fix: restore RepairOrphaned middleware export and repoint model fake to deferred_model boundary

agent/analyzer.py

@@ -46,8 +46,8 @@ from .server import (
 from .tools.read_finding_outcomes import read_finding_outcomes
 from .tools.save_review_style import save_review_style_prompt
 from .utils.analyzer_skills import SKILLS_ROUTE, skill_path_for_mode
-from .utils.github_app import get_github_app_installation_token
 from .utils.deferred_model import make_model_or_defer
+from .utils.github_app import get_github_app_installation_token
 from .utils.model import DEFAULT_LLM_REASONING, provider_model_kwargs
 from .utils.sandbox_paths import aresolve_sandbox_work_dir
 from .utils.sandbox_state import unwrap_sandbox_backend

agent/middleware/__init__.py

@@ -11,6 +11,7 @@ _MIDDLEWARE_MODULES = {
     "PlanModeMiddleware": ".plan_mode",
     "refresh_github_proxy_before_model": ".refresh_github_proxy",
     "SlackAssistantStatusMiddleware": ".refresh_slack_status",
+    "RepairOrphanedToolCallsMiddleware": ".repair_orphaned_tool_calls",
     "SandboxCircuitBreakerMiddleware": ".sandbox_circuit_breaker",
     "SanitizeThinkingBlocksMiddleware": ".sanitize_thinking_blocks",
     "SanitizeToolInputsMiddleware": ".sanitize_tool_inputs",
@@ -37,6 +38,7 @@ __all__ = [
     "WorkflowPushGuardMiddleware",
     "SandboxCircuitBreakerMiddleware",
     "SlackAssistantStatusMiddleware",
+    "RepairOrphanedToolCallsMiddleware",
     "check_message_queue_before_model",
     "ensure_no_empty_msg",
     "notify_step_limit_reached",
@@ -55,6 +57,7 @@ if TYPE_CHECKING:
     from .plan_mode import PlanModeMiddleware
     from .refresh_github_proxy import refresh_github_proxy_before_model
     from .refresh_slack_status import SlackAssistantStatusMiddleware
+    from .repair_orphaned_tool_calls import RepairOrphanedToolCallsMiddleware
     from .sandbox_circuit_breaker import SandboxCircuitBreakerMiddleware
     from .sanitize_thinking_blocks import SanitizeThinkingBlocksMiddleware
     from .sanitize_tool_inputs import SanitizeToolInputsMiddleware

agent/reviewer.py

@@ -83,9 +83,9 @@ from .tools import (
 )
 from .utils.agents_md import fetch_agents_md
 from .utils.api_standards_skill import fetch_api_standards_skill
+from .utils.deferred_model import make_model_or_defer
 from .utils.github_app import get_github_app_installation_token_with_expiry
 from .utils.github_token import cache_github_token_for_thread
-from .utils.deferred_model import make_model_or_defer
 from .utils.model import DEFAULT_LLM_REASONING, provider_model_kwargs
 from .utils.repo_prep import materialize_trusted_skills, prepare_review_repo
 from .utils.sandbox_paths import aresolve_sandbox_work_dir

tests/e2e/patches.py

@@ -29,7 +29,6 @@ def apply() -> None:
 
     import importlib
 
-    from agent import server
     from agent.utils import auth, authorship
     from agent.utils import slack as slack_utils
 
@@ -52,7 +51,11 @@ def apply() -> None:
         def _fake_make_model(model_id: str, **kwargs: object):  # noqa: ARG001
             return FakeScriptedChatModel(script=build_script())
 
-        server.make_model = _fake_make_model
+        # ``make_model_or_defer`` (used by server/reviewer/analyzer) resolves
+        # ``make_model`` via the module global at call time, so patch it there.
+        import agent.utils.deferred_model as deferred_model
+
+        deferred_model.make_model = _fake_make_model
 
     async def _dummy_install_token_with_expiry() -> tuple[str, str | None]:
         return "dummy-installation-token", None

tests/test_agent_assembly_context.py

@@ -65,7 +65,7 @@ async def _capture_create_deep_agent_kwargs() -> dict[str, object]:
         ),
         patch("agent.server.load_profile", new_callable=AsyncMock, return_value=None),
         patch("agent.server.fallback_model_id_for", return_value=None),
-        patch("agent.server.make_model", side_effect=[MagicMock(), MagicMock()]),
+        patch("agent.utils.deferred_model.make_model", side_effect=[MagicMock(), MagicMock()]),
         patch("agent.server.construct_system_prompt", return_value="prompt"),
         patch("agent.server.create_deep_agent", side_effect=fake_create_deep_agent),
     ):

tests/test_agent_subagent_models.py

@@ -66,7 +66,9 @@ async def test_agent_uses_profile_subagent_model_override() -> None:
             },
         ),
         patch("agent.server.fallback_model_id_for", return_value=None),
-        patch("agent.server.make_model", side_effect=[main_model, subagent_model]) as make_model,
+        patch(
+            "agent.utils.deferred_model.make_model", side_effect=[main_model, subagent_model]
+        ) as make_model,
         patch("agent.server.construct_system_prompt", return_value="prompt"),
         patch("agent.server.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
@@ -142,7 +144,9 @@ async def test_agent_subagent_inherits_profile_model_override_without_explicit_p
             },
         ),
         patch("agent.server.fallback_model_id_for", return_value=None),
-        patch("agent.server.make_model", side_effect=[main_model, subagent_model]) as make_model,
+        patch(
+            "agent.utils.deferred_model.make_model", side_effect=[main_model, subagent_model]
+        ) as make_model,
         patch("agent.server.construct_system_prompt", return_value="prompt"),
         patch("agent.server.create_deep_agent", side_effect=fake_create_deep_agent),
     ):

tests/test_corridor_mcp.py

@@ -154,7 +154,7 @@ async def test_get_agent_passes_corridor_prompt_state() -> None:
                 return_value=(("openai:gpt-5.5", "medium"), ("openai:gpt-5.5", "low")),
             ),
             patch.object(server, "fallback_model_id_for", return_value=None),
-            patch.object(server, "make_model", return_value=MagicMock()),
+            patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
             patch.object(
                 server, "_load_observability_tools", new_callable=AsyncMock, return_value=[]
             ),

tests/test_dispatch.py

@@ -105,7 +105,9 @@ async def test_create_durable_run_no_webhook_when_none(monkeypatch: pytest.Monke
 
 
 @pytest.mark.asyncio
-async def test_dispatch_agent_run_preserves_multitask_strategy(monkeypatch: pytest.MonkeyPatch) -> None:
+async def test_dispatch_agent_run_preserves_multitask_strategy(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
     client = _FakeClient()
     monkeypatch.setattr(dispatch, "COMPLETION_WEBHOOK_URL", None)
 

tests/test_proxy_auth.py

@@ -333,7 +333,7 @@ class TestRefreshProxyOnSandboxReuse:
                 new_callable=AsyncMock,
                 return_value=mock_sandbox,
             ),
-            patch("agent.server.make_model", return_value=MagicMock()),
+            patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
             patch("agent.server.construct_system_prompt", return_value="prompt"),
             patch("agent.server.create_deep_agent", return_value=_DummyAgent()),
             patch.dict(
@@ -386,7 +386,7 @@ class TestRefreshProxyOnSandboxReuse:
                 new_callable=AsyncMock,
                 return_value="/workspace",
             ),
-            patch("agent.server.make_model", return_value=MagicMock()),
+            patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
             patch("agent.server.construct_system_prompt", return_value="prompt"),
             patch("agent.server.create_deep_agent", return_value=_DummyAgent()),
             patch.dict("agent.server.SANDBOX_BACKENDS", {}, clear=True),

tests/test_reviewer.py

@@ -204,7 +204,7 @@ async def test_reviewer_resolves_app_installation_token_at_run_start() -> None:
             new_callable=AsyncMock,
             return_value="/workspace",
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", return_value=dummy_agent) as create_agent,
     ):
         await reviewer.get_reviewer_agent(config)
@@ -260,7 +260,7 @@ async def test_reviewer_reuses_app_token_for_sandbox_proxy() -> None:
             new_callable=AsyncMock,
         ),
         patch("agent.reviewer.fetch_agents_md", new_callable=AsyncMock, return_value=None),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", return_value=_DummyAgent()),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -335,7 +335,7 @@ async def test_reviewer_applies_eval_model_and_effort_overrides() -> None:
             new_callable=AsyncMock,
             return_value="/workspace",
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()) as make_model,
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()) as make_model,
         patch("agent.reviewer.create_deep_agent", return_value=dummy_agent),
         patch(
             "agent.reviewer.fetch_agents_md",
@@ -383,7 +383,7 @@ async def test_reviewer_subagent_inherits_eval_model_without_explicit_override()
             new_callable=AsyncMock,
             return_value="/workspace",
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()) as make_model,
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()) as make_model,
         patch("agent.reviewer.create_deep_agent", return_value=dummy_agent),
         patch(
             "agent.reviewer.fetch_agents_md",
@@ -441,7 +441,7 @@ async def test_reviewer_injects_repo_style_during_eval() -> None:
             new_callable=AsyncMock,
             return_value="Flag table rerender regressions.",
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
         patch(
             "agent.reviewer.fetch_agents_md",
@@ -507,7 +507,7 @@ async def test_reviewer_inlines_org_guidelines_into_system_prompt() -> None:
             new_callable=AsyncMock,
             return_value=None,
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -572,7 +572,7 @@ async def test_reviewer_inlines_agents_md_into_system_prompt() -> None:
             new_callable=AsyncMock,
             return_value="Always use the design system IconButton.",
         ) as mock_fetch_agents_md,
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -624,7 +624,7 @@ async def test_reviewer_inlines_claude_md_when_agents_md_absent() -> None:
             new_callable=AsyncMock,
             return_value="# CLAUDE.md\nUse semantic tokens only.",
         ) as mock_fetch_agents_md,
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -989,7 +989,7 @@ async def test_reviewer_injects_pr_review_threads_into_first_review_context() ->
             new_callable=AsyncMock,
             return_value=fake_threads,
         ) as mock_fetch_threads,
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1065,7 +1065,7 @@ async def test_reviewer_injects_pr_review_threads_into_re_review_context() -> No
             new_callable=AsyncMock,
             return_value=[],
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1123,7 +1123,7 @@ async def test_reviewer_omits_threads_block_when_fetch_returns_empty() -> None:
             new_callable=AsyncMock,
             return_value=[],
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1182,7 +1182,7 @@ async def test_reviewer_continues_when_thread_fetch_raises() -> None:
             new_callable=AsyncMock,
             side_effect=RuntimeError("network down"),
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1254,7 +1254,7 @@ async def test_reviewer_populates_diff_line_set_from_github_api() -> None:
             new_callable=AsyncMock,
             return_value=pr_diff,
         ) as mock_fetch_diff,
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1319,7 +1319,7 @@ async def test_reviewer_leaves_validation_disabled_when_diff_fetch_fails() -> No
             new_callable=AsyncMock,
             return_value=None,
         ),
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)
@@ -1385,7 +1385,7 @@ async def test_reviewer_injects_pr_title_and_body_into_context() -> None:
             new_callable=AsyncMock,
             return_value=("Add retry logic for uploads", "Retries flaky uploads up to 3 times."),
         ) as mock_fetch_metadata,
-        patch("agent.reviewer.make_model", return_value=MagicMock()),
+        patch("agent.utils.deferred_model.make_model", return_value=MagicMock()),
         patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
     ):
         await reviewer.get_reviewer_agent(config)