fix(title): publish conversation context for aux title calls

Title generation via the aux client ran without the Agent's ambient
conversation context, so relay targets that key on it (OpenCode Go)
rejected the completion call with HTTP 400.

Publish the webui session id as the Agent conversation context around
the aux completion (restoring it in a finally block) and pass it from
all three title entry paths: background update, background refresh and
the on-demand regenerate helper.

Refs #7470

api/streaming.py

@@ -4560,7 +4560,7 @@ def _generate_llm_session_title_for_agent(agent, user_text: str, assistant_text:
     return None, 'llm_invalid', str(raw)[:120]
 
 
-def _generate_llm_session_title_via_aux(user_text: str, assistant_text: str, agent=None, *, use_agent_model: bool = False) -> tuple[Optional[str], str, str]:
+def _generate_llm_session_title_via_aux(user_text: str, assistant_text: str, agent=None, *, use_agent_model: bool = False, conversation_id: str = '') -> tuple[Optional[str], str, str]:
     """Generate a title via dedicated auxiliary LLM route, then sanitize/validate result.
 
     When use_agent_model is False (default), the auxiliary client resolves
@@ -4568,6 +4568,11 @@ def _generate_llm_session_title_via_aux(user_text: str, assistant_text: str, age
     prevents the session's chat model (e.g. a Chinese model) from overriding
     the dedicated title model.  When True, the agent's attrs are passed through
     (legacy fallback behaviour).
+
+    conversation_id republishes the webui session id as the Agent's ambient
+    conversation context for the duration of the aux call, so OpenCode relay
+    targets receive the same ``x-opencode-session`` sticky key as the session's
+    main turns (#7470). The context is reset in all exit paths.
     """
     if use_agent_model and agent:
         provider = getattr(agent, 'provider', '')
@@ -4577,13 +4582,30 @@ def _generate_llm_session_title_via_aux(user_text: str, assistant_text: str, age
         provider = ''
         model = ''
         base_url = ''
-    raw, status = generate_title_raw_via_aux(
-        user_text,
-        assistant_text,
-        provider=provider,
-        model=model,
-        base_url=base_url,
-    )
+    ctx_token = None
+    if conversation_id:
+        try:
+            from agent.portal_tags import set_conversation_context
+            ctx_token = set_conversation_context(str(conversation_id))
+        except Exception:
+            # Older/absent agent runtime: proceed without conversation context
+            # (previous behaviour) rather than failing title generation.
+            ctx_token = None
+    try:
+        raw, status = generate_title_raw_via_aux(
+            user_text,
+            assistant_text,
+            provider=provider,
+            model=model,
+            base_url=base_url,
+        )
+    finally:
+        if ctx_token is not None:
+            try:
+                from agent.portal_tags import reset_conversation_context
+                reset_conversation_context(ctx_token)
+            except Exception:
+                pass
     if not raw:
         return None, status, ''
     title = _sanitize_generated_title(raw)
@@ -4733,9 +4755,9 @@ def _run_background_title_update(session_id: str, user_text: str, assistant_text
             if agent and not aux_title_configured:
                 next_title, llm_status, raw_preview = _generate_llm_session_title_for_agent(agent, user_text, assistant_text)
                 if not next_title and llm_status in ('llm_error', 'llm_invalid'):
-                    next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent, use_agent_model=True)
+                    next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent, use_agent_model=True, conversation_id=session_id)
             else:
-                next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text)
+                next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, conversation_id=session_id)
                 if not next_title and agent and llm_status in ('llm_error_aux', 'llm_invalid_aux'):
                     next_title, llm_status, raw_preview = _generate_llm_session_title_for_agent(agent, user_text, assistant_text)
             source = llm_status
@@ -4831,9 +4853,9 @@ def _run_background_title_refresh(session_id: str, user_text: str, assistant_tex
             if agent and not aux_title_configured:
                 next_title, llm_status, raw_preview = _generate_llm_session_title_for_agent(agent, user_text, assistant_text)
                 if not next_title and llm_status in ('llm_error', 'llm_invalid'):
-                    next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent, use_agent_model=True)
+                    next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent, use_agent_model=True, conversation_id=session_id)
             else:
-                next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text)
+                next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, conversation_id=session_id)
                 if not next_title and agent and llm_status in ('llm_error_aux', 'llm_invalid_aux'):
                     next_title, llm_status, raw_preview = _generate_llm_session_title_for_agent(agent, user_text, assistant_text)
         if not next_title:
@@ -4896,7 +4918,7 @@ def generate_session_title_for_session(session, *, prefer_latest: bool = False,
     with profiles_api.profile_env_for_background_worker(session, "manual title regeneration", logger_override=logger):
         if not _aux_title_generation_enabled():
             return None, 'title_generation_disabled', ''
-        next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent)
+        next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent, conversation_id=getattr(session, 'session_id', '') or '')
     if next_title:
         return next_title, llm_status, raw_preview
     fallback_title = _fallback_title_from_exchange(user_text, assistant_text)

tests/_aux_client_helpers.py

@@ -33,13 +33,45 @@ def auxiliary_client_modules():
     The stubs are fresh objects owned by this helper, so the ``auxiliary_client``
     attribute is set on our own module and no pre-existing ``agent`` module is
     mutated. Restoration is therefore entirely ``patch.dict``'s job.
+
+    ``agent.portal_tags`` is stubbed with a real ``ContextVar`` behind the same
+    set/reset/get surface as the Agent runtime (tokens are truthy objects, and
+    reset restores the previous value), so title-generation tests can assert
+    conversation-context publication/restoration without a real Agent runtime
+    (#7470).
     """
+    import contextvars as _contextvars
+
     agent_stub = types.ModuleType("agent")
     auxiliary_client_stub = types.ModuleType("agent.auxiliary_client")
+    portal_tags_stub = types.ModuleType("agent.portal_tags")
     agent_stub.auxiliary_client = auxiliary_client_stub
+    portal_state = {"set_calls": [], "reset_calls": []}
+    _conversation_cv = _contextvars.ContextVar("webui_test_conversation_id", default=None)
+
+    def _set_conversation_context(conversation_id):
+        portal_state["set_calls"].append(conversation_id)
+        return _conversation_cv.set(conversation_id or None)
+
+    def _reset_conversation_context(token):
+        portal_state["reset_calls"].append(token)
+        _conversation_cv.reset(token)
+
+    def _get_conversation_context():
+        return _conversation_cv.get()
+
+    portal_tags_stub.set_conversation_context = _set_conversation_context
+    portal_tags_stub.reset_conversation_context = _reset_conversation_context
+    portal_tags_stub.get_conversation_context = _get_conversation_context
+    portal_tags_stub._portal_state = portal_state
+    agent_stub.portal_tags = portal_tags_stub
     return mock.patch.dict(
         sys.modules,
-        {"agent": agent_stub, "agent.auxiliary_client": auxiliary_client_stub},
+        {
+            "agent": agent_stub,
+            "agent.auxiliary_client": auxiliary_client_stub,
+            "agent.portal_tags": portal_tags_stub,
+        },
     )
 
 

tests/test_title_aux_routing.py

@@ -1082,5 +1082,196 @@ class TestAuxInvalidAuxTriggersAgentFallback(unittest.TestCase):
         mock_agent_title.assert_not_called()
 
 
+class TestAuxTitleConversationContext(unittest.TestCase):
+    """#7470: aux title-generation calls republish the session id as the Agent's
+    ambient conversation context so OpenCode relay targets get the same
+    x-opencode-session sticky key as the session's main turns."""
+
+    MOCK_RESP = types.SimpleNamespace(
+        choices=[
+            types.SimpleNamespace(
+                message=types.SimpleNamespace(content='Weather Title'),
+                finish_reason='stop',
+            )
+        ]
+    )
+
+    def _portal_state(self):
+        from agent.portal_tags import _portal_state
+        return _portal_state
+
+    def _current_context(self):
+        from agent.portal_tags import get_conversation_context
+        return get_conversation_context()
+
+    def test_direct_call_publishes_and_restores_context_after_success(self):
+        from api.streaming import _generate_llm_session_title_via_aux
+
+        seen = {}
+
+        def fake_call_llm(**kwargs):
+            from agent.portal_tags import get_conversation_context
+            seen['context_during_call'] = get_conversation_context()
+            return self.MOCK_RESP
+
+        with _patch_tg_config({'provider': '', 'model': 'gpt-4o', 'base_url': ''}):
+            with patch('agent.auxiliary_client.call_llm', side_effect=fake_call_llm, create=True):
+                title, status, _raw = _generate_llm_session_title_via_aux(
+                    'What is the weather?', 'It is sunny.', conversation_id='sid-123',
+                )
+
+        self.assertEqual(title, 'Weather Title')
+        self.assertEqual(status, 'llm_aux')
+        # Context published before the aux call and visible inside call_llm.
+        self.assertEqual(seen['context_during_call'], 'sid-123')
+        state = self._portal_state()
+        self.assertEqual(state['set_calls'], ['sid-123'])
+        # Reset restored the pre-call value (None) in the finally block.
+        self.assertIsNone(self._current_context())
+        self.assertEqual(len(state['reset_calls']), 1)
+
+    def test_context_restored_after_exception_escapes_aux_call(self):
+        from api.streaming import _generate_llm_session_title_via_aux
+
+        with _patch_tg_config({'provider': '', 'model': 'gpt-4o', 'base_url': ''}):
+            with patch('api.streaming.generate_title_raw_via_aux', side_effect=RuntimeError('boom')):
+                with self.assertRaises(RuntimeError):
+                    _generate_llm_session_title_via_aux(
+                        'What is the weather?', 'It is sunny.', conversation_id='sid-456',
+                    )
+        state = self._portal_state()
+        self.assertEqual(state['set_calls'], ['sid-456'])
+        self.assertIsNone(self._current_context())
+        self.assertEqual(len(state['reset_calls']), 1)
+
+    def test_no_conversation_id_skips_context_publication(self):
+        from api.streaming import _generate_llm_session_title_via_aux
+
+        def fake_call_llm(**kwargs):
+            return self.MOCK_RESP
+
+        with _patch_tg_config({'provider': '', 'model': 'gpt-4o', 'base_url': ''}):
+            with patch('agent.auxiliary_client.call_llm', side_effect=fake_call_llm, create=True):
+                title, status, _raw = _generate_llm_session_title_via_aux(
+                    'What is the weather?', 'It is sunny.',
+                )
+
+        self.assertEqual(title, 'Weather Title')
+        state = self._portal_state()
+        self.assertEqual(state['set_calls'], [])
+        self.assertEqual(state['reset_calls'], [])
+
+    def test_opencode_target_agent_model_publishes_non_empty_session_context(self):
+        """use_agent_model=True through an OpenCode-style target still carries the key."""
+        from api.streaming import _generate_llm_session_title_via_aux
+
+        seen = {}
+        agent = types.SimpleNamespace(
+            provider='opencode', model='opencode-go', base_url='http://localhost:3456',
+        )
+
+        def fake_call_llm(**kwargs):
+            from agent.portal_tags import get_conversation_context
+            seen['context_during_call'] = get_conversation_context()
+            seen['provider'] = kwargs.get('provider')
+            seen['base_url'] = kwargs.get('base_url')
+            return self.MOCK_RESP
+
+        with patch('agent.auxiliary_client.call_llm', side_effect=fake_call_llm, create=True):
+            title, status, _raw = _generate_llm_session_title_via_aux(
+                'What is the weather?', 'It is sunny.',
+                agent=agent, use_agent_model=True, conversation_id='oc-sid',
+            )
+
+        self.assertEqual(title, 'Weather Title')
+        self.assertEqual(seen['context_during_call'], 'oc-sid')
+        self.assertEqual(seen['provider'], 'opencode')
+        self.assertEqual(seen['base_url'], 'http://localhost:3456')
+        self.assertIsNone(self._current_context())
+
+    @patch('api.streaming._aux_title_configured', return_value=True)
+    @patch('api.streaming.get_session')
+    @patch('api.streaming._generate_llm_session_title_for_agent')
+    @patch('api.streaming._generate_llm_session_title_via_aux')
+    def test_background_update_entry_passes_session_id(
+        self, mock_aux_title, mock_agent_title, mock_get_session, mock_configured,
+    ):
+        """_run_background_title_update forwards the session id as conversation id."""
+        from api.streaming import _run_background_title_update
+
+        mock_session = MagicMock()
+        mock_session.title = 'Untitled'
+        mock_session.llm_title_generated = False
+        mock_session.messages = [
+            {'role': 'user', 'content': 'Hello'},
+            {'role': 'assistant', 'content': 'Hi there'},
+        ]
+        mock_get_session.return_value = mock_session
+        mock_aux_title.return_value = ('Greeting', 'llm_aux', '')
+
+        _run_background_title_update(
+            session_id='entry-update-sid',
+            user_text='Hello',
+            assistant_text='Hi there',
+            placeholder_title='Untitled',
+            put_event=lambda _t, _d: None,
+            agent=MagicMock(),
+        )
+
+        self.assertEqual(mock_aux_title.call_args.kwargs.get('conversation_id'), 'entry-update-sid')
+        mock_agent_title.assert_not_called()
+
+    @patch('api.streaming._aux_title_configured', return_value=True)
+    @patch('api.streaming.get_session')
+    @patch('api.streaming._generate_llm_session_title_for_agent')
+    @patch('api.streaming._generate_llm_session_title_via_aux')
+    def test_background_refresh_entry_passes_session_id(
+        self, mock_aux_title, mock_agent_title, mock_get_session, mock_configured,
+    ):
+        """_run_background_title_refresh forwards the session id as conversation id."""
+        from api.streaming import _run_background_title_refresh
+
+        mock_session = MagicMock()
+        mock_session.title = 'Old Title'
+        mock_session.messages = [
+            {'role': 'user', 'content': 'Hello again'},
+            {'role': 'assistant', 'content': 'More context'},
+        ]
+        mock_get_session.return_value = mock_session
+        mock_aux_title.return_value = ('Newer Title', 'llm_aux', '')
+
+        _run_background_title_refresh(
+            session_id='entry-refresh-sid',
+            user_text='Hello again',
+            assistant_text='More context',
+            current_title='Old Title',
+            put_event=lambda _t, _d: None,
+            agent=MagicMock(),
+        )
+
+        self.assertEqual(mock_aux_title.call_args.kwargs.get('conversation_id'), 'entry-refresh-sid')
+        mock_agent_title.assert_not_called()
+
+    @patch('api.streaming._generate_llm_session_title_via_aux')
+    def test_sync_regenerate_entry_passes_session_id(self, mock_aux_title):
+        """generate_session_title_for_session forwards the session id (sync path)."""
+        from api.streaming import generate_session_title_for_session
+
+        mock_session = MagicMock()
+        mock_session.session_id = 'entry-sync-sid'
+        mock_session.title = 'Untitled'
+        mock_session.messages = [
+            {'role': 'user', 'content': 'Hello sync'},
+            {'role': 'assistant', 'content': 'Sync reply'},
+        ]
+        mock_aux_title.return_value = ('Sync Title', 'llm_aux', '')
+
+        title, status, _raw = generate_session_title_for_session(mock_session)
+
+        self.assertEqual(title, 'Sync Title')
+        self.assertEqual(status, 'llm_aux')
+        self.assertEqual(mock_aux_title.call_args.kwargs.get('conversation_id'), 'entry-sync-sid')
+
+
 if __name__ == '__main__':
     unittest.main()