fix: keep assistant-only stream deltas on current turn

api/streaming.py

@@ -1516,6 +1516,33 @@ def _merge_display_messages_after_agent_result(previous_display, previous_contex
     merged = previous_display[:]
     seen = {_message_identity(m) for m in merged}
     current_user_key = _message_identity({'role': 'user', 'content': msg_text})
+    current_user_in_candidates = any(
+        _message_identity(m) == current_user_key for m in candidates
+    )
+    current_user_already_checkpointed = bool(
+        merged and _message_identity(merged[-1]) == current_user_key
+    )
+    if (
+        current_user_key is not None
+        and not current_user_in_candidates
+        and not current_user_already_checkpointed
+        and any(
+            isinstance(m, dict) and m.get('role') in ('assistant', 'tool')
+            for m in candidates
+        )
+    ):
+        # Some provider retry/fallback paths can return an assistant/tool delta
+        # without echoing the current user turn. In deferred session-save mode
+        # the prompt exists only in pending_user_message, so appending that delta
+        # directly would make the assistant bubble appear attached to the prior
+        # exchange and then clear the pending prompt. Materialize the current
+        # turn at the transcript boundary before the assistant/tool response.
+        current_user_msg = {'role': 'user', 'content': msg_text}
+        insert_at = 0
+        while insert_at < len(candidates) and _is_context_compression_marker(candidates[insert_at]):
+            insert_at += 1
+        candidates = candidates[:insert_at] + [current_user_msg] + candidates[insert_at:]
+
     for msg in candidates:
         key = _message_identity(msg)
         if (

tests/test_session_save_mode.py

@@ -130,3 +130,30 @@ def test_eager_checkpointed_user_is_not_duplicated_after_agent_result():
         msg_text="repeat me",
     )
     assert [m["role"] for m in merged] == ["user", "assistant"]
+
+
+def test_deferred_turn_is_materialized_when_agent_returns_assistant_only_delta():
+    merged = streaming._merge_display_messages_after_agent_result(
+        previous_display=[
+            {"role": "user", "content": "older prompt"},
+            {"role": "assistant", "content": "older answer"},
+        ],
+        previous_context=[
+            {"role": "user", "content": "older prompt"},
+            {"role": "assistant", "content": "older answer"},
+        ],
+        result_messages=[
+            {"role": "user", "content": "older prompt"},
+            {"role": "assistant", "content": "older answer"},
+            {"role": "assistant", "content": "current answer"},
+        ],
+        msg_text="latest prompt",
+    )
+
+    assert [m["role"] for m in merged] == [
+        "user",
+        "assistant",
+        "user",
+        "assistant",
+    ]
+    assert [m["content"] for m in merged[-2:]] == ["latest prompt", "current answer"]