fix(model): repair bare custom-provider suffix on resolver slow path (#5225)

Mirror the #5128 fast-path suffix repair when profile context is present
but requested_provider is omitted, which async delegation completion can
trigger via process_wakeup.

Co-authored-by: b3nw <b3nw@users.noreply.github.com>

api/routes.py

@@ -5578,6 +5578,24 @@ def _resolve_compatible_session_model_state(
             _target = _profile_default or default_model
             return _target, profile_provider, True
 
+        # Async server-side continuations (for example delegate_task completion
+        # re-entry) can arrive here with profile context but without a usable
+        # requested_provider, bypassing the fast-path custom-provider repair
+        # above. If the profile's configured custom-provider default is a
+        # slash-qualified model whose suffix matches the bare session model,
+        # repair back to the profile default before the provider call (#5225).
+        if (
+            "/" not in model
+            and _profile_default
+            and "/" in _profile_default
+            and _profile_default.rsplit("/", 1)[-1] == model
+            and (
+                _profile_provider_normalized == "custom"
+                or str(profile_provider).startswith("custom:")
+            )
+        ):
+            return _profile_default, profile_provider, True
+
         return model, profile_provider, False
 
     if not model:

tests/test_issue5127_process_wakeup_bare_model.py

@@ -82,6 +82,47 @@ class TestIssue5127CustomProviderBareSuffixRepair:
         assert mock_catalog.call_count == 0
         assert result == ("x-ai/grok-composer-2.5-fast", "custom:my-proxy", False)
 
+    def test_slow_path_repairs_bare_suffix_to_profile_default_for_custom_provider(self):
+        """Regression #5225: async continuations may arrive without usable model_provider."""
+        from api.routes import _resolve_compatible_session_model_state
+
+        with patch("api.routes.get_available_models") as mock_catalog:
+            mock_catalog.return_value = {
+                "active_provider": "openrouter",
+                "default_model": "openai/gpt-5.5",
+                "groups": [],
+            }
+            result = _resolve_compatible_session_model_state(
+                "grok-composer-2.5-fast",
+                None,
+                profile_provider="custom:my-proxy",
+                profile_default_model="x-ai/grok-composer-2.5-fast",
+                prefer_cached_catalog=True,
+            )
+
+        assert mock_catalog.call_count == 1
+        assert result == ("x-ai/grok-composer-2.5-fast", "custom:my-proxy", True)
+
+    def test_slow_path_does_not_rewrite_unrelated_bare_model_for_custom_provider(self):
+        from api.routes import _resolve_compatible_session_model_state
+
+        with patch("api.routes.get_available_models") as mock_catalog:
+            mock_catalog.return_value = {
+                "active_provider": "openrouter",
+                "default_model": "openai/gpt-5.5",
+                "groups": [],
+            }
+            result = _resolve_compatible_session_model_state(
+                "other-model",
+                None,
+                profile_provider="custom:my-proxy",
+                profile_default_model="x-ai/grok-composer-2.5-fast",
+                prefer_cached_catalog=True,
+            )
+
+        assert mock_catalog.call_count == 1
+        assert result == ("other-model", "custom:my-proxy", False)
+
     def test_openai_codex_stale_openai_slash_still_uses_slow_path(self):
         """Regression: openai/... under openai-codex must not take bare fast return."""
         from api.routes import _resolve_compatible_session_model_state