fix(cli): TypeError concatenating queued note onto multimodal (image) message

cli.py

@@ -2116,6 +2116,41 @@ def _schedule():
             pass
 
 
+def _prepend_note_to_message(message, note: str):
+    """Prepend a one-shot system-style note to a user message.
+
+    ``message`` is normally a plain string, but when the user attaches an image
+    to a vision-capable model it becomes a list of OpenAI-style content parts
+    (text + ``image_url`` blocks). Naively doing ``note + "\\n\\n" + message``
+    then raises ``TypeError: can only concatenate str (not "list") to str`` —
+    e.g. running ``/model ...`` (which queues a model-switch note) and then
+    sending a pasted image in the same turn.
+
+    Returns the message with ``note`` prepended:
+      * ``str``  → ``f"{note}\\n\\n{message}"`` (just ``note`` when empty)
+      * ``list`` → note folded into the first text part, or inserted as a new
+        leading ``{"type": "text"}`` part when there is no text part.
+    Unknown shapes are returned unchanged (fail-open).
+    """
+    note = str(note or "").strip()
+    if not note:
+        return message
+    if isinstance(message, str):
+        return f"{note}\n\n{message}" if message else note
+    if isinstance(message, list):
+        parts = list(message)
+        for i, part in enumerate(parts):
+            if isinstance(part, dict) and part.get("type") == "text":
+                merged = dict(part)
+                text = merged.get("text", "")
+                merged["text"] = f"{note}\n\n{text}" if text else note
+                parts[i] = merged
+                return parts
+        # No text part (image-only) — insert the note as a leading text block.
+        return [{"type": "text", "text": note}, *parts]
+    return message
+
+
 # ---------------------------------------------------------------------------
 # File-drop / local attachment detection — extracted as pure helpers for tests.
 # ---------------------------------------------------------------------------
@@ -12135,17 +12170,21 @@ def run_agent():
                     reset_current_session_key = None  # type: ignore[assignment]
                     _approval_session_token = None
                 agent_message = _voice_prefix + message if _voice_prefix else message
-                # Prepend pending model switch note so the model knows about the switch
+                # Prepend pending notes via _prepend_note_to_message, which
+                # handles both plain-string and multimodal content-parts list
+                # messages. Naive ``note + "\n\n" + agent_message`` crashed with
+                # TypeError when an image was attached (agent_message is a list)
+                # and a /model or /reload-skills note was queued for the turn.
                 _msn = getattr(self, '_pending_model_switch_note', None)
                 if _msn:
-                    agent_message = _msn + "\n\n" + agent_message
+                    agent_message = _prepend_note_to_message(agent_message, _msn)
                     self._pending_model_switch_note = None
                 # Prepend pending /reload-skills note so the model sees which
                 # skills were added/removed before handling this turn. Same
                 # one-shot queue pattern as the model-switch note above.
                 _srn = getattr(self, '_pending_skills_reload_note', None)
                 if _srn:
-                    agent_message = _srn + "\n\n" + agent_message
+                    agent_message = _prepend_note_to_message(agent_message, _srn)
                     self._pending_skills_reload_note = None
                 try:
                     result = self.agent.run_conversation(

tests/cli/test_prepend_note_to_message.py

@@ -0,0 +1,80 @@
+"""Tests for cli._prepend_note_to_message.
+
+Regression coverage for the TypeError raised when a queued /model or
+/reload-skills note was prepended to a multimodal (image-attached) message:
+``can only concatenate str (not "list") to str``.
+"""
+
+from cli import _prepend_note_to_message
+
+
+def test_string_message_gets_note_prepended():
+    assert _prepend_note_to_message("hello", "NOTE") == "NOTE\n\nhello"
+
+
+def test_empty_note_returns_message_unchanged():
+    assert _prepend_note_to_message("hello", "") == "hello"
+    assert _prepend_note_to_message("hello", "   ") == "hello"
+    parts = [{"type": "text", "text": "hi"}]
+    assert _prepend_note_to_message(parts, "") == parts
+
+
+def test_note_is_stripped():
+    assert _prepend_note_to_message("hello", "  NOTE  ") == "NOTE\n\nhello"
+
+
+def test_empty_string_message_yields_just_note():
+    # No trailing blank lines when the user message is empty.
+    assert _prepend_note_to_message("", "NOTE") == "NOTE"
+
+
+def test_empty_text_part_yields_just_note():
+    message = [
+        {"type": "text", "text": ""},
+        {"type": "image_url", "image_url": {"url": "x"}},
+    ]
+    result = _prepend_note_to_message(message, "NOTE")
+    assert result[0]["text"] == "NOTE"
+    assert result[1]["type"] == "image_url"
+
+
+def test_list_message_folds_note_into_first_text_part():
+    message = [
+        {"type": "text", "text": "describe this"},
+        {"type": "image_url", "image_url": {"url": "data:..."}},
+    ]
+    result = _prepend_note_to_message(message, "NOTE")
+
+    assert result[0]["type"] == "text"
+    assert result[0]["text"] == "NOTE\n\ndescribe this"
+    # Image part is preserved untouched.
+    assert result[1] == {"type": "image_url", "image_url": {"url": "data:..."}}
+    # Original message is not mutated.
+    assert message[0]["text"] == "describe this"
+
+
+def test_image_only_list_gets_leading_text_part():
+    message = [{"type": "image_url", "image_url": {"url": "data:..."}}]
+    result = _prepend_note_to_message(message, "NOTE")
+
+    assert result[0] == {"type": "text", "text": "NOTE"}
+    assert result[1]["type"] == "image_url"
+
+
+def test_list_message_does_not_raise_typeerror():
+    # The exact #repro shape: multimodal list + queued note must not raise
+    # "can only concatenate str (not 'list') to str".
+    message = [
+        {"type": "text", "text": "look"},
+        {"type": "image_url", "image_url": {"url": "x"}},
+    ]
+    result = _prepend_note_to_message(
+        message, "Model switched to gpt-5.5 (provider: openai-codex)."
+    )
+    assert isinstance(result, list)
+    assert result[0]["text"].startswith("Model switched to gpt-5.5")
+
+
+def test_unknown_shape_returned_unchanged():
+    assert _prepend_note_to_message(123, "NOTE") == 123
+    assert _prepend_note_to_message(None, "NOTE") is None