Fix handoff workflow context management and improve AG-UI demo (#5136)

python/packages/core/agent_framework/_workflows/_agent_executor.py

@@ -339,6 +339,13 @@ class AgentExecutor(Executor):
             ctx.get_state(WORKFLOW_RUN_KWARGS_KEY, {})
         )
 
+        if not self._cache:
+            logger.warning(
+                "AgentExecutor %s: Running agent with empty message cache. "
+                "This could lead to service error for some LLM providers.",
+                self.id,
+            )
+
         run_agent = cast(Callable[..., Awaitable[AgentResponse[Any]]], self._agent.run)
         response = await run_agent(
             self._cache,
@@ -371,6 +378,13 @@ class AgentExecutor(Executor):
             ctx.get_state(WORKFLOW_RUN_KWARGS_KEY, {})
         )
 
+        if not self._cache:
+            logger.warning(
+                "AgentExecutor %s: Running agent with empty message cache. "
+                "This could lead to service error for some LLM providers.",
+                self.id,
+            )
+
         updates: list[AgentResponseUpdate] = []
         streamed_user_input_requests: list[Content] = []
         run_agent_stream = cast(Callable[..., ResponseStream[AgentResponseUpdate, AgentResponse[Any]]], self._agent.run)

python/packages/core/agent_framework/_workflows/_runner.py

@@ -186,6 +186,12 @@ class Runner:
                 """Inner loop to deliver a single message through an edge runner."""
                 return await edge_runner.send_message(message, self._state, self._ctx)
 
+            async def _deliver_messages_for_edge_runner(edge_runner: EdgeRunner) -> None:
+                # Preserve message order per edge runner (and therefore per routed target path)
+                # while still allowing parallelism across different edge runners.
+                for message in source_messages:
+                    await _deliver_message_inner(edge_runner, message)
+
             # Route all messages through normal workflow edges
             associated_edge_runners = self._edge_runner_map.get(source_executor_id, [])
             if not associated_edge_runners:
@@ -193,12 +199,6 @@ class Runner:
                 logger.debug(f"No outgoing edges found for executor {source_executor_id}; dropping messages.")
                 return
 
-            async def _deliver_messages_for_edge_runner(edge_runner: EdgeRunner) -> None:
-                # Preserve message order per edge runner (and therefore per routed target path)
-                # while still allowing parallelism across different edge runners.
-                for message in source_messages:
-                    await _deliver_message_inner(edge_runner, message)
-
             tasks = [_deliver_messages_for_edge_runner(edge_runner) for edge_runner in associated_edge_runners]
             await asyncio.gather(*tasks)