ollama_cloud_manual_credential_shows_group(monkeypatch, tmp_path):
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "ollama-cloud": [
+                {
+                    "id": "abc123",
+                    "label": "ollama-manual",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                    "base_url": "https://ollama.com/v1",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "Ollama Cloud" in groups, f"Expected Ollama Cloud in {list(groups)}"
+    model_ids = [m["id"] for m in groups["Ollama Cloud"]]
+    assert model_ids == ["@ollama-cloud:gpt-oss:20b", "@ollama-cloud:qwen3:30b-a3b"], model_ids
+
+
+def test_copilot_gh_cli_only_credential_hidden(monkeypatch, tmp_path):
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "def456",
+                    "label": "gh auth token",
+                    "source": "gh_cli",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" not in groups, (
+        "GitHub Copilot should be hidden when only ambient gh auth token is present; "
+        f"got {list(groups)}"
+    )
+
+
+def test_copilot_mixed_credential_pool_remains_visible(monkeypatch, tmp_path):
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "def456",
+                    "label": "gh auth token",
+                    "source": "gh_cli",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                },
+                {
+                    "id": "ghi789",
+                    "label": "explicit-copilot",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                },
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" in groups, f"Expected GitHub Copilot in {list(groups)}"
+    model_ids = [m["id"] for m in groups["GitHub Copilot"]]
+    assert "@copilot:gpt-5.4" in model_ids, model_ids
+    assert "@copilot:claude-opus-4.6" in model_ids, model_ids
+
+
+def test_copilot_empty_field_entries_are_treated_as_explicit(monkeypatch, tmp_path):
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "jkl012",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" in groups, f"Expected GitHub Copilot in {list(groups)}"
+
+
+def test_copilot_oauth_credential_is_visible(monkeypatch, tmp_path):
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "mno345",
+                    "label": "github-oauth",
+                    "source": "oauth",
+                    "auth_type": "oauth",
+                    "base_url": "https://api.githubcopilot.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" in groups, f"Expected GitHub Copilot in {list(groups)}"
+
+
+# --- load_pool path (suppression-aware) ---
+
+
+def test_load_pool_copilot_ambient_only_remains_hidden(monkeypatch, tmp_path):
+    """load_pool path: copilot with only ambient gh-cli entries is suppressed."""
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "lp001",
+                    "label": "gh auth token",
+                    "source": "gh_cli",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload, with_load_pool=True)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" not in groups, (
+        "GitHub Copilot must be hidden when load_pool returns no usable entries; "
+        f"got {list(groups)}"
+    )
+
+
+def test_load_pool_copilot_ambient_key_source_only_remains_hidden(monkeypatch, tmp_path):
+    """load_pool path: key_source-only ambient markers must also be suppressed."""
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "lp001b",
+                    "label": "copilot-token",
+                    "source": "manual",
+                    "key_source": "gh auth token",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload, with_load_pool=True)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" not in groups, (
+        "GitHub Copilot must stay hidden when load_pool entries only differ by key_source ambient markers; "
+        f"got {list(groups)}"
+    )
+
+
+def test_load_pool_alias_provider_key_is_resolved(monkeypatch, tmp_path):
+    """load_pool path: aliased pool keys should resolve to canonical provider ids."""
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "google": [
+                {
+                    "id": "gp001",
+                    "label": "explicit-gemini",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                    "base_url": "https://generativelanguage.googleapis.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload, with_load_pool=True)
+    groups = _group_by_provider(result)
+    assert "Gemini" in groups, f"Expected Gemini in {list(groups)}"
+    assert "Google" not in groups, f"Aliased provider key should not render under raw alias name: {list(groups)}"
+
+
+def test_load_pool_explicit_credential_shows_provider(monkeypatch, tmp_path):
+    """load_pool path: provider with at least one explicit entry is visible."""
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "lp002",
+                    "label": "gh auth token",
+                    "source": "gh_cli",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                },
+                {
+                    "id": "lp003",
+                    "label": "explicit-pat",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                },
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload, with_load_pool=True)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" in groups, (
+        f"GitHub Copilot must appear when load_pool has at least one usable entry; got {list(groups)}"
+    )
+
+
+# --- _apply_provider_prefix helper ---
+
+
+def test_apply_provider_prefix_ollama_cloud_non_active():
+    """Bare ollama-cloud model ids get @ollama-cloud: prefix when not active."""
+    from api.config import _apply_provider_prefix
+
+    raw = [{"id": "gpt-oss:20b", "label": "gpt-oss:20b"}, {"id": "qwen3:30b-a3b", "label": "qwen3:30b-a3b"}]
+    result = _apply_provider_prefix(raw, "ollama-cloud", "openai-codex")
+    ids = [m["id"] for m in result]
+    assert ids == ["@ollama-cloud:gpt-oss:20b", "@ollama-cloud:qwen3:30b-a3b"], ids
+
+
+def test_apply_provider_prefix_copilot_non_active():
+    """Bare copilot model ids get @copilot: prefix when not active."""
+    from api.config import _apply_provider_prefix
+
+    raw = [{"id": "gpt-5.4", "label": "GPT-5.4"}, {"id": "claude-opus-4.6", "label": "Claude Opus 4.6"}]
+    result = _apply_provider_prefix(raw, "copilot", "openai-codex")
+    ids = [m["id"] for m in result]
+    assert ids == ["@copilot:gpt-5.4", "@copilot:claude-opus-4.6"], ids
+
+
+def test_apply_provider_prefix_no_double_prefix():
+    """Already-prefixed or provider/model ids are not double-prefixed."""
+    from api.config import _apply_provider_prefix
+
+    raw = [
+        {"id": "@copilot:gpt-5.4", "label": "already prefixed"},
+        {"id": "openai/gpt-5.4", "label": "slash form"},
+        {"id": "bare-model", "label": "bare"},
+    ]
+    result = _apply_provider_prefix(raw, "copilot", "openai-codex")
+    ids = [m["id"] for m in result]
+    assert ids == ["@copilot:gpt-5.4", "openai/gpt-5.4", "@copilot:bare-model"], ids
+
+
+def test_apply_provider_prefix_active_provider_no_prefix():
+    """No prefix is added when the provider is already the active one."""
+    from api.config import _apply_provider_prefix
+
+    raw = [{"id": "gpt-5.4", "label": "GPT-5.4"}]
+    result = _apply_provider_prefix(raw, "openai-codex", "openai-codex")
+    ids = [m["id"] for m in result]
+    assert ids == ["gpt-5.4"], ids
+
+
+def test_copilot_mixed_pool_prefixed_models(monkeypatch, tmp_path):
+    """Copilot with mixed pool and non-active provider has @copilot: prefixed model ids."""
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "copilot": [
+                {
+                    "id": "lp010",
+                    "label": "explicit-copilot",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                    "base_url": "https://api.githubcopilot.com",
+                }
+            ]
+        },
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    assert "GitHub Copilot" in groups
+    model_ids = [m["id"] for m in groups["GitHub Copilot"]]
+    assert all(mid.startswith("@copilot:") for mid in model_ids), model_ids
+
+
+def test_auth_store_active_provider_alias_is_resolved(monkeypatch, tmp_path):
+    """active_provider read from auth.json must be alias-normalized.
+
+    Regression: previously the alias table was applied only to config.yaml's
+    active_provider, so an aliased name in auth.json (e.g. 'google') would
+    not match the canonical pid ('gemini') and the prefixing logic would
+    add an unwanted '@gemini:' prefix to the active provider's models.
+    """
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        # Aliased name: 'google' → 'gemini' per _PROVIDER_ALIASES.
+        "active_provider": "google",
+        "credential_pool": {},
+    }
+
+    result = _call_get_available_models(monkeypatch, tmp_path, auth_payload)
+    groups = _group_by_provider(result)
+    # Gemini should appear under its canonical display name and its model
+    # ids should NOT be prefixed (it's the active provider).
+    assert "Gemini" in groups, f"Expected Gemini in {list(groups)}"
+    model_ids = [m["id"] for m in groups["Gemini"]]
+    assert model_ids, "Gemini group should have models"
+    assert not any(mid.startswith("@") for mid in model_ids), (
+        f"Active provider models must not be prefixed; got {model_ids}"
+    )
+
+
+def test_ollama_cloud_empty_catalog_skips_group(monkeypatch, tmp_path):
+    """When hermes_cli returns no models for ollama-cloud, the group is omitted.
+
+    Matches the named-custom and unknown-provider branches: we don't invent a
+    catalog we can't enumerate. The logger.warning in the except branch keeps
+    diagnostics available for operators.
+    """
+    _install_fake_hermes_cli(monkeypatch)
+
+    # Override the stub to return empty for ollama-cloud.
+    import sys as _sys
+    _sys.modules["hermes_cli.models"].provider_model_ids = lambda pid: []
+
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "ollama-cloud": [
+                {
+                    "id": "oc-empty",
+                    "label": "ollama-manual",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                }
+            ]
+        },
+    }
+
+    (tmp_path / "auth.json").write_text(json.dumps(auth_payload), encoding="utf-8")
+    monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: tmp_path)
+
+    old_cfg = dict(config.cfg)
+    old_mtime = config._cfg_mtime
+    config.cfg.clear()
+    config.cfg["model"] = {}
+    try:
+        config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+    except Exception:
+        config._cfg_mtime = 0.0
+
+    try:
+        result = config.get_available_models()
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+        config._cfg_mtime = old_mtime
+
+    groups = _group_by_provider(result)
+    assert "Ollama Cloud" not in groups, (
+        f"Ollama Cloud group should be skipped when catalog is empty; got {list(groups)}"
+    )
+
+
+# --- _format_ollama_label helper ---
+
+
+def test_format_ollama_label_simple():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("kimi-k2.5") == "Kimi K2.5"
+
+
+def test_format_ollama_label_with_variant():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("qwen3-vl:235b-instruct") == "Qwen3 VL (235B Instruct)"
+
+
+def test_format_ollama_label_short_acronym():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("glm-5.1") == "GLM 5.1"
+
+
+def test_format_ollama_label_gpt_oss_with_size():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("gpt-oss:20b") == "GPT OSS (20B)"
+
+
+def test_format_ollama_label_empty_string():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("") == ""
+
+
+def test_format_ollama_label_no_variant():
+    from api.config import _format_ollama_label
+
+    assert _format_ollama_label("nemotron-3-super") == "Nemotron 3 Super"
+
+
+# --- Fallback-path (ImportError branch) alias resolution ---
+
+
+def test_fallback_path_resolves_alias_when_load_pool_unavailable(monkeypatch, tmp_path):
+    """When agent.credential_pool can't be imported, the manual-inspection
+    branch must still canonicalize pool keys so aliased names (e.g. 'google')
+    end up under their canonical provider id ('gemini')."""
+    _install_fake_hermes_cli(monkeypatch)
+    # Ensure agent.credential_pool is not importable so the fallback branch runs.
+    monkeypatch.setitem(sys.modules, "agent.credential_pool", None)
+
+    auth_payload = {
+        "version": 1,
+        "providers": {},
+        "active_provider": "openai-codex",
+        "credential_pool": {
+            "google": [
+                {
+                    "id": "gp-fallback",
+                    "label": "explicit-gemini",
+                    "source": "manual",
+                    "auth_type": "api_key",
+                }
+            ]
+        },
+    }
+
+    (tmp_path / "auth.json").write_text(json.dumps(auth_payload), encoding="utf-8")
+    monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: tmp_path)
+
+    old_cfg = dict(config.cfg)
+    old_mtime = config._cfg_mtime
+    config.cfg.clear()
+    config.cfg["model"] = {}
+    try:
+        config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+    except Exception:
+        config._cfg_mtime = 0.0
+
+    try:
+        result = config.get_available_models()
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+        config._cfg_mtime = old_mtime
+
+    groups = _group_by_provider(result)
+    assert "Gemini" in groups, (
+        f"Fallback path must resolve 'google' -> 'gemini'; got {list(groups)}"
+    )
+    assert "Google" not in groups, (
+        f"Raw alias name must not leak when fallback path runs; got {list(groups)}"
+    )

tests/test_cron_refresh_button_835.py

@@ -0,0 +1,110 @@
+"""Tests for #835 — refresh button in Tasks / Scheduled Jobs panel."""
+import os
+import re
+
+
+_SRC = os.path.join(os.path.dirname(__file__), "..")
+
+
+def _read(name):
+    return open(os.path.join(_SRC, name), encoding="utf-8").read()
+
+
+class TestCronRefreshButtonHtml:
+    """index.html must expose a refresh button in the Tasks panel header."""
+
+    def test_refresh_button_present(self):
+        html = _read("static/index.html")
+        assert 'id="cronRefreshBtn"' in html, (
+            "Tasks panel must have a #cronRefreshBtn element"
+        )
+
+    def test_refresh_button_has_accessibility_labels(self):
+        """Icon-only buttons need aria-label + title so screen readers and
+        hover tooltips work."""
+        html = _read("static/index.html")
+        m = re.search(r'<button[^>]*id="cronRefreshBtn"[^>]*>', html)
+        assert m, "cronRefreshBtn tag not found"
+        tag = m.group(0)
+        assert 'aria-label=' in tag, (
+            "#cronRefreshBtn is icon-only and must have aria-label"
+        )
+        assert 'title=' in tag, (
+            "#cronRefreshBtn should have a title tooltip"
+        )
+
+    def test_refresh_button_calls_load_crons_with_animate(self):
+        html = _read("static/index.html")
+        m = re.search(r'<button[^>]*id="cronRefreshBtn"[^>]*>', html)
+        assert m
+        tag = m.group(0)
+        assert 'loadCrons(true)' in tag, (
+            "#cronRefreshBtn must call loadCrons(true) to enable the dim-while-fetching animation"
+        )
+
+    def test_refresh_button_sits_next_to_new_job_button(self):
+        """Refresh button should appear in the same header row as the New Job
+        button so the header layout stays tight."""
+        html = _read("static/index.html")
+        ref_pos = html.find('id="cronRefreshBtn"')
+        newjob_pos = html.find('openCronCreate()')
+        assert ref_pos != -1 and newjob_pos != -1
+        # Must be close enough to be in the same header row (single SVG-inline
+        # button can be around 500 chars by itself due to inline styles/attrs).
+        assert abs(ref_pos - newjob_pos) < 1000, (
+            "Refresh button and New Job button should be in the same header row"
+        )
+
+
+class TestLoadCronsAnimateFlag:
+    """panels.js loadCrons() must accept an optional animate flag that dims
+    the refresh button while fetching."""
+
+    def test_load_crons_accepts_animate_param(self):
+        js = _read("static/panels.js")
+        assert re.search(r'async function loadCrons\s*\(\s*animate\s*\)', js), (
+            "loadCrons must accept an `animate` parameter"
+        )
+
+    def test_load_crons_restores_button_in_finally(self):
+        """The opacity/disabled restore MUST be in a finally block so a
+        throwing fetch doesn't leave the button stuck at 0.5 / disabled."""
+        js = _read("static/panels.js")
+        m = re.search(r'async function loadCrons\(.*?\n\}', js, re.DOTALL)
+        assert m, "loadCrons body not found"
+        fn = m.group(0)
+        assert 'finally' in fn, (
+            "loadCrons must restore the refresh button's opacity/disabled state "
+            "in a finally block so errors during fetch don't leave the button stuck"
+        )
+        # The restore block sets opacity='' (not '1') so CSS cascade wins
+        assert "opacity = ''" in fn or "opacity=''" in fn, (
+            "restore must use opacity='' to clear the inline override"
+        )
+
+
+class TestCronCreatedEventListener:
+    """A global `hermes:cron_created` listener must be registered so
+    future chat paths can trigger the cron list refresh."""
+
+    def test_listener_registered_at_module_scope(self):
+        js = _read("static/panels.js")
+        assert re.search(
+            r"addEventListener\(\s*['\"]hermes:cron_created['\"]",
+            js,
+        ), (
+            "panels.js must register a window-level 'hermes:cron_created' event listener"
+        )
+
+    def test_listener_triggers_load_crons(self):
+        js = _read("static/panels.js")
+        m = re.search(
+            r"addEventListener\(\s*['\"]hermes:cron_created['\"].*?\}\s*\)",
+            js,
+            re.DOTALL,
+        )
+        assert m, "hermes:cron_created listener body not found"
+        body = m.group(0)
+        assert 'loadCrons' in body, (
+            "hermes:cron_created listener must call loadCrons() to refresh the list"
+        )

tests/test_cron_session_title.py

@@ -0,0 +1,148 @@
+"""Tests for cron session title fallback in get_cli_sessions().
+
+When a CLI session originates from cron and has no title in state.db, the
+WebUI sidebar should display the human-friendly job name from cron/jobs.json
+instead of a generic "Cron Session" label.
+
+Session ID format produced by hermes-agent: cron_<job_id>_<YYYYMMDD>_<HHMMSS>
+"""
+import json
+import sqlite3
+
+import pytest
+
+import api.models as models
+
+
+def _make_state_db(path, sessions):
+    """Create a state.db with the schema get_cli_sessions() expects.
+
+    `sessions` is a list of (id, title, source) tuples.
+    """
+    conn = sqlite3.connect(str(path))
+    conn.execute("""
+        CREATE TABLE sessions (
+            id TEXT PRIMARY KEY,
+            title TEXT,
+            model TEXT,
+            message_count INTEGER,
+            started_at REAL,
+            source TEXT
+        )
+    """)
+    conn.execute("""
+        CREATE TABLE messages (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            session_id TEXT,
+            timestamp REAL
+        )
+    """)
+    for sid, title, source in sessions:
+        conn.execute(
+            "INSERT INTO sessions (id, title, model, message_count, started_at, source) "
+            "VALUES (?, ?, ?, ?, ?, ?)",
+            (sid, title, "gpt-x", 1, 1700000000.0, source),
+        )
+        conn.execute(
+            "INSERT INTO messages (session_id, timestamp) VALUES (?, ?)",
+            (sid, 1700000001.0),
+        )
+    conn.commit()
+    conn.close()
+
+
+def _write_jobs_json(hermes_home, jobs):
+    """Write cron/jobs.json with the given jobs list."""
+    cron_dir = hermes_home / "cron"
+    cron_dir.mkdir(parents=True, exist_ok=True)
+    (cron_dir / "jobs.json").write_text(
+        json.dumps({"jobs": jobs}), encoding="utf-8"
+    )
+
+
+@pytest.fixture
+def fake_hermes_home(tmp_path, monkeypatch):
+    """Point get_cli_sessions() at a temporary HERMES_HOME and disable
+    profile lookups so the test runs hermetically."""
+    home = tmp_path / "hermes"
+    home.mkdir()
+
+    # Both profile helpers are imported lazily inside get_cli_sessions(),
+    # so patching the api.profiles module reaches them.
+    import api.profiles as profiles
+    monkeypatch.setattr(profiles, "get_active_hermes_home", lambda: home)
+    monkeypatch.setattr(profiles, "get_active_profile_name", lambda: None)
+
+    return home
+
+
+def test_cron_session_uses_job_name_when_title_missing(fake_hermes_home):
+    """A cron session with no title should display the friendly job name."""
+    _write_jobs_json(fake_hermes_home, [
+        {"id": "cd65df6fc1a8", "name": "wiki-auto-ingest"},
+    ])
+    _make_state_db(fake_hermes_home / "state.db", [
+        ("cron_cd65df6fc1a8_20260417_191049", None, "cron"),
+    ])
+
+    sessions = models.get_cli_sessions()
+
+    assert len(sessions) == 1
+    assert sessions[0]["title"] == "wiki-auto-ingest"
+
+
+def test_cron_session_falls_back_when_jobs_json_missing(fake_hermes_home):
+    """No jobs.json should not crash; title falls back to 'Cron Session'."""
+    _make_state_db(fake_hermes_home / "state.db", [
+        ("cron_abc123_20260417_191049", None, "cron"),
+    ])
+
+    sessions = models.get_cli_sessions()
+
+    assert sessions[0]["title"] == "Cron Session"
+
+
+def test_cron_session_falls_back_when_job_id_not_in_jobs_json(fake_hermes_home):
+    """Stale session whose job has been deleted falls back gracefully."""
+    _write_jobs_json(fake_hermes_home, [
+        {"id": "different_job", "name": "Some Other Job"},
+    ])
+    _make_state_db(fake_hermes_home / "state.db", [
+        ("cron_orphan_20260417_191049", None, "cron"),
+    ])
+
+    sessions = models.get_cli_sessions()
+
+    assert sessions[0]["title"] == "Cron Session"
+
+
+def test_explicit_title_is_preserved(fake_hermes_home):
+    """If state.db already has a title, the cron job lookup should not
+    override it."""
+    _write_jobs_json(fake_hermes_home, [
+        {"id": "cd65df6fc1a8", "name": "wiki-auto-ingest"},
+    ])
+    _make_state_db(fake_hermes_home / "state.db", [
+        ("cron_cd65df6fc1a8_20260417_191049", "User-edited title", "cron"),
+    ])
+
+    sessions = models.get_cli_sessions()
+
+    assert sessions[0]["title"] == "User-edited title"
+
+
+def test_non_cron_sessions_unaffected(fake_hermes_home):
+    """The cron-name lookup must not run for cli-source sessions, so the
+    generic 'Cli Session' fallback still applies when title is empty."""
+    _write_jobs_json(fake_hermes_home, [
+        {"id": "cd65df6fc1a8", "name": "wiki-auto-ingest"},
+    ])
+    # A 'cli' session whose ID coincidentally starts with 'cron_' must not
+    # pick up the job name — the source check guards against this.
+    _make_state_db(fake_hermes_home / "state.db", [
+        ("cron_cd65df6fc1a8_xx", None, "cli"),
+    ])
+
+    sessions = models.get_cli_sessions()
+
+    assert sessions[0]["title"] == "Cli Session"

tests/test_custom_provider_display_name.py

@@ -0,0 +1,165 @@
+"""
+Tests for named custom provider display in the model dropdown (issue #557).
+
+When a custom_providers entry carries a `name` field (e.g. "Agent37"), the
+web UI model picker should show that name as the group header rather than the
+generic "Custom" label.
+"""
+import pytest
+import api.config as config
+
+
+@pytest.fixture(autouse=True)
+def _isolate_models_cache():
+    """Invalidate the models TTL cache before and after every test in this file."""
+    try:
+        config.invalidate_models_cache()
+    except Exception:
+        pass
+    yield
+    try:
+        config.invalidate_models_cache()
+    except Exception:
+        pass
+
+
+def _models_with_cfg(model_cfg=None, custom_providers=None, active_provider=None):
+    """Temporarily patch config.cfg, call get_available_models(), restore.
+
+    Also pins _cfg_mtime to the current config.yaml mtime before calling
+    get_available_models().  Without this, if a prior test wrote config.yaml
+    (changing its mtime), the mtime-guard inside get_available_models() fires
+    reload_config() which overwrites config.cfg with the real on-disk values,
+    silently discarding the patch and causing ordering-dependent failures.
+    This matches the pattern used in test_model_resolver.py.
+    """
+    old_cfg = dict(config.cfg)
+    old_mtime = config._cfg_mtime
+    config.cfg.clear()
+    if model_cfg:
+        config.cfg["model"] = model_cfg
+    if custom_providers is not None:
+        config.cfg["custom_providers"] = custom_providers
+    # Pin mtime so get_available_models() skips its reload_config() guard.
+    try:
+        config._cfg_mtime = config.Path(config._get_config_path()).stat().st_mtime
+    except Exception:
+        config._cfg_mtime = 0.0  # no config.yaml present; reload guard is a no-op
+    try:
+        return config.get_available_models()
+    finally:
+        config.cfg.clear()
+        config.cfg.update(old_cfg)
+        config._cfg_mtime = old_mtime
+
+
+# ── Named provider shows its name in the dropdown ─────────────────────────────
+
+class TestNamedCustomProviderGroup:
+
+    def test_named_provider_uses_name_as_group_header(self):
+        """A custom_provider entry with name='Agent37' should produce
+        a group whose 'provider' key is 'Agent37', not 'Custom'."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom", "base_url": "https://agent37.example.com/v1"},
+            custom_providers=[
+                {"name": "Agent37", "model": "default", "base_url": "https://agent37.example.com/v1"}
+            ],
+        )
+        group_names = [g["provider"] for g in result.get("groups", [])]
+        assert "Agent37" in group_names, (
+            f"Expected 'Agent37' in group names, got {group_names}"
+        )
+
+    def test_named_provider_does_not_produce_generic_custom(self):
+        """When all custom_provider entries have names, no group called 'Custom'
+        should appear alongside them."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom", "base_url": "https://agent37.example.com/v1"},
+            custom_providers=[
+                {"name": "Agent37", "model": "default", "base_url": "https://agent37.example.com/v1"}
+            ],
+        )
+        group_names = [g["provider"] for g in result.get("groups", [])]
+        assert "Custom" not in group_names, (
+            f"Expected no generic 'Custom' group when all entries are named, got {group_names}"
+        )
+
+    def test_named_provider_model_appears_in_its_group(self):
+        """The model ID from the named entry should be inside the named group."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom"},
+            custom_providers=[
+                {"name": "Agent37", "model": "my-llm", "base_url": "https://agent37.example.com/v1"}
+            ],
+        )
+        agent37_group = next(
+            (g for g in result.get("groups", []) if g["provider"] == "Agent37"), None
+        )
+        assert agent37_group is not None, "Expected an 'Agent37' group"
+        model_ids = [m["id"] for m in agent37_group.get("models", [])]
+        assert "my-llm" in model_ids, (
+            f"Expected 'my-llm' in Agent37 group models, got {model_ids}"
+        )
+
+    def test_multiple_named_providers_each_get_their_own_group(self):
+        """Two named custom providers should produce two distinct groups."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom"},
+            custom_providers=[
+                {"name": "Agent37", "model": "fast-model"},
+                {"name": "PrivateProxy", "model": "private-llm"},
+            ],
+        )
+        group_names = [g["provider"] for g in result.get("groups", [])]
+        assert "Agent37" in group_names, f"Expected 'Agent37' group, got {group_names}"
+        assert "PrivateProxy" in group_names, f"Expected 'PrivateProxy' group, got {group_names}"
+        assert "Custom" not in group_names, f"No generic 'Custom' group expected, got {group_names}"
+
+    def test_multiple_models_in_same_named_provider(self):
+        """Multiple entries with the same name should be collapsed into one group."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom"},
+            custom_providers=[
+                {"name": "Agent37", "model": "model-a"},
+                {"name": "Agent37", "model": "model-b"},
+            ],
+        )
+        agent37_groups = [g for g in result.get("groups", []) if g["provider"] == "Agent37"]
+        assert len(agent37_groups) == 1, (
+            f"Expected exactly one 'Agent37' group, got {len(agent37_groups)}"
+        )
+        model_ids = [m["id"] for m in agent37_groups[0].get("models", [])]
+        assert "model-a" in model_ids
+        assert "model-b" in model_ids
+
+
+# ── Unnamed entry still falls back to 'Custom' ─────────────────────────────────
+
+class TestUnnamedCustomProviderFallback:
+
+    def test_unnamed_entry_still_produces_custom_group(self):
+        """A custom_provider entry without a name should still show as 'Custom'."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom"},
+            custom_providers=[
+                {"model": "unnamed-model"}
+            ],
+        )
+        group_names = [g["provider"] for g in result.get("groups", [])]
+        assert "Custom" in group_names, (
+            f"Expected generic 'Custom' group for unnamed entry, got {group_names}"
+        )
+
+    def test_mixed_named_and_unnamed_entries(self):
+        """Named and unnamed entries should appear in their respective groups."""
+        result = _models_with_cfg(
+            model_cfg={"provider": "custom"},
+            custom_providers=[
+                {"name": "Agent37", "model": "named-model"},
+                {"model": "unnamed-model"},
+            ],
+        )
+        group_names = [g["provider"] for g in result.get("groups", [])]
+        assert "Agent37" in group_names, f"Expected 'Agent37' group, got {group_names}"
+        assert "Custom" in group_names, f"Expected 'Custom' group for unnamed entry, got {group_names}"

tests/test_default_workspace_fallback.py

@@ -0,0 +1,148 @@
+import json
+from pathlib import Path
+
+import api.config as config
+
+
+def test_resolve_default_workspace_falls_back_to_existing_home_work(monkeypatch, tmp_path):
+    preferred = tmp_path / "work"
+    preferred.mkdir()
+    state_dir = tmp_path / "state"
+
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+
+    resolved = config.resolve_default_workspace("/definitely/not/usable")
+
+    assert resolved == preferred.resolve()
+
+
+
+def test_save_settings_rewrites_bad_default_workspace_to_fallback(monkeypatch, tmp_path):
+    preferred = tmp_path / "work"
+    preferred.mkdir()
+    state_dir = tmp_path / "state"
+    settings_file = tmp_path / "settings.json"
+
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    monkeypatch.setattr(config, "SETTINGS_FILE", settings_file)
+    monkeypatch.setattr(config, "DEFAULT_WORKSPACE", preferred)
+
+    saved = config.save_settings({"default_workspace": "/definitely/not/usable"})
+    on_disk = json.loads(settings_file.read_text(encoding="utf-8"))
+
+    assert saved["default_workspace"] == str(preferred.resolve())
+    assert on_disk["default_workspace"] == str(preferred.resolve())
+
+
+def test_resolve_default_workspace_creates_home_workspace_when_missing(monkeypatch, tmp_path):
+    """When no preferred dir exists, resolve falls back to creating ~/workspace."""
+    state_dir = tmp_path / "state"
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    # Neither ~/work nor ~/workspace exists yet
+    resolved = config.resolve_default_workspace(None)
+    assert resolved == (tmp_path / "workspace").resolve()
+    assert resolved.is_dir()
+
+
+def test_resolve_default_workspace_raises_when_all_candidates_fail(monkeypatch, tmp_path):
+    """RuntimeError is raised when every candidate is unwritable."""
+    import stat, pytest
+    # Make tmp_path read-only so mkdir inside it fails
+    tmp_path.chmod(stat.S_IRUSR | stat.S_IXUSR)
+    state_dir = tmp_path / "state"
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    monkeypatch.delenv("HERMES_WEBUI_DEFAULT_WORKSPACE", raising=False)
+    try:
+        with pytest.raises(RuntimeError, match="Could not create or access"):
+            config.resolve_default_workspace(None)
+    finally:
+        tmp_path.chmod(stat.S_IRWXU)  # restore for cleanup
+
+
+def test_workspace_candidates_deduplicates_home_workspace(monkeypatch, tmp_path):
+    """~/workspace must appear at most once in the candidates list even if it exists."""
+    ws = tmp_path / "workspace"
+    ws.mkdir()
+    state_dir = tmp_path / "state"
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    monkeypatch.delenv("HERMES_WEBUI_DEFAULT_WORKSPACE", raising=False)
+    candidates = config._workspace_candidates(None)
+    paths = [str(p) for p in candidates]
+    assert paths.count(str(ws.resolve())) <= 1, "~/workspace must not appear twice"
+
+
+def test_env_var_workspace_takes_priority_over_passed_raw(monkeypatch, tmp_path):
+    """HERMES_WEBUI_DEFAULT_WORKSPACE env var overrides a None raw arg but not a valid one."""
+    env_ws = tmp_path / "env_workspace"
+    env_ws.mkdir()
+    state_dir = tmp_path / "state"
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    monkeypatch.setenv("HERMES_WEBUI_DEFAULT_WORKSPACE", str(env_ws))
+    # When raw is None, env var should be used
+    resolved = config.resolve_default_workspace(None)
+    assert resolved == env_ws.resolve()
+
+
+def test_ensure_workspace_dir_returns_false_for_unwritable_path(monkeypatch, tmp_path):
+    """_ensure_workspace_dir returns False for a path that can't be created."""
+    import stat
+    # Make parent read-only so mkdir fails
+    parent = tmp_path / "ro_parent"
+    parent.mkdir()
+    parent.chmod(stat.S_IRUSR | stat.S_IXUSR)
+    try:
+        result = config._ensure_workspace_dir(parent / "child")
+        assert result is False
+    finally:
+        parent.chmod(stat.S_IRWXU)
+
+
+def test_env_var_wins_over_settings_json_on_startup(monkeypatch, tmp_path):
+    """HERMES_WEBUI_DEFAULT_WORKSPACE must not be overridden by settings.json at startup.
+
+    Regression for GitHub issue #609: Docker deployments set the env var to a
+    volume mount, but settings.json from a previous container run used to
+    silently win, reverting the files panel to the old path.
+    """
+    import json as _json
+    import os as _os
+
+    env_ws = tmp_path / "env_workspace"
+    env_ws.mkdir()
+    settings_ws = tmp_path / "settings_workspace"
+    settings_ws.mkdir()
+    state_dir = tmp_path / "state"
+    state_dir.mkdir()
+    settings_file = state_dir / "settings.json"
+    settings_file.write_text(
+        _json.dumps({"default_workspace": str(settings_ws)}), encoding="utf-8"
+    )
+
+    monkeypatch.setattr(config, "HOME", tmp_path)
+    monkeypatch.setattr(config, "STATE_DIR", state_dir)
+    monkeypatch.setattr(config, "SETTINGS_FILE", settings_file)
+    # Simulate DEFAULT_WORKSPACE already set correctly from env var at import time
+    monkeypatch.setattr(config, "DEFAULT_WORKSPACE", env_ws.resolve())
+    monkeypatch.setenv("HERMES_WEBUI_DEFAULT_WORKSPACE", str(env_ws))
+
+    # Execute the patched startup block logic inline — env var present → skip override
+    current_ws = config.DEFAULT_WORKSPACE
+    startup_settings = config.load_settings()
+    if not _os.getenv("HERMES_WEBUI_DEFAULT_WORKSPACE"):
+        # This branch must be skipped because env var is set
+        current_ws = config.resolve_default_workspace(
+            startup_settings.get("default_workspace")
+        )
+
+    # env var was set → the if block was skipped → env path wins over settings.json
+    assert current_ws == env_ws.resolve(), (
+        f"Expected {env_ws.resolve()}, got {current_ws}. "
+        "settings.json must not override HERMES_WEBUI_DEFAULT_WORKSPACE."
+    )
+

tests/test_font_size_setting.py

@@ -0,0 +1,228 @@
+"""Tests for font size setting (#833) — 3-toggle Small/Default/Large in Appearance."""
+import os
+import re
+
+_SRC = os.path.join(os.path.dirname(__file__), "..")
+
+def _read(name):
+    return open(os.path.join(_SRC, name), encoding="utf-8").read()
+
+
+class TestFontSizeCssModifiers:
+    """CSS must define font-size overrides for small and large via data attribute."""
+
+    def test_small_font_size_rule_exists(self):
+        css = _read("static/style.css")
+        assert 'data-font-size="small"' in css, (
+            "style.css must have :root[data-font-size=\"small\"] font-size rule"
+        )
+
+    def test_large_font_size_rule_exists(self):
+        css = _read("static/style.css")
+        assert 'data-font-size="large"' in css, (
+            "style.css must have :root[data-font-size=\"large\"] font-size rule"
+        )
+
+    def test_small_is_smaller_than_default(self):
+        css = _read("static/style.css")
+        # Match both compact {font-size:12px} and spaced { font-size: 12px; } formats
+        m_small = re.search(r':root\[data-font-size="small"\][^{]*\{[^}]*font-size:\s*(\d+)px', css)
+        m_large = re.search(r':root\[data-font-size="large"\][^{]*\{[^}]*font-size:\s*(\d+)px', css)
+        assert m_small and m_large, "Both small and large font-size rules must set px values"
+        assert int(m_small.group(1)) < 14, "Small font size must be < 14px (default)"
+        assert int(m_large.group(1)) > 14, "Large font size must be > 14px (default)"
+
+
+class TestFontSizeBootScript:
+    """The boot script must apply font size from localStorage before page renders."""
+
+    def test_boot_script_reads_hermes_font_size(self):
+        html = _read("static/index.html")
+        assert "hermes-font-size" in html, (
+            "index.html boot script must read 'hermes-font-size' from localStorage"
+        )
+        assert "data-font-size" in html, (
+            "boot script must set document.documentElement.dataset.fontSize"
+        )
+
+    def test_font_size_picker_html_present(self):
+        html = _read("static/index.html")
+        assert "fontSizePickerGrid" in html, (
+            "Appearance pane must contain a fontSizePickerGrid element"
+        )
+        assert "settingsFontSize" in html, (
+            "Appearance pane must contain a hidden #settingsFontSize input"
+        )
+        assert "font-size-pick-btn" in html, (
+            "Font size picker buttons must have font-size-pick-btn class"
+        )
+
+    def test_three_font_size_values_present(self):
+        html = _read("static/index.html")
+        assert 'data-font-size-val="small"' in html, "Small button must exist"
+        assert 'data-font-size-val="default"' in html, "Default button must exist"
+        assert 'data-font-size-val="large"' in html, "Large button must exist"
+
+    def test_font_size_picker_not_duplicated(self):
+        """Regression guard: the font size picker grid must appear exactly once
+        in index.html. Earlier versions of this PR accidentally injected the
+        block into both settingsPaneAppearance (correct) and
+        settingsPanePreferences (copy-paste duplicate), creating duplicate IDs
+        that break _syncFontSizePicker visual sync on one of the grids."""
+        html = _read("static/index.html")
+        assert html.count('id="fontSizePickerGrid"') == 1, (
+            "fontSizePickerGrid must appear exactly once — duplicate IDs "
+            "violate HTML spec and break querySelectorAll-based sync."
+        )
+        assert html.count('id="settingsFontSize"') == 1, (
+            "settingsFontSize hidden input must appear exactly once"
+        )
+
+    def test_font_size_picker_lives_in_appearance_pane(self):
+        """The font size picker must be under settingsPaneAppearance,
+        not Preferences/System/Conversation."""
+        html = _read("static/index.html")
+        appearance_start = html.find('id="settingsPaneAppearance"')
+        next_pane_markers = [
+            'id="settingsPanePreferences"',
+            'id="settingsPaneSystem"',
+            'id="settingsPaneConversation"',
+        ]
+        next_pane_starts = [
+            html.find(m, appearance_start + 1) for m in next_pane_markers
+        ]
+        after_appearance = min(
+            [p for p in next_pane_starts if p != -1] or [len(html)]
+        )
+        picker_pos = html.find('id="fontSizePickerGrid"')
+        assert appearance_start != -1, "settingsPaneAppearance not found"
+        assert picker_pos != -1, "fontSizePickerGrid not found"
+        assert appearance_start < picker_pos < after_appearance, (
+            "Font size picker must live inside settingsPaneAppearance "
+            "(same section as Theme and Skin)"
+        )
+
+
+class TestFontSizeJsFunctions:
+    """JS must expose _pickFontSize, _applyFontSize, and _syncFontSizePicker."""
+
+    def test_pick_font_size_function_exists(self):
+        boot = _read("static/boot.js")
+        assert "function _pickFontSize(" in boot, (
+            "boot.js must define _pickFontSize()"
+        )
+
+    def test_apply_font_size_function_exists(self):
+        boot = _read("static/boot.js")
+        assert "function _applyFontSize(" in boot, (
+            "boot.js must define _applyFontSize()"
+        )
+
+    def test_sync_font_size_picker_function_exists(self):
+        boot = _read("static/boot.js")
+        assert "function _syncFontSizePicker(" in boot, (
+            "boot.js must define _syncFontSizePicker()"
+        )
+
+    def test_pick_font_size_persists_to_localstorage(self):
+        boot = _read("static/boot.js")
+        idx = boot.find("function _pickFontSize(")
+        block = boot[idx:idx+400]
+        assert "localStorage.setItem('hermes-font-size'" in block, (
+            "_pickFontSize must persist choice to localStorage"
+        )
+
+    def test_apply_font_size_sets_data_attribute(self):
+        boot = _read("static/boot.js")
+        idx = boot.find("function _applyFontSize(")
+        block = boot[idx:idx+300]
+        assert "dataset.fontSize" in block, (
+            "_applyFontSize must set document.documentElement.dataset.fontSize"
+        )
+
+
+class TestFontSizeI18nCoverage:
+    """All locales must include the font size i18n keys."""
+
+    def _get_locale_keys(self, src, locale_marker_after, stop_marker):
+        """Extract keys from a locale block."""
+        start = src.find(locale_marker_after)
+        if start < 0:
+            return set()
+        end = src.find(stop_marker, start)
+        block = src[start:end if end > 0 else start + 20000]
+        return set(re.findall(r"(\w[\w_]+):", block))
+
+    REQUIRED_KEYS = {"settings_label_font_size", "font_size_small", "font_size_default", "font_size_large"}
+
+    def test_all_locales_have_font_size_keys(self):
+        src = _read("static/i18n.js")
+        count = src.count("settings_label_font_size")
+        # 6 locales: en, ru, es, de, zh, zh-Hant
+        assert count >= 6, (
+            f"settings_label_font_size must appear in all 6 locales, found {count}"
+        )
+
+    def test_font_size_small_key_in_all_locales(self):
+        src = _read("static/i18n.js")
+        count = src.count("font_size_small")
+        assert count >= 6, f"font_size_small must appear in all 6 locales, found {count}"
+
+    def test_font_size_large_key_in_all_locales(self):
+        src = _read("static/i18n.js")
+        count = src.count("font_size_large")
+        assert count >= 6, f"font_size_large must appear in all 6 locales, found {count}"
+
+
+class TestFontSizeCssTargetedOverrides:
+    """CSS must override px-unit text in key UI elements, not just :root font-size.
+
+    The original PR only set :root font-size, but the stylesheet uses hardcoded px
+    values throughout — changing :root has no effect on those. This test class locks
+    in the targeted overrides for the most visible UI surfaces.
+    """
+
+    def test_msg_body_overridden_for_small(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="small"] .msg-body' in css, \
+            "Chat message text must be explicitly scaled for small"
+
+    def test_msg_body_overridden_for_large(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="large"] .msg-body' in css, \
+            "Chat message text must be explicitly scaled for large"
+
+    def test_session_item_overridden_for_small(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="small"] .session-item' in css, \
+            "Sidebar session list text must be explicitly scaled for small"
+
+    def test_session_item_overridden_for_large(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="large"] .session-item' in css, \
+            "Sidebar session list text must be explicitly scaled for large"
+
+    def test_composer_overridden_for_small(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="small"] #msg' in css, \
+            "Composer textarea must be explicitly scaled for small"
+
+    def test_composer_overridden_for_large(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="large"] #msg' in css, \
+            "Composer textarea must be explicitly scaled for large"
+        # Large composer must not equal the default 16px — that's a no-op
+        import re
+        m = re.search(r':root\[data-font-size="large"\] #msg \{ font-size: (\d+)px', css)
+        assert m and int(m.group(1)) != 16, \
+            "Large composer font-size must differ from default (16px) to have visible effect"
+
+    def test_file_item_overridden_for_small(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="small"] .file-item' in css, \
+            "Workspace file tree text must be explicitly scaled for small"
+
+    def test_file_item_overridden_for_large(self):
+        css = _read("static/style.css")
+        assert ':root[data-font-size="large"] .file-item' in css, \
+            "Workspace file tree text must be explicitly scaled for large"

tests/test_gateway_sync.py

@@ -0,0 +1,631 @@
+"""
+Tests for Phase 1: Real-time Gateway Session Sync.
+
+Tests are ordered TDD-style:
+  1. Gateway sessions appear in /api/sessions when setting enabled
+  2. Gateway sessions excluded when setting disabled
+  3. Gateway sessions have correct metadata (source_tag, is_cli_session)
+  4. SSE stream endpoint opens and receives events
+  5. Watcher detects new sessions inserted into state.db
+  6. Settings UI has renamed label
+"""
+import json
+import os
+import pathlib
+import sqlite3
+import time
+import urllib.error
+import urllib.request
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
+from tests._pytest_port import BASE
+
+
+def get(path):
+    with urllib.request.urlopen(BASE + path, timeout=10) as r:
+        return json.loads(r.read()), r.status
+
+
+def post(path, body=None):
+    data = json.dumps(body or {}).encode()
+    req = urllib.request.Request(BASE + path, data=data,
+                                  headers={"Content-Type": "application/json"})
+    try:
+        with urllib.request.urlopen(req, timeout=10) as r:
+            return json.loads(r.read()), r.status
+    except urllib.error.HTTPError as e:
+        try:
+            return json.loads(e.read()), e.code
+        except Exception:
+            return {}, e.code
+
+
+def _get_test_state_dir():
+    """Return the test state directory (matches conftest.py TEST_STATE_DIR).
+
+    conftest.py sets HERMES_WEBUI_TEST_STATE_DIR in the test-process environment
+    (via os.environ.setdefault) so that tests writing directly to state.db always
+    use the same path the test server was started with.  If the env var is not
+    set (e.g. when running this file standalone), fall back to the conftest
+    formula: HERMES_HOME/webui-mvp-test.
+    """
+    # Use _pytest_port which applies the same auto-derivation as conftest.py
+    from tests._pytest_port import TEST_STATE_DIR as _ptsd
+    return _ptsd
+
+
+def _get_state_db_path():
+    """Return path to the test state.db."""
+    return _get_test_state_dir() / 'state.db'
+
+
+def _ensure_state_db():
+    """Create state.db with sessions and messages tables if it doesn't exist.
+    Returns a connection. Does NOT delete existing data (safe for parallel tests).
+    """
+    db_path = _get_state_db_path()
+    db_path.parent.mkdir(parents=True, exist_ok=True)
+    conn = sqlite3.connect(str(db_path))
+    conn.row_factory = sqlite3.Row
+    conn.execute("PRAGMA journal_mode=WAL")
+    conn.executescript("""
+        CREATE TABLE IF NOT EXISTS sessions (
+            id TEXT PRIMARY KEY,
+            source TEXT NOT NULL,
+            user_id TEXT,
+            model TEXT,
+            started_at REAL NOT NULL,
+            message_count INTEGER DEFAULT 0,
+            title TEXT
+        );
+        CREATE TABLE IF NOT EXISTS messages (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            session_id TEXT NOT NULL,
+            role TEXT NOT NULL,
+            content TEXT,
+            timestamp REAL NOT NULL
+        );
+    """)
+    conn.commit()
+    return conn
+
+
+def _insert_gateway_session(conn, session_id='20260401_120000_abcdefgh', source='telegram',
+                             title='Telegram Chat', model='anthropic/claude-sonnet-4-5',
+                             started_at=None, message_count=2):
+    """Insert a gateway session into state.db."""
+    conn.execute(
+        "INSERT OR REPLACE INTO sessions (id, source, title, model, started_at, message_count) "
+        "VALUES (?, ?, ?, ?, ?, ?)",
+        (session_id, source, title, model, started_at or time.time(), message_count)
+    )
+    # Delete any existing messages for this session (idempotent re-insert)
+    conn.execute("DELETE FROM messages WHERE session_id = ?", (session_id,))
+    # Insert some messages
+    conn.execute(
+        "INSERT INTO messages (session_id, role, content, timestamp) VALUES (?, 'user', ?, ?)",
+        (session_id, 'Hello from Telegram', started_at or time.time())
+    )
+    conn.execute(
+        "INSERT INTO messages (session_id, role, content, timestamp) VALUES (?, 'assistant', ?, ?)",
+        (session_id, 'Hi there!', (started_at or time.time()) + 1)
+    )
+    conn.commit()
+
+
+def _remove_test_sessions(conn, *session_ids):
+    """Remove specific test sessions from state.db (parallel-safe cleanup)."""
+    for sid in session_ids:
+        conn.execute("DELETE FROM messages WHERE session_id = ?", (sid,))
+        conn.execute("DELETE FROM sessions WHERE id = ?", (sid,))
+    conn.commit()
+
+
+def _cleanup_state_db():
+    """Remove state.db if it exists (only used for tests that need a blank slate)."""
+    db_path = _get_state_db_path()
+    for p in [db_path, db_path.parent / 'state.db-wal', db_path.parent / 'state.db-shm']:
+        try:
+            p.unlink(missing_ok=True)
+        except Exception:
+            pass
+
+
+# ── Tests ──────────────────────────────────────────────────────────────────
+
+def test_gateway_sessions_appear_when_enabled():
+    """Gateway sessions from state.db appear in /api/sessions when show_cli_sessions is on."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_test_tg_001', source='telegram', title='TG Test Chat')
+
+        # Enable the setting
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        gw_ids = [s['session_id'] for s in sessions if s.get('session_id') == 'gw_test_tg_001']
+        assert len(gw_ids) == 1, f"Expected gateway session gw_test_tg_001, got {[s['session_id'] for s in sessions]}"
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_test_tg_001')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_sessions_without_messages_are_hidden_from_sidebar():
+    """Regression: empty agent session rows must not appear as broken sidebar entries."""
+    conn = _ensure_state_db()
+    empty_sid = 'gw_empty_no_messages_001'
+    try:
+        conn.execute(
+            "INSERT OR REPLACE INTO sessions (id, source, title, model, started_at, message_count) "
+            "VALUES (?, ?, ?, ?, ?, ?)",
+            (empty_sid, 'cron', 'Cron Session', 'openai/gpt-5', time.time(), 0),
+        )
+        conn.execute("DELETE FROM messages WHERE session_id = ?", (empty_sid,))
+        conn.commit()
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        assert empty_sid not in {s.get('session_id') for s in sessions}, (
+            "Agent sessions with no readable message rows should be filtered before "
+            "they reach the sidebar; otherwise clicking them fails during import."
+        )
+    finally:
+        try:
+            _remove_test_sessions(conn, empty_sid)
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_watcher_hides_sessions_without_messages(monkeypatch):
+    """Regression: SSE watcher must use the same importable-agent filter."""
+    conn = _ensure_state_db()
+    empty_sid = 'gw_empty_watcher_001'
+    live_sid = 'gw_live_watcher_001'
+    try:
+        conn.execute(
+            "INSERT OR REPLACE INTO sessions (id, source, title, model, started_at, message_count) "
+            "VALUES (?, ?, ?, ?, ?, ?)",
+            (empty_sid, 'cron', 'Empty Cron Session', 'openai/gpt-5', time.time(), 0),
+        )
+        conn.execute("DELETE FROM messages WHERE session_id = ?", (empty_sid,))
+        _insert_gateway_session(
+            conn,
+            session_id=live_sid,
+            source='cron',
+            title='Live Cron Session',
+            message_count=0,
+        )
+
+        import api.gateway_watcher as gateway_watcher
+
+        monkeypatch.setattr(gateway_watcher, '_get_state_db_path', _get_state_db_path)
+
+        sessions = gateway_watcher._get_agent_sessions_from_db()
+        ids = {s.get('session_id') for s in sessions}
+        live = next((s for s in sessions if s.get('session_id') == live_sid), None)
+
+        assert empty_sid not in ids
+        assert live is not None
+        assert live.get('message_count') == 2, (
+            "Watcher should fall back to actual message rows when stored "
+            "message_count is zero, matching the sidebar route."
+        )
+    finally:
+        try:
+            _remove_test_sessions(conn, empty_sid, live_sid)
+            conn.close()
+        except Exception:
+            pass
+
+
+def test_gateway_sessions_excluded_when_disabled():
+    """Gateway sessions are NOT returned when show_cli_sessions is off."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_test_dc_001', source='discord', title='DC Test Chat')
+
+        # Ensure setting is off
+        post('/api/settings', {'show_cli_sessions': False})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        gw_ids = [s['session_id'] for s in sessions if s.get('session_id') == 'gw_test_dc_001']
+        assert len(gw_ids) == 0, "Gateway session should not appear when setting is off"
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_test_dc_001')
+            conn.close()
+        except Exception:
+            pass
+
+
+def test_gateway_session_has_correct_metadata():
+    """Gateway sessions include source_tag and is_cli_session fields."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_meta_001', source='telegram', title='Meta Test')
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        gw = next((s for s in sessions if s['session_id'] == 'gw_meta_001'), None)
+        assert gw is not None, "Gateway session not found"
+        assert gw.get('source_tag') == 'telegram', f"Expected source_tag=telegram, got {gw.get('source_tag')}"
+        assert gw.get('is_cli_session') is True, "is_cli_session should be True for agent sessions"
+        assert gw.get('title') == 'Meta Test'
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_meta_001')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_session_has_message_count():
+    """Gateway sessions report correct message_count from state.db."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_msg_001', source='discord', title='Msg Count Test', message_count=5)
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        gw = next((s for s in sessions if s['session_id'] == 'gw_msg_001'), None)
+        assert gw is not None
+        assert gw.get('message_count') == 5, f"Expected message_count=5, got {gw.get('message_count')}"
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_msg_001')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_sessions_multiple_sources():
+    """Sessions from multiple gateway sources (telegram, discord, slack) all appear."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_multi_tg', source='telegram', title='TG Chat')
+        _insert_gateway_session(conn, session_id='gw_multi_dc', source='discord', title='DC Chat')
+        _insert_gateway_session(conn, session_id='gw_multi_sl', source='slack', title='SL Chat')
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        gw_ids = {s['session_id'] for s in sessions if s.get('session_id') in ('gw_multi_tg', 'gw_multi_dc', 'gw_multi_sl')}
+        assert len(gw_ids) == 3, f"Expected 3 gateway sessions, got {len(gw_ids)}: {gw_ids}"
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_multi_tg', 'gw_multi_dc', 'gw_multi_sl')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_session_messages_readable():
+    """Gateway session messages can be loaded via /api/session."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='gw_read_001', source='telegram', title='Readable')
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get(f'/api/session?session_id=gw_read_001')
+        assert status == 200
+        msgs = data.get('session', {}).get('messages', [])
+        assert len(msgs) >= 2, f"Expected at least 2 messages, got {len(msgs)}"
+        assert msgs[0].get('role') == 'user'
+        assert msgs[0].get('content') == 'Hello from Telegram'
+    finally:
+        try:
+            _remove_test_sessions(conn, 'gw_read_001')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_importing_older_gateway_session_preserves_original_timestamps_and_order():
+    """Importing an older gateway session should not bump it above newer WebUI sessions."""
+    conn = _ensure_state_db()
+    older_started_at = time.time() - 1800
+    imported_sid = 'gw_import_old_001'
+    newer_webui_sid = None
+    try:
+        newer_webui, status = post('/api/session/new', {'model': 'openai/gpt-5'})
+        assert status == 200, newer_webui
+        newer_webui_sid = newer_webui['session']['session_id']
+
+        rename, rename_status = post(
+            '/api/session/rename',
+            {'session_id': newer_webui_sid, 'title': 'Newer WebUI Session'},
+        )
+        assert rename_status == 200, rename
+
+        _insert_gateway_session(
+            conn,
+            session_id=imported_sid,
+            source='discord',
+            title='Older imported gateway session',
+            started_at=older_started_at,
+        )
+        post('/api/settings', {'show_cli_sessions': True})
+
+        imported, imported_status = post('/api/session/import_cli', {'session_id': imported_sid})
+        assert imported_status == 200, imported
+        imported_session = imported['session']
+        assert abs(imported_session['created_at'] - older_started_at) < 2, imported_session
+        assert abs(imported_session['updated_at'] - older_started_at) < 5, imported_session
+
+        sessions_payload, sessions_status = get('/api/sessions')
+        assert sessions_status == 200, sessions_payload
+        ordered_ids = [item['session_id'] for item in sessions_payload.get('sessions', [])]
+        assert newer_webui_sid in ordered_ids, ordered_ids
+        assert imported_sid in ordered_ids, ordered_ids
+        assert ordered_ids.index(newer_webui_sid) < ordered_ids.index(imported_sid), ordered_ids
+    finally:
+        try:
+            _remove_test_sessions(conn, imported_sid)
+            conn.close()
+        except Exception:
+            pass
+        if imported_sid:
+            try:
+                post('/api/session/delete', {'session_id': imported_sid})
+            except Exception:
+                pass
+        if newer_webui_sid:
+            try:
+                post('/api/session/delete', {'session_id': newer_webui_sid})
+            except Exception:
+                pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+
+def test_gateway_sse_stream_endpoint_exists():
+    """GET /api/sessions/gateway/stream returns a response (200 or 200-range)."""
+    # The SSE endpoint requires show_cli_sessions to be enabled
+    post('/api/settings', {'show_cli_sessions': True})
+    try:
+        req = urllib.request.Request(BASE + '/api/sessions/gateway/stream')
+        with urllib.request.urlopen(req, timeout=5) as r:
+            assert r.status in (200, 204), f"Expected 200/204, got {r.status}"
+            # SSE should have content-type text/event-stream
+            ctype = r.headers.get('Content-Type', '')
+            assert 'text/event-stream' in ctype, f"Expected text/event-stream, got {ctype}"
+    except Exception as e:
+        # Timeout is acceptable — means the connection is held open (SSE behavior)
+        if 'timed out' in str(e).lower() or 'timeout' in str(e).lower():
+            pass  # Good: SSE keeps the connection open
+        else:
+            raise
+    finally:
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_sse_stream_probe_reports_status():
+    """Probe mode returns JSON watcher status instead of holding open an SSE stream."""
+    post('/api/settings', {'show_cli_sessions': True})
+    try:
+        req = urllib.request.Request(BASE + '/api/sessions/gateway/stream?probe=1')
+        with urllib.request.urlopen(req, timeout=5) as r:
+            assert r.status == 200, f"Expected 200, got {r.status}"
+            ctype = r.headers.get('Content-Type', '')
+            assert 'application/json' in ctype, f"Expected application/json, got {ctype}"
+            data = json.loads(r.read().decode('utf-8'))
+            assert data['enabled'] is True
+            assert 'watcher_running' in data
+            assert data['fallback_poll_ms'] == 30000
+    finally:
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_webui_sessions_not_duplicated():
+    """If a session_id exists both in WebUI store and state.db, it's not duplicated."""
+    # Create a WebUI session with a known ID
+    body = {}
+    d, _ = post('/api/session/new', body)
+    webui_sid = d['session']['session_id']
+
+    try:
+        # Insert the same session_id into state.db as a gateway session
+        conn = _ensure_state_db()
+        _insert_gateway_session(conn, session_id=webui_sid, source='telegram', title='Dup Test')
+        conn.close()
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        matching = [s for s in sessions if s['session_id'] == webui_sid]
+        assert len(matching) == 1, f"Expected 1 entry for {webui_sid}, got {len(matching)}"
+    finally:
+        try:
+            conn2 = sqlite3.connect(str(_get_state_db_path()))
+            _remove_test_sessions(conn2, webui_sid)
+            conn2.close()
+        except Exception:
+            pass
+        post('/api/session/delete', {'session_id': webui_sid})
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_gateway_sessions_no_state_db():
+    """When state.db doesn't exist, /api/sessions works fine (no gateway sessions)."""
+    _cleanup_state_db()
+
+    post('/api/settings', {'show_cli_sessions': True})
+    try:
+        data, status = get('/api/sessions')
+        assert status == 200
+        # Should succeed with just webui sessions (or empty)
+        assert 'sessions' in data
+    finally:
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+def test_cli_sessions_still_work():
+    """CLI sessions (source='cli') still appear alongside gateway sessions."""
+    conn = _ensure_state_db()
+    try:
+        _insert_gateway_session(conn, session_id='cli_legacy_001', source='cli', title='CLI Legacy')
+        _insert_gateway_session(conn, session_id='gw_new_001', source='telegram', title='GW New')
+
+        post('/api/settings', {'show_cli_sessions': True})
+
+        data, status = get('/api/sessions')
+        assert status == 200
+        sessions = data.get('sessions', [])
+        agent_ids = {s['session_id'] for s in sessions if s.get('session_id') in ('cli_legacy_001', 'gw_new_001')}
+        assert len(agent_ids) == 2, f"Expected 2 agent sessions (cli + gateway), got {len(agent_ids)}"
+    finally:
+        try:
+            _remove_test_sessions(conn, 'cli_legacy_001', 'gw_new_001')
+            conn.close()
+        except Exception:
+            pass
+        post('/api/settings', {'show_cli_sessions': False})
+
+
+# ── Unit tests for _gateway_sse_probe_payload ────────────────────────────────
+# These replace the deleted repo-root test_gateway_sse_probe_unit.py and account
+# for the watcher_alive check (thread existence + is_alive()).
+
+import sys
+import threading
+sys.path.insert(0, str(REPO_ROOT))
+from api.routes import _gateway_sse_probe_payload
+
+
+def test_probe_payload_when_disabled():
+    """Probe returns 404 when show_cli_sessions is False."""
+    body, status = _gateway_sse_probe_payload({'show_cli_sessions': False}, watcher=None)
+    assert status == 404
+    assert body['ok'] is False
+    assert body['enabled'] is False
+    assert body['watcher_running'] is False
+    assert body['error'] == 'agent sessions not enabled'
+    assert body['fallback_poll_ms'] == 30000
+
+
+def test_probe_payload_when_watcher_missing():
+    """Probe returns 503 when enabled but no watcher instance."""
+    body, status = _gateway_sse_probe_payload({'show_cli_sessions': True}, watcher=None)
+    assert status == 503
+    assert body['ok'] is False
+    assert body['enabled'] is True
+    assert body['watcher_running'] is False
+    assert body['error'] == 'watcher not started'
+    assert body['fallback_poll_ms'] == 30000
+
+
+def test_probe_payload_when_watcher_instance_no_thread():
+    """Probe returns 503 when watcher exists but _thread attribute is missing/None."""
+    class _FakeWatcher:
+        _thread = None
+    body, status = _gateway_sse_probe_payload({'show_cli_sessions': True}, watcher=_FakeWatcher())
+    assert status == 503
+    assert body['watcher_running'] is False
+
+
+def test_probe_payload_when_watcher_thread_alive():
+    """Probe returns 200 when enabled and watcher thread is alive."""
+    class _FakeWatcher:
+        pass
+    w = _FakeWatcher()
+    t = threading.Thread(target=lambda: None)
+    t.daemon = True
+    t.start()
+    w._thread = t
+    # Thread may finish fast — loop-start a live daemon thread for reliability
+    import time as _time
+    done = threading.Event()
+    live = threading.Thread(target=done.wait, daemon=True)
+    live.start()
+    w._thread = live
+    try:
+        body, status = _gateway_sse_probe_payload({'show_cli_sessions': True}, watcher=w)
+        assert status == 200
+        assert body['ok'] is True
+        assert body['watcher_running'] is True
+        assert body['fallback_poll_ms'] == 30000
+    finally:
+        done.set()
+        live.join(timeout=1)
+
+
+def test_probe_payload_when_watcher_thread_dead():
+    """Probe returns 503 when watcher instance exists but thread has exited."""
+    class _FakeWatcher:
+        pass
+    w = _FakeWatcher()
+    t = threading.Thread(target=lambda: None)
+    t.start()
+    t.join()  # wait for it to finish
+    w._thread = t
+    body, status = _gateway_sse_probe_payload({'show_cli_sessions': True}, watcher=w)
+    assert status == 503
+    assert body['watcher_running'] is False
+    assert body['ok'] is False
+
+
+def test_gateway_watcher_is_alive_public_method():
+    """GatewayWatcher.is_alive() is the public API the probe uses. Cover all
+    three states: before start(), while running, after stop()."""
+    from api.gateway_watcher import GatewayWatcher
+    w = GatewayWatcher()
+    # Before start(): no thread
+    assert w.is_alive() is False, "is_alive() must be False before start()"
+    # After start(): thread running
+    w.start()
+    try:
+        assert w.is_alive() is True, "is_alive() must be True while running"
+    finally:
+        w.stop()
+    # After stop(): thread cleared
+    assert w.is_alive() is False, "is_alive() must be False after stop()"
+
+
+def test_probe_payload_prefers_public_is_alive():
+    """Regression guard: _gateway_sse_probe_payload must call watcher.is_alive()
+    rather than poking at _thread directly when the public method exists."""
+    calls = []
+
+    class _WatcherWithPublicApi:
+        def is_alive(self):
+            calls.append('is_alive')
+            return True
+        # _thread is deliberately absent — must not be accessed.
+
+    body, status = _gateway_sse_probe_payload(
+        {'show_cli_sessions': True},
+        watcher=_WatcherWithPublicApi(),
+    )
+    assert status == 200
+    assert body['watcher_running'] is True
+    assert calls == ['is_alive'], (
+        "probe must prefer the public is_alive() method over poking _thread"
+    )

tests/test_ime_composition.py

@@ -0,0 +1,61 @@
+import pathlib
+import re
+
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
+BOOT_JS = (REPO_ROOT / "static" / "boot.js").read_text(encoding="utf-8")
+UI_JS = (REPO_ROOT / "static" / "ui.js").read_text(encoding="utf-8")
+SESSIONS_JS = (REPO_ROOT / "static" / "sessions.js").read_text(encoding="utf-8")
+
+
+def _ime_guarded_enter_pattern(event_var_pattern, require_no_shift=False):
+    no_shift = rf"\s*&&\s*!\s*{event_var_pattern}\.shiftKey" if require_no_shift else ""
+    return (
+        rf"if\s*\(\s*{event_var_pattern}\.key\s*===\s*'Enter'{no_shift}\s*\)\s*\{{\s*"
+        rf"if\s*\(\s*{event_var_pattern}\.isComposing\s*\)\s*"
+        rf"(?:\{{\s*return\s*;?\s*\}}|return\s*;?)"
+    )
+
+
+def test_boot_chat_enter_send_respects_ime_composition():
+    assert re.search(
+        _ime_guarded_enter_pattern("e"),
+        BOOT_JS,
+        re.DOTALL,
+    ), "Chat composer Enter handler must ignore IME composition Enter in static/boot.js"
+    assert re.search(
+        _ime_guarded_enter_pattern("e", require_no_shift=True),
+        BOOT_JS,
+        re.DOTALL,
+    ), "Command dropdown Enter handler must ignore IME composition Enter in static/boot.js"
+
+
+def test_ui_enter_submit_paths_respect_ime_composition():
+    assert re.search(
+        rf"document\.addEventListener\('keydown',e=>\{{[\s\S]*?{_ime_guarded_enter_pattern('e')}",
+        UI_JS,
+        re.DOTALL,
+    ), \
+        "App dialog Enter handler must ignore IME composition Enter in static/ui.js"
+    assert re.search(
+        _ime_guarded_enter_pattern("e", require_no_shift=True),
+        UI_JS,
+        re.DOTALL,
+    ), \
+        "Message edit Enter-to-save handler must ignore IME composition Enter in static/ui.js"
+    assert re.search(
+        rf"inp\.onkeydown=\(e2\)=>\{{\s*{_ime_guarded_enter_pattern('e2')}",
+        UI_JS,
+        re.DOTALL,
+    ), \
+        "Workspace rename Enter handler must ignore IME composition Enter in static/ui.js"
+
+
+def test_sessions_enter_submit_paths_respect_ime_composition():
+    matches = re.findall(
+        _ime_guarded_enter_pattern(r"e2?"),
+        SESSIONS_JS,
+        re.DOTALL,
+    )
+    assert len(matches) >= 3, \
+        "Session and project rename/create Enter handlers must ignore IME composition Enter in static/sessions.js"

tests/test_issue1014_model_not_found.py

@@ -0,0 +1,198 @@
+"""
+Tests for issue #1014 — model-not-found error classification.
+
+Covers:
+  1. streaming.py: 404/model-not-found errors detected and classified as 'model_not_found'
+  2. streaming.py: HTML tags stripped from provider error messages before classification
+  3. static/messages.js: apperror handler has model_not_found branch
+  4. static/i18n.js: model_not_found_label key present in all locales
+  5. streaming.py: model_not_found checked after auth but before generic error
+"""
+import pathlib
+import re
+
+REPO_ROOT = pathlib.Path(__file__).parent.parent.resolve()
+
+
+def _read(rel_path: str) -> str:
+    return (REPO_ROOT / rel_path).read_text(encoding="utf-8")
+
+
+# ── 1. streaming.py: model-not-found error detection ─────────────────────────
+
+class TestStreamingModelNotFoundDetection:
+    """streaming.py must classify 404/model-not-found errors as model_not_found."""
+
+    def test_model_not_found_type_defined_in_streaming(self):
+        """'model_not_found' type must be emitted for 404 errors."""
+        src = _read("api/streaming.py")
+        assert "model_not_found" in src, (
+            "model_not_found type not found in streaming.py — "
+            "404 errors will not be surfaced with a helpful message"
+        )
+
+    def test_is_not_found_flag_defined(self):
+        """_exc_is_not_found variable must exist in the exception handler."""
+        src = _read("api/streaming.py")
+        assert "_exc_is_not_found" in src, (
+            "_exc_is_not_found flag not found in streaming.py"
+        )
+
+    def test_not_found_detects_404(self):
+        """'404' must be part of the model-not-found detection logic."""
+        src = _read("api/streaming.py")
+        idx = src.find("_exc_is_not_found")
+        assert idx != -1, "_exc_is_not_found not found"
+        block = src[idx:idx + 600]
+        assert "'404'" in block or '"404"' in block, (
+            "'404' not in model-not-found detection block"
+        )
+
+    def test_not_found_detects_not_found_string(self):
+        """'not found' must be part of the detection logic."""
+        src = _read("api/streaming.py")
+        idx = src.find("_exc_is_not_found")
+        block = src[idx:idx + 600]
+        assert "not found" in block.lower(), (
+            "'not found' not in model-not-found detection block"
+        )
+
+    def test_not_found_detects_does_not_exist(self):
+        """'does not exist' must be part of the detection logic."""
+        src = _read("api/streaming.py")
+        idx = src.find("_exc_is_not_found")
+        block = src[idx:idx + 600]
+        assert "does not exist" in block.lower(), (
+            "'does not exist' not in model-not-found detection block"
+        )
+
+    def test_not_found_detects_invalid_model(self):
+        """'invalid model' must be part of the detection logic."""
+        src = _read("api/streaming.py")
+        idx = src.find("_exc_is_not_found")
+        block = src[idx:idx + 600]
+        assert "invalid model" in block.lower(), (
+            "'invalid model' not in model-not-found detection block"
+        )
+
+    def test_not_found_hint_mentions_settings(self):
+        """The model_not_found hint must mention Settings or hermes model."""
+        src = _read("api/streaming.py")
+        idx = src.find("model_not_found")
+        block = src[idx:idx + 500]
+        assert "Settings" in block or "hermes model" in block, (
+            "model_not_found hint must mention Settings or hermes model command"
+        )
+
+    def test_not_found_check_order_after_auth(self):
+        """model_not_found must be checked after auth_mismatch (auth first)."""
+        src = _read("api/streaming.py")
+        auth_idx = src.find("elif _exc_is_auth")
+        nf_idx = src.find("elif _exc_is_not_found")
+        assert auth_idx != -1, "_exc_is_auth not found"
+        assert nf_idx != -1, "_exc_is_not_found not found"
+        assert auth_idx < nf_idx, (
+            "auth_mismatch should be checked before model_not_found — "
+            "auth errors must not be mistaken for not-found errors"
+        )
+
+
+# ── 2. streaming.py: HTML sanitization ───────────────────────────────────────
+
+class TestStreamingHtmlSanitization:
+    """Provider error messages containing HTML must be stripped."""
+
+    def test_html_strip_before_classification(self):
+        """HTML tags must be stripped before error classification."""
+        src = _read("api/streaming.py")
+        # Find the HTML sanitization block in the exception handler
+        # It should appear before _exc_lower = err_str.lower()
+        sanitize_idx = src.find("re.sub(r'<[^>]+>'")
+        exc_lower_idx = src.find("_exc_lower = err_str.lower()")
+        assert sanitize_idx != -1, (
+            "HTML tag stripping (re.sub) not found in streaming.py exception handler"
+        )
+        assert exc_lower_idx != -1, "_exc_lower not found"
+        assert sanitize_idx < exc_lower_idx, (
+            "HTML sanitization must happen before error classification"
+        )
+
+    def test_whitespace_normalization(self):
+        """Stripped HTML must have whitespace collapsed."""
+        src = _read("api/streaming.py")
+        sanitize_idx = src.find("re.sub(r'<[^>]+>'")
+        block = src[sanitize_idx:sanitize_idx + 300]
+        assert r"\s+" in block, (
+            "Whitespace normalization (\\s+) not found after HTML strip"
+        )
+
+
+# ── 3. static/messages.js: apperror handler ──────────────────────────────────
+
+class TestApperrorModelNotFound:
+    """messages.js apperror handler must handle model_not_found type."""
+
+    def test_model_not_found_type_handled(self):
+        """apperror handler must check for type='model_not_found'."""
+        src = _read("static/messages.js")
+        assert "model_not_found" in src, (
+            "model_not_found type not handled in messages.js apperror handler"
+        )
+
+    def test_model_not_found_label(self):
+        """'Model not found' label must appear in the error handling."""
+        src = _read("static/messages.js")
+        assert "Model not found" in src, (
+            "'Model not found' label not found in messages.js"
+        )
+
+    def test_is_model_not_found_variable(self):
+        """isModelNotFound variable must be defined."""
+        src = _read("static/messages.js")
+        assert "isModelNotFound" in src, (
+            "isModelNotFound variable not found in messages.js apperror handler"
+        )
+
+
+# ── 4. static/i18n.js: all locales ───────────────────────────────────────────
+
+class TestI18nModelNotFound:
+    """All locales must have model_not_found_label."""
+
+    REQUIRED_KEY = "model_not_found_label"
+
+    def _locale_names(self, src: str) -> list:
+        pattern = re.compile(
+            r"^\s{2}(?:'(?P<quoted>[A-Za-z0-9-]+)'|(?P<plain>[A-Za-z0-9-]+))\s*:\s*\{",
+            re.MULTILINE,
+        )
+        names = []
+        for match in pattern.finditer(src):
+            names.append(match.group("quoted") or match.group("plain"))
+        return names
+
+    def _count_key(self, src: str, key: str) -> int:
+        return len(re.findall(r'\b' + re.escape(key) + r'\b', src))
+
+    def test_all_locales_have_model_not_found_label(self):
+        """model_not_found_label must appear in all locales."""
+        src = _read("static/i18n.js")
+        locale_count = len(self._locale_names(src))
+        count = self._count_key(src, self.REQUIRED_KEY)
+        assert count >= locale_count, (
+            f"model_not_found_label found {count} times, expected >= {locale_count} "
+            f"(one per locale)"
+        )
+
+    def test_english_label_is_plain_string(self):
+        """English model_not_found_label must be a plain string, not a function."""
+        src = _read("static/i18n.js")
+        en_start = src.find("\n  en: {")
+        es_start = src.find("\n  es: {")
+        en_block = src[en_start:es_start]
+        assert self.REQUIRED_KEY in en_block, "Key not in en block"
+        idx = en_block.find(self.REQUIRED_KEY)
+        line = en_block[idx:idx + 200]
+        assert "=>" not in line, (
+            "model_not_found_label should be a plain string, not an arrow function"
+        )

tests/test_issue341.py

@@ -0,0 +1,34 @@
+"""Tests for GitHub issue #341: .msg-body table CSS styles."""
+import os
+
+CSS_PATH = os.path.join(os.path.dirname(__file__), "..", "static", "style.css")
+
+
+def _read_css():
+    with open(CSS_PATH, "r") as f:
+        return f.read()
+
+
+def test_msg_body_table_css_present():
+    css = _read_css()
+    assert ".msg-body table" in css, ".msg-body table rule missing from style.css"
+    assert "border-collapse:collapse" in css, "border-collapse:collapse missing from style.css"
+
+
+def test_msg_body_table_th_td_present():
+    css = _read_css()
+    assert ".msg-body th" in css, ".msg-body th rule missing from style.css"
+    assert ".msg-body td" in css, ".msg-body td rule missing from style.css"
+
+
+def test_msg_body_table_tr_stripe_present():
+    css = _read_css()
+    assert ".msg-body tr:nth-child(even)" in css, ".msg-body tr:nth-child(even) rule missing from style.css"
+
+
+def test_msg_body_light_theme_overrides():
+    css = _read_css()
+    assert ':root:not(.dark) .msg-body th' in css, \
+        'Light-mode override for .msg-body th missing from style.css'
+    assert ':root:not(.dark) .msg-body td' in css, \
+        'Light-mode override for .msg-body td missing from style.css'

tests/test_issue342.py

@@ -0,0 +1,124 @@
+"""
+Tests for GitHub issue #342: auto-link plain URLs in chat messages.
+
+These are structural tests that verify the fix is present in static/ui.js
+without requiring a running server or JavaScript engine.
+"""
+import os
+import re
+
+UI_JS = os.path.join(os.path.dirname(__file__), '..', 'static', 'ui.js')
+
+
+def read_ui_js():
+    with open(UI_JS, 'r') as f:
+        return f.read()
+
+
+def test_autolink_comment_present():
+    """The Autolink comment should be present in renderMd() to document the feature."""
+    content = read_ui_js()
+    assert 'Autolink: convert plain URLs' in content, (
+        "Expected 'Autolink: convert plain URLs' comment not found in static/ui.js. "
+        "Did the autolink pass get added?"
+    )
+
+
+def test_autolink_regex_in_rendermd():
+    """The autolink regex pattern (https?://) should appear in renderMd()."""
+    content = read_ui_js()
+    # Locate the renderMd function body
+    rendermd_start = content.find('function renderMd(raw){')
+    assert rendermd_start != -1, "renderMd function not found in ui.js"
+    # Find the closing brace after renderMd (look for the autolink pattern within it)
+    rendermd_body = content[rendermd_start:rendermd_start + 5000]
+    assert 'https?:\\/\\/' in rendermd_body, (
+        "Autolink regex (https?:\\/\\/) not found inside renderMd() body."
+    )
+
+
+def test_autolink_uses_esc_for_xss_safety():
+    """The autolink code must use esc() to escape the display text of URLs, preventing XSS.
+    Note: esc() is intentionally NOT applied to the href value (that would corrupt & in
+    query strings). It IS applied to the visible link text (esc(clean)) to prevent XSS."""
+    content = read_ui_js()
+    # Find the autolink section (between the SAFE_TAGS pass and paragraph wrap)
+    autolink_idx = content.find('// Autolink: convert plain URLs')
+    assert autolink_idx != -1, "Autolink comment not found in ui.js"
+    # Extract the autolink block (next ~600 chars after the comment)
+    autolink_block = content[autolink_idx:autolink_idx + 600]
+    # esc() must be used on the visible link text to prevent XSS
+    assert 'esc(clean)' in autolink_block, (
+        "Autolink block should use esc(clean) for the link display text (XSS safety), "
+        "but it was not found."
+    )
+    # esc() must NOT be used on the href value — that breaks URLs containing &
+    assert 'href="${esc(clean)}"' not in autolink_block, (
+        "Autolink block should use href=\"${clean}\" (not esc'd) to preserve & in query strings."
+    )
+
+
+def test_autolink_in_inline_md():
+    """The autolink pass should also be present inside the inlineMd() helper."""
+    content = read_ui_js()
+    # Find inlineMd function
+    inline_start = content.find('function inlineMd(t){')
+    assert inline_start != -1, "inlineMd function not found in ui.js"
+    # Find closing brace of inlineMd by looking for 'return t;' followed by '}'
+    inline_end = content.find('return t;\n  }', inline_start)
+    assert inline_end != -1, "Could not locate end of inlineMd function"
+    inline_body = content[inline_start:inline_end + 20]
+    assert 'https?:\\/\\/' in inline_body, (
+        "Autolink regex not found inside inlineMd() — plain URLs in list items "
+        "and blockquotes won't be autolinked."
+    )
+
+
+def test_autolink_after_safe_tags_pass():
+    """The autolink pass must come AFTER the SAFE_TAGS escape pass (ordering matters)."""
+    content = read_ui_js()
+    safe_tags_idx = content.find('s=s.replace(/<\\/?[a-z][^>]*>/gi,tag=>SAFE_TAGS.test(tag)?tag:esc(tag));')
+    autolink_idx = content.find('// Autolink: convert plain URLs')
+    parts_idx = content.find('const parts=s.split(/\\n{2,}/);')
+    assert safe_tags_idx != -1, "SAFE_TAGS pass not found"
+    assert autolink_idx != -1, "Autolink pass not found"
+    assert parts_idx != -1, "Paragraph-wrap parts line not found"
+    assert safe_tags_idx < autolink_idx < parts_idx, (
+        f"Ordering wrong: SAFE_TAGS at {safe_tags_idx}, autolink at {autolink_idx}, "
+        f"parts (paragraph wrap) at {parts_idx}. "
+        "Autolink must come between SAFE_TAGS pass and paragraph wrap."
+    )
+
+
+def test_autolink_target_blank_and_rel():
+    """Autolinked URLs should open in a new tab with rel=noopener for security."""
+    content = read_ui_js()
+    autolink_idx = content.find('// Autolink: convert plain URLs')
+    assert autolink_idx != -1, "Autolink comment not found"
+    # Use a larger window to account for the stash preamble added by the fix
+    autolink_block = content[autolink_idx:autolink_idx + 700]
+    assert 'target="_blank"' in autolink_block, (
+        'Autolinked URLs should have target="_blank"'
+    )
+    assert 'rel="noopener"' in autolink_block, (
+        'Autolinked URLs should have rel="noopener" for security'
+    )
+
+
+def test_safe_tags_includes_anchor():
+    """SAFE_TAGS regex must include 'a' so <a> tags from autolink are not escaped."""
+    content = read_ui_js()
+    # Find the SAFE_TAGS definition line — the pattern contains slashes so we
+    # search for the line directly rather than extracting the regex literal.
+    safe_tags_line = None
+    for line in content.splitlines():
+        if 'const SAFE_TAGS=' in line:
+            safe_tags_line = line
+            break
+    assert safe_tags_line is not None, "SAFE_TAGS const definition not found in ui.js"
+    # The pattern should include 'a' as a tag alternative (e.g. |a|)
+    assert '|a|' in safe_tags_line or '|a)' in safe_tags_line, (
+        f"SAFE_TAGS line does not include 'a' tag — "
+        "<a> tags emitted by autolink would be escaped!\n"
+        f"Line: {safe_tags_line}"
+    )

tests/test_issue347.py

@@ -0,0 +1,348 @@
+"""
+Tests for GitHub issue #347: KaTeX / LaTeX math rendering in chat and workspace previews.
+
+Structural tests — no server required. Verify:
+- renderMd() stashes and restores $..$ and $$...$$ math delimiters
+- KaTeX lazy-load function exists and follows the mermaid pattern
+- KaTeX JS loaded from CDN with SRI integrity hash
+- KaTeX CSS loaded in index.html with SRI hash
+- CSS rules present for .katex-block and .katex-inline
+- SAFE_TAGS updated to allow <span> (for inline math)
+- renderKatexBlocks() is wired into the requestAnimationFrame call
+"""
+import pathlib
+import re
+
+REPO = pathlib.Path(__file__).parent.parent
+UI_JS   = (REPO / 'static' / 'ui.js').read_text(encoding='utf-8')
+INDEX   = (REPO / 'static' / 'index.html').read_text(encoding='utf-8')
+CSS     = (REPO / 'static' / 'style.css').read_text(encoding='utf-8')
+
+
+# ── renderMd pipeline ──────────────────────────────────────────────────────────
+
+def test_display_math_stash_present():
+    """renderMd must stash $$...$$ display math before other processing."""
+    assert r'\$\$([\s\S]+?)\$\$' in UI_JS or '$$' in UI_JS, \
+        'Display math $$..$$ stash regex not found in ui.js'
+    # The stash uses \\x00M token
+    assert '\\x00M' in UI_JS, 'Math stash token \\x00M not found in renderMd'
+
+
+def test_inline_math_stash_present():
+    """renderMd must stash $..$ inline math."""
+    # Inline math regex must be present
+    assert 'math_stash' in UI_JS, 'math_stash array not found in renderMd'
+
+
+def test_katex_block_placeholder_emitted():
+    """renderMd restore pass must emit .katex-block divs for display math."""
+    assert 'katex-block' in UI_JS, \
+        '.katex-block placeholder div not emitted by renderMd restore pass'
+
+
+def test_katex_inline_placeholder_emitted():
+    """renderMd restore pass must emit .katex-inline spans for inline math."""
+    assert 'katex-inline' in UI_JS, \
+        '.katex-inline placeholder span not emitted by renderMd restore pass'
+
+
+def test_data_katex_attribute_present():
+    """Placeholders must carry data-katex attribute for display/inline distinction."""
+    assert 'data-katex' in UI_JS, \
+        'data-katex attribute not found — renderKatexBlocks cannot distinguish display from inline'
+
+
+# ── renderKatexBlocks() ────────────────────────────────────────────────────────
+
+def test_render_katex_blocks_function_exists():
+    """renderKatexBlocks() function must exist in ui.js."""
+    assert 'function renderKatexBlocks()' in UI_JS, \
+        'renderKatexBlocks() function not found in ui.js'
+
+
+def test_katex_lazy_load_follows_mermaid_pattern():
+    """KaTeX must use the same lazy-load pattern as mermaid (load on first use)."""
+    assert '_katexLoading' in UI_JS, '_katexLoading flag not found'
+    assert '_katexReady' in UI_JS,   '_katexReady flag not found'
+
+
+def test_katex_js_loaded_from_cdn():
+    """KaTeX JS must be loaded from jsdelivr CDN."""
+    assert 'katex@0.16' in UI_JS, \
+        'KaTeX JS CDN URL not found in ui.js — expected katex@0.16.x'
+
+
+def test_katex_js_has_sri_hash():
+    """KaTeX JS CDN tag must have an SRI integrity hash."""
+    # The hash is in the script.integrity assignment
+    assert "script.integrity='sha384-" in UI_JS or 'script.integrity="sha384-' in UI_JS, \
+        'KaTeX JS SRI integrity hash not found in ui.js'
+
+
+def test_katex_display_mode_used():
+    """renderKatexBlocks must pass displayMode based on data-katex attribute."""
+    assert 'displayMode' in UI_JS, \
+        'displayMode not passed to katex.render() — display math will render inline'
+
+
+def test_katex_throw_on_error_false():
+    """KaTeX must be configured with throwOnError:false to degrade gracefully."""
+    assert 'throwOnError:false' in UI_JS, \
+        'throwOnError:false not set — bad LaTeX will throw and break the message'
+
+
+def test_render_katex_blocks_wired_into_raf():
+    """renderKatexBlocks() must be called in the same requestAnimationFrame as renderMermaidBlocks()."""
+    # Check that renderKatexBlocks appears somewhere near requestAnimationFrame
+    raf_idx = UI_JS.find('requestAnimationFrame')
+    # Find the rAF call that also contains renderKatexBlocks
+    has_katex_in_raf = any(
+        'renderKatexBlocks' in UI_JS[m.start():m.start()+200]
+        for m in re.finditer(r'requestAnimationFrame', UI_JS)
+    )
+    assert has_katex_in_raf, \
+        'renderKatexBlocks() not found in any requestAnimationFrame call — math will not render'
+
+
+# ── index.html ────────────────────────────────────────────────────────────────
+
+def test_katex_css_in_index_html():
+    """KaTeX CSS must be loaded in index.html."""
+    assert 'katex@0.16' in INDEX, \
+        'KaTeX CSS CDN link not found in index.html'
+
+
+def test_katex_css_has_sri_hash():
+    """KaTeX CSS link in index.html must have an SRI integrity hash."""
+    assert 'sha384-5TcZemv2l' in INDEX or 'integrity' in INDEX and 'katex' in INDEX, \
+        'KaTeX CSS SRI integrity hash not found in index.html'
+
+
+# ── style.css ─────────────────────────────────────────────────────────────────
+
+def test_katex_block_css_present():
+    """.katex-block CSS rule must exist for centered display math."""
+    assert '.katex-block' in CSS, \
+        '.katex-block CSS rule missing from style.css — display math will have no layout'
+
+
+def test_katex_inline_css_present():
+    """.katex-inline CSS rule must exist."""
+    assert '.katex-inline' in CSS, \
+        '.katex-inline CSS rule missing from style.css'
+
+
+def test_katex_block_text_align_center():
+    """.katex-block must be text-align:center for display math."""
+    assert 'text-align:center' in CSS, \
+        'text-align:center not found for .katex-block'
+
+
+# ── SAFE_TAGS ──────────────────────────────────────────────────────────────────
+
+def test_safe_tags_includes_span():
+    """SAFE_TAGS must include <span> to allow .katex-inline spans through the escape pass."""
+    # The SAFE_TAGS regex should contain 'span'
+    safe_tags_match = re.search(r'SAFE_TAGS\s*=\s*/.*?/i', UI_JS)
+    assert safe_tags_match, 'SAFE_TAGS pattern not found in ui.js'
+    assert 'span' in safe_tags_match.group(), \
+        '<span> not in SAFE_TAGS — inline math spans will be HTML-escaped and rendered as text'
+
+
+# ── Stash ordering: fence must protect code spans from math extraction ─────────
+
+WORKSPACE_JS = (REPO / 'static' / 'workspace.js').read_text(encoding='utf-8')
+
+
+def test_fence_stash_before_math_stash():
+    """fence_stash must be initialized and populated BEFORE math_stash in renderMd.
+
+    If math_stash runs first, dollar signs inside backtick code spans are extracted
+    as math, leaving placeholder tokens inside the stashed code string. The code span
+    then renders with KaTeX inside <code> instead of the literal dollar-sign text.
+    """
+    fence_pos = UI_JS.find("const fence_stash=[]")
+    math_pos = UI_JS.find("const math_stash=[]")
+    assert fence_pos != -1, "fence_stash not found in renderMd"
+    assert math_pos != -1, "math_stash not found in renderMd"
+    assert fence_pos < math_pos, (
+        "fence_stash must be declared BEFORE math_stash in renderMd "
+        f"(fence at char {fence_pos}, math at char {math_pos}). "
+        "If math runs first, `$x$` inside backticks gets extracted as math instead of code."
+    )
+
+
+def test_fence_stash_populated_before_math_stash():
+    """The fence_stash s.replace call must appear before any math_stash s.replace calls."""
+    # Find the s.replace call that populates each stash
+    fence_replace_pos = UI_JS.find("fence_stash.push(m)")
+    math_replace_pos = UI_JS.find("math_stash.push(")
+    assert fence_replace_pos != -1, "fence_stash population call not found"
+    assert math_replace_pos != -1, "math_stash population call not found"
+    assert fence_replace_pos < math_replace_pos, (
+        "fence_stash must be populated before math_stash to protect code span contents"
+    )
+
+
+def test_math_stash_comment_says_after_fence():
+    """The math stash comment should explain it runs AFTER fence_stash, not before."""
+    # Should not have the old misleading comment
+    assert "Must run BEFORE fence_stash" not in UI_JS, (
+        "Old misleading comment still present. Math stash runs AFTER fence_stash. "
+        "The comment should say 'Runs AFTER fence_stash'."
+    )
+
+
+# ── Pipeline regression: code spans protect their contents ────────────────────
+
+def test_math_restore_after_fence_restore():
+    """Math stash tokens are restored AFTER fence restore, so code spans get
+    their raw text back (not KaTeX placeholders)."""
+    fence_restore_pos = UI_JS.find("fence_stash[+i]")
+    math_restore_pos = UI_JS.find("math_stash[+i]")
+    assert fence_restore_pos != -1, "fence_stash restore not found"
+    assert math_restore_pos != -1, "math_stash restore not found"
+    # Both restores must exist; their relative order doesn't matter for correctness
+    # (they use different tokens: \x00F vs \x00M), but we assert both exist
+    assert fence_restore_pos != math_restore_pos, "fence and math restore must be separate calls"
+
+
+def test_stash_tokens_distinct():
+    """fence_stash and math_stash must use distinct sentinel tokens to avoid collisions."""
+    # fence uses \x00F, math uses \x00M (or similar unique prefix)
+    # The JS source uses escaped \\x00F and \\x00M as sentinel characters
+    # In the Python string read from the file these appear as '\\\\x00F' and '\\\\x00M'
+    assert "'\\\\x00F'" in UI_JS or 'x00F' in UI_JS, (
+        "fence stash token (\\x00F) not found — must be distinct from math token"
+    )
+    assert "'\\\\x00M'" in UI_JS or 'x00M' in UI_JS, (
+        "math stash token (\\x00M) not found — must be distinct from fence token"
+    )
+    # The two tokens must use different discriminator characters
+    assert 'x00F' in UI_JS and 'x00M' in UI_JS, (
+        "Both \\x00F (fence) and \\x00M (math) tokens must exist"
+    )
+
+
+# ── Workspace preview renderKatexBlocks wiring ────────────────────────────────
+
+def test_workspace_calls_render_katex_after_preview():
+    """workspace.js must call renderKatexBlocks() after setting previewMd.innerHTML.
+
+    Without this, math placeholders appear in workspace file previews but are never
+    rendered by KaTeX (renderKatexBlocks is only wired into renderMessages rAF).
+    """
+    assert "renderKatexBlocks" in WORKSPACE_JS, (
+        "workspace.js must call renderKatexBlocks() after renderMd() for file previews"
+    )
+
+
+def test_workspace_renders_katex_after_file_open():
+    """workspace.js renderKatexBlocks call must come after the renderMd(data.content) assignment."""
+    preview_md_pos = WORKSPACE_JS.find("renderMd(data.content)")
+    # Use the actual call string (not a stray regex match on 'M' characters)
+    katex_call_str = "renderKatexBlocks==='function'"
+    katex_call_pos = WORKSPACE_JS.find(katex_call_str)
+    assert preview_md_pos != -1, "renderMd(data.content) not found in workspace.js"
+    assert katex_call_pos != -1, (
+        "renderKatexBlocks guard (typeof renderKatexBlocks==='function') not found in workspace.js"
+    )
+    # The call after 'renderMd(data.content)' — find the LAST occurrence
+    # (there may be an earlier one in the save path at line ~153)
+    last_katex_pos = WORKSPACE_JS.rfind(katex_call_str)
+    assert last_katex_pos > preview_md_pos, (
+        "renderKatexBlocks must be called AFTER renderMd(data.content) in workspace.js "
+        f"(renderMd at {preview_md_pos}, last renderKatexBlocks at {last_katex_pos})"
+    )
+
+
+def test_workspace_katex_guarded_by_typeof():
+    """workspace.js renderKatexBlocks call must guard with typeof check for safety
+    in case KaTeX feature is not loaded (e.g. test environments, offline)."""
+    assert "typeof renderKatexBlocks" in WORKSPACE_JS, (
+        "workspace.js must guard renderKatexBlocks call with typeof check: "
+        "if(typeof renderKatexBlocks==='function')renderKatexBlocks()"
+    )
+
+
+# ── SAFE_TAGS: span addition should not expand attack surface ─────────────────
+
+def test_safe_tags_span_is_narrowly_scoped():
+    """SAFE_TAGS adding <span> is only a bypass if span carries dangerous attributes.
+    Verify the SAFE_TAGS regex tests the tag NAME only, not arbitrary attributes.
+    The rest of the pipeline uses esc() for user content, so attribute injection
+    into KaTeX spans isn't possible.
+    """
+    # The SAFE_TAGS regex must still require a word boundary / tag-end pattern
+    safe_tags_match = re.search(r"SAFE_TAGS\s*=\s*/(.+?)/i", UI_JS)
+    if not safe_tags_match:
+        safe_tags_match = re.search(r'SAFE_TAGS\s*=\s*/(.*?)/i', UI_JS)
+    assert safe_tags_match, "SAFE_TAGS regex not found"
+    pattern = safe_tags_match.group(1)
+    # Must have a trailing boundary check — ([\s>]|$) or similar
+    assert r"[\s>]" in pattern or r'[\s>]' in pattern, (
+        "SAFE_TAGS must enforce a boundary after the tag name to prevent "
+        "<spanxss> from matching when checking for <span>"
+    )
+
+
+# ── False-positive prevention ─────────────────────────────────────────────────
+
+def test_inline_math_regex_requires_non_space_boundaries():
+    """The $...$ inline regex must require non-space at both boundaries.
+
+    This prevents 'costs $5 and $10' from matching — the space after the opening
+    $ means it's a currency amount, not math.
+    """
+    # The inline math stash push is type:'inline' — find its containing replace() line
+    inline_push_idx = UI_JS.find("type:'inline',src:m")
+    assert inline_push_idx != -1, "Inline math stash push not found"
+    # Get the text from the start of that line back to find the regex
+    line_start = UI_JS.rfind('\n', 0, inline_push_idx) + 1
+    inline_line = UI_JS[line_start:inline_push_idx + 50]
+    # The regex must use \s (via [^\s...]) to exclude spaces at boundaries
+    assert '\\s' in inline_line or '[^' in inline_line, (
+        f"Inline math regex must exclude spaces at boundaries to prevent false "
+        f"positives on currency like $5. Found: {inline_line[:120]}"
+    )
+def test_display_math_stashed_before_inline():
+    """$$...$$ display math must be stashed before $...$ inline math.
+
+    If inline runs first on '$$x$$', it could match '$' + 'x' + '$' leaving
+    a stray outer '$', corrupting the output.
+    """
+    display_pos = UI_JS.find("type:'display',src:m")
+    inline_pos = UI_JS.find("type:'inline',src:m")
+    assert display_pos != -1, "display math stash not found"
+    assert inline_pos != -1, "inline math stash not found"
+    # First occurrence of display must be before first occurrence of inline
+    assert display_pos < inline_pos, (
+        "Display math ($$...$$) must be stashed before inline math ($...$) "
+        "to prevent $$ from being parsed as two adjacent inline delimiters"
+    )
+
+
+def test_math_stash_token_uses_single_backslash_null_byte():
+    """Math stash tokens must use the null-byte form (single backslash x00M).
+
+    The restore regex expects a null byte character. If the stash emits
+    a literal backslash+x00M (double backslash = 5-char string), the restore
+    regex never matches and the tokens appear verbatim in the rendered output.
+
+    The fence_stash correctly uses the null byte convention. Math stash must be consistent.
+    """
+    # In the source file, the correct form is: return '\x00M'
+    # The wrong form (double backslash) would be: return '\\x00M'
+    # Check that no double-backslash form exists in the math stash return statements
+    import re
+    bad_returns = re.findall(r"return\s+'\\\\x00M'", UI_JS)
+    assert not bad_returns, (
+        f"Found {len(bad_returns)} math stash return(s) using double-backslash \\\\x00M. "
+        "Must use single backslash '\x00M' (null byte) to match the restore regex."
+    )
+    # Positive check: single-backslash form must exist
+    good_returns = re.findall(r"math_stash\.push.*?return '\\x00M'", UI_JS, re.DOTALL)
+    assert good_returns, (
+        "Math stash return must use single-backslash '\x00M' (null byte convention)"
+    )

tests/test_issue357.py

@@ -0,0 +1,202 @@
+"""
+Tests for GitHub issue #357: Docker container fails to start without internet access.
+
+Structural tests — verify Dockerfile and docker_init.bash contain the expected
+patterns for pre-installed uv and workspace permission fixes.
+
+Two problems fixed:
+1. uv was downloaded at container startup; fails in air-gapped / firewalled environments.
+   Fix: pre-install uv in the Docker image at build time (system-wide in /usr/local/bin).
+2. workspace directory created with plain mkdir (as root); bind-mount dirs created by
+   Docker as root are unwritable by the hermeswebui user.
+   Fix: sudo mkdir + sudo chown for workspace directory.
+"""
+import pathlib
+import re
+
+REPO = pathlib.Path(__file__).parent.parent
+DOCKERFILE = (REPO / "Dockerfile").read_text(encoding="utf-8")
+INIT_SCRIPT = (REPO / "docker_init.bash").read_text(encoding="utf-8")
+
+
+# ── Dockerfile: uv pre-installed at build time ───────────────────────────────
+
+class TestDockerfileUvPreinstall:
+
+    def test_dockerfile_installs_uv_at_build_time(self):
+        """Dockerfile must install uv via RUN curl at build time (not only at runtime)."""
+        assert "RUN curl" in DOCKERFILE and "uv/install.sh" in DOCKERFILE, (
+            "Dockerfile must install uv at build time via RUN curl .../uv/install.sh"
+        )
+
+    def test_dockerfile_uv_installed_system_wide(self):
+        """uv must be installed to a system-wide directory (/usr/local/bin) accessible
+        to all users, not to a user-specific ~/.local/bin that another user can't see."""
+        # The install command must target /usr/local/bin or use root to install globally
+        uv_install_line = next(
+            (line for line in DOCKERFILE.splitlines() if "uv/install.sh" in line),
+            None,
+        )
+        assert uv_install_line is not None, "Could not find uv install line in Dockerfile"
+        # Must either use UV_INSTALL_DIR pointing to /usr/local/bin, or run as root
+        # (so the default install location is accessible to hermeswebui user)
+        has_system_dir = "/usr/local/bin" in uv_install_line or "UV_INSTALL_DIR=/usr/local/bin" in DOCKERFILE
+        assert has_system_dir, (
+            "uv must be installed to /usr/local/bin (system-wide) so hermeswebui user "
+            "can find it. Installing as hermeswebuitoo puts it in /home/hermeswebuitoo/.local/bin "
+            "which is NOT on hermeswebui's PATH."
+        )
+
+    def test_dockerfile_uv_installed_before_copy(self):
+        """uv installation must happen before COPY . /apptoo so it's in the image."""
+        import re
+        uv_pos = DOCKERFILE.find("uv/install.sh")
+        # Match COPY regardless of flags (e.g. --chown=...) — only the destination matters.
+        m = re.search(r"^COPY\b.*\s/apptoo\b", DOCKERFILE, re.MULTILINE)
+        assert uv_pos != -1, "uv install not found in Dockerfile"
+        assert m is not None, "COPY ... /apptoo not found in Dockerfile"
+        copy_pos = m.start()
+        assert uv_pos < copy_pos, "uv must be installed before COPY . /apptoo"
+
+    def test_dockerfile_uv_installed_as_root_or_before_user_switch(self):
+        """uv must be installed as root (USER root) to reach /usr/local/bin.
+        If installed as hermeswebuitoo, it lands in ~hermeswebuitoo/.local/bin,
+        which the hermeswebui user at runtime can't see.
+        """
+        lines = DOCKERFILE.splitlines()
+        uv_line_idx = next(i for i, l in enumerate(lines) if "uv/install.sh" in l)
+        # Find the last USER directive before the uv install line
+        user_before = None
+        for i in range(uv_line_idx - 1, -1, -1):
+            if lines[i].strip().startswith("USER "):
+                user_before = lines[i].strip().split()[1]
+                break
+        assert user_before == "root", (
+            f"uv install must run as USER root (found USER {user_before!r}). "
+            "Installing as hermeswebuitoo puts uv in /home/hermeswebuitoo/.local/bin "
+            "which is not accessible to the hermeswebui runtime user."
+        )
+
+
+# ── docker_init.bash: skip uv download when already present ─────────────────
+
+class TestInitScriptUvSkip:
+
+    def test_init_script_checks_uv_before_download(self):
+        """docker_init.bash must check 'command -v uv' before attempting download."""
+        assert "command -v uv" in INIT_SCRIPT, (
+            "docker_init.bash must check 'command -v uv' to skip download "
+            "when uv is already pre-installed in the image (#357)"
+        )
+
+    def test_init_script_skips_download_if_present(self):
+        """Init script must use conditional logic (if/else) around the uv download."""
+        # Pattern: if command -v uv ... else ... fi
+        assert re.search(r'if\s+command\s+-v\s+uv', INIT_SCRIPT), (
+            "docker_init.bash must use 'if command -v uv' guard around the download"
+        )
+
+    def test_init_script_curl_download_in_else_branch(self):
+        """The curl download must be in the else branch (only runs if uv not found)."""
+        # Find the conditional block
+        m = re.search(
+            r'if\s+command\s+-v\s+uv.*?fi',
+            INIT_SCRIPT, re.DOTALL
+        )
+        assert m, "Could not find uv conditional block in docker_init.bash"
+        block = m.group(0)
+        # curl must appear after 'else' not in the 'then' branch
+        else_pos = block.find("else")
+        curl_pos = block.find("curl")
+        assert else_pos != -1, "No 'else' branch in uv conditional"
+        assert curl_pos != -1, "No 'curl' in uv conditional block"
+        assert curl_pos > else_pos, (
+            "curl download must be in the 'else' branch, not the 'if/then' branch"
+        )
+
+    def test_init_script_error_exit_on_download_failure(self):
+        """Curl download must call error_exit on failure (not silently continue)."""
+        assert "error_exit" in INIT_SCRIPT and "Failed to install uv" in INIT_SCRIPT, (
+            "docker_init.bash must call error_exit if uv download fails, "
+            "so the container exits with a clear message instead of failing silently"
+        )
+
+    def test_init_script_path_includes_hermeswebui_local_bin(self):
+        """PATH must include /home/hermeswebui/.local/bin for fallback runtime install."""
+        assert "/home/hermeswebui/.local/bin" in INIT_SCRIPT, (
+            "docker_init.bash must include /home/hermeswebui/.local/bin in PATH "
+            "for the case where uv is installed at runtime via curl"
+        )
+
+
+# ── docker_init.bash: workspace directory permissions ────────────────────────
+
+class TestWorkspacePermissions:
+
+    def test_workspace_uses_sudo_mkdir(self):
+        """docker_init.bash must use 'sudo mkdir' for the workspace directory.
+
+        Docker auto-creates bind-mount directories as root if they don't exist,
+        leaving them unwritable by hermeswebui. sudo mkdir + chown fixes this.
+        """
+        # Find the workspace section
+        ws_section = INIT_SCRIPT[
+            INIT_SCRIPT.find("HERMES_WEBUI_DEFAULT_WORKSPACE"):
+            INIT_SCRIPT.find("HERMES_WEBUI_DEFAULT_WORKSPACE") + 800
+        ]
+        assert "sudo mkdir" in ws_section, (
+            "docker_init.bash must use 'sudo mkdir -p' for the workspace directory "
+            "to handle the case where Docker created the bind-mount dir as root (#357)"
+        )
+
+    def test_workspace_uses_sudo_chown(self):
+        """docker_init.bash must chown the workspace to hermeswebui when writable.
+
+        The chown is now conditional on the workspace being writable, to allow
+        read-only (:ro) workspace mounts without crashing (#670). The sudo chown
+        must still be present in the script (just guarded by [ -w ]).
+        """
+        assert 'sudo chown hermeswebui:hermeswebui "$HERMES_WEBUI_DEFAULT_WORKSPACE"' in INIT_SCRIPT, (
+            "docker_init.bash must 'sudo chown hermeswebui:hermeswebui' the workspace "
+            "when it is writable, so the app user can write to it (#357)"
+        )
+
+    def test_workspace_mkdir_before_chown(self):
+        """sudo mkdir must come before sudo chown in docker_init.bash."""
+        mkdir_pos = INIT_SCRIPT.find('sudo mkdir -p "$HERMES_WEBUI_DEFAULT_WORKSPACE"')
+        chown_pos = INIT_SCRIPT.find('sudo chown hermeswebui:hermeswebui "$HERMES_WEBUI_DEFAULT_WORKSPACE"')
+        assert mkdir_pos != -1, "sudo mkdir for workspace not found"
+        assert chown_pos != -1, "sudo chown for workspace not found"
+        assert mkdir_pos < chown_pos, "sudo mkdir must come before sudo chown"
+
+    def test_workspace_error_exit_on_mkdir_failure(self):
+        """sudo mkdir must call error_exit on failure."""
+        assert 'sudo mkdir -p "$HERMES_WEBUI_DEFAULT_WORKSPACE" || error_exit' in INIT_SCRIPT, (
+            "sudo mkdir for workspace must call error_exit on failure"
+        )
+
+    def test_workspace_chown_is_conditional_on_writable(self):
+        """chown and write-test must be skipped for read-only workspace mounts (#670).
+
+        The script must check [ -w "$HERMES_WEBUI_DEFAULT_WORKSPACE" ] before
+        attempting chown or a write test, so :ro bind-mounts don't crash startup.
+        """
+        assert '[ -w "$HERMES_WEBUI_DEFAULT_WORKSPACE" ]' in INIT_SCRIPT, (
+            "docker_init.bash must guard chown with [ -w ] to support read-only "
+            "workspace mounts (:ro) without crashing (#670)"
+        )
+        # Read-only path must log a clear message rather than calling error_exit
+        assert "read-only workspace is supported" in INIT_SCRIPT, (
+            "docker_init.bash must print a clear message when workspace is read-only (#670)"
+        )
+
+    def test_init_script_syntax_valid(self):
+        """docker_init.bash must pass bash -n syntax check."""
+        import subprocess
+        result = subprocess.run(
+            ["bash", "-n", str(REPO / "docker_init.bash")],
+            capture_output=True, text=True
+        )
+        assert result.returncode == 0, (
+            f"docker_init.bash failed bash -n syntax check:\n{result.stderr}"
+        )