fix(gateway): restore MoA one-shot model override on failed turns

The MoA one-shot restore ran inside the try block after
_handle_message_with_agent returned. When that call raised an
exception (agent init failure, interpreter shutdown, OOM), the
restore was skipped and the MoA model override stayed permanently
on _session_model_overrides — silently routing all subsequent
messages through the MoA reference fan-out with no user-visible
indication.

Move the restore to the finally block so it fires on every exit
path (success, exception, interrupt). The restore data lives on
the per-turn event object and would be lost if not consumed here.

gateway/run.py

@@ -9020,16 +9020,6 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
 
         try:
             _agent_result = await self._handle_message_with_agent(event, source, _quick_key, _run_generation)
-            if getattr(event, "_moa_disable_after_turn", False):
-                try:
-                    _restore = getattr(event, "_moa_restore_override", None)
-                    if _restore is None:
-                        self._session_model_overrides.pop(_quick_key, None)
-                    else:
-                        self._session_model_overrides[_quick_key] = _restore
-                    self._evict_cached_agent(_quick_key)
-                except Exception:
-                    pass
             # Goal continuation: after the agent returns a final response
             # for this turn, check any standing /goal — the judge will
             # either mark it done, pause it (budget), or enqueue a
@@ -9060,13 +9050,16 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
                 logger.debug("goal continuation hook failed: %s", _goal_exc)
             return _agent_result
         finally:
-            # Unconditional release covers every exit path. _release_running_agent_state
-            # is idempotent (pop-on-absent is harmless) and, called without a
-            # run_generation guard, always clears the slot regardless of which
-            # generation it holds. This evicts the zombie left when session_reset
-            # bumps the generation (N -> N+1) mid-flight: gen-N's guarded release
-            # inside _run_agent returns False, and the old sentinel-only check here
-            # missed the leftover real agent — locking the session out forever (#28686).
+            if getattr(event, "_moa_disable_after_turn", False):
+                try:
+                    _restore = getattr(event, "_moa_restore_override", None)
+                    if _restore is None:
+                        self._session_model_overrides.pop(_quick_key, None)
+                    else:
+                        self._session_model_overrides[_quick_key] = _restore
+                    self._evict_cached_agent(_quick_key)
+                except Exception:
+                    pass
             self._release_running_agent_state(_quick_key)
 
     async def _prepare_inbound_message_text(

tests/gateway/test_moa_one_shot_restore.py

@@ -0,0 +1,130 @@
+"""MoA one-shot model override must be restored on both success and failure."""
+
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+
+def _make_runner():
+    """Build a minimal GatewayRunner-like object with the fields the
+    MoA one-shot restore path reads/writes."""
+    runner = MagicMock()
+    runner._session_model_overrides = {}
+    runner._release_running_agent_state = MagicMock()
+    runner._evict_cached_agent = MagicMock()
+    runner._begin_session_run_generation = MagicMock(return_value=1)
+    return runner
+
+
+def _make_event(text="hello", moa_disable=False, moa_restore=None):
+    event = SimpleNamespace()
+    event.text = text
+    if moa_disable:
+        event._moa_disable_after_turn = True
+        event._moa_restore_override = moa_restore
+    return event
+
+
+class TestMoaOneShotRestore:
+    """The gateway's MoA one-shot restore must fire in the finally block
+    so a failed turn still reverts the model override."""
+
+    def test_restore_fires_on_success(self):
+        """Normal successful turn restores the previous model override."""
+        runner = _make_runner()
+        key = "agent:main:telegram:dm:123"
+        runner._session_model_overrides[key] = {
+            "provider": "moa", "model": "default",
+        }
+        event = _make_event(
+            moa_disable=True,
+            moa_restore={"provider": "openrouter", "model": "gpt-4"},
+        )
+
+        # Simulate: try block succeeds, finally runs
+        try:
+            pass  # _handle_message_with_agent succeeds
+        finally:
+            if getattr(event, "_moa_disable_after_turn", False):
+                _restore = getattr(event, "_moa_restore_override", None)
+                if _restore is None:
+                    runner._session_model_overrides.pop(key, None)
+                else:
+                    runner._session_model_overrides[key] = _restore
+
+        assert runner._session_model_overrides[key] == {
+            "provider": "openrouter", "model": "gpt-4",
+        }
+
+    def test_restore_fires_on_exception(self):
+        """A failed turn (exception) must still restore the previous model."""
+        runner = _make_runner()
+        key = "agent:main:telegram:dm:123"
+        runner._session_model_overrides[key] = {
+            "provider": "moa", "model": "default",
+        }
+        event = _make_event(
+            moa_disable=True,
+            moa_restore={"provider": "openrouter", "model": "gpt-4"},
+        )
+
+        # Simulate: try block raises, finally still runs
+        try:
+            raise RuntimeError("provider error")
+        except RuntimeError:
+            pass
+        finally:
+            if getattr(event, "_moa_disable_after_turn", False):
+                _restore = getattr(event, "_moa_restore_override", None)
+                if _restore is None:
+                    runner._session_model_overrides.pop(key, None)
+                else:
+                    runner._session_model_overrides[key] = _restore
+
+        assert runner._session_model_overrides[key] == {
+            "provider": "openrouter", "model": "gpt-4",
+        }
+
+    def test_restore_none_clears_override(self):
+        """When the user had no model override before /moa, the override
+        should be removed (not left as MoA)."""
+        runner = _make_runner()
+        key = "agent:main:discord:guild:456"
+        runner._session_model_overrides[key] = {
+            "provider": "moa", "model": "default",
+        }
+        event = _make_event(moa_disable=True, moa_restore=None)
+
+        try:
+            raise RuntimeError("timeout")
+        except RuntimeError:
+            pass
+        finally:
+            if getattr(event, "_moa_disable_after_turn", False):
+                _restore = getattr(event, "_moa_restore_override", None)
+                if _restore is None:
+                    runner._session_model_overrides.pop(key, None)
+                else:
+                    runner._session_model_overrides[key] = _restore
+
+        assert key not in runner._session_model_overrides
+
+    def test_no_restore_when_not_one_shot(self):
+        """Normal (non-MoA) turns must not touch model overrides."""
+        runner = _make_runner()
+        key = "agent:main:slack:channel:789"
+        runner._session_model_overrides[key] = {
+            "provider": "openrouter", "model": "gpt-4",
+        }
+        event = _make_event()  # no _moa_disable_after_turn
+
+        try:
+            pass
+        finally:
+            if getattr(event, "_moa_disable_after_turn", False):
+                runner._session_model_overrides.pop(key, None)
+
+        assert runner._session_model_overrides[key] == {
+            "provider": "openrouter", "model": "gpt-4",
+        }