fixes

Author: Benny Chen

eval_protocol/integrations/deepeval.py

@@ -80,6 +80,9 @@ def _build_case_kwargs() -> Dict[str, Any]:
             return case_kwargs
 
         if BaseConversationalMetric is not None and isinstance(metric, BaseConversationalMetric):
+            # Narrow types for optional imports to satisfy the type checker
+            assert LLMTestCase is not None
+            assert ConversationalTestCase is not None
             turns = []
             for i, msg in enumerate(messages):
                 turn_input = messages[i - 1].get("content", "") if i > 0 else ""
@@ -93,6 +96,8 @@ def _build_case_kwargs() -> Dict[str, Any]:
                 output = messages[-1].get("content", "")
             test_case = ConversationalTestCase(turns=turns)
         else:
+            # Narrow types for optional imports to satisfy the type checker
+            assert LLMTestCase is not None
             case_kwargs = _build_case_kwargs()
             test_case = LLMTestCase(**case_kwargs)
 

eval_protocol/mcp_servers/tau2/tau2_mcp.py

@@ -422,7 +422,7 @@ def _register_tools(self):
         """Register mock-specific MCP tools matching τ²-Bench schemas"""
 
         @self.mcp.tool(name="create_task", description="Create a new task for a user.")
-        def create_task(user_id: str, title: str, ctx: Context, description: str = None) -> Dict[str, Any]:
+        def create_task(user_id: str, title: str, ctx: Context, description: Optional[str] = None) -> Dict[str, Any]:
             """Create a new task for a user"""
             session_id = self._get_session_id(ctx)