Responses: include assistant function_call items but omit reasoning passthrough

- Keep assistant function_call in input so paired tool outputs validate
- Do not echo prior 'reasoning' items to avoid ordering constraints

Co-authored-by: openhands <[email protected]>

Author: enyst

openhands-sdk/openhands/sdk/llm/message.py

@@ -466,9 +466,11 @@ def to_responses_dict(self, *, vision_enabled: bool) -> list[dict[str, Any]]:
             # Do not echo back prior turn's reasoning item. The Responses API
             # does not require sending these back, and including them can cause
             # validation constraints (ordering requirements).
-            # Do NOT emit assistant function_call items in input.
-            # The server already has the assistant tool calls from its previous
-            # output; the client should only send function_call_output items.
+            # Emit assistant function_call items; the client will also send
+            # the corresponding function_call_output items in the same request.
+            if self.tool_calls:
+                for tc in self.tool_calls:
+                    items.append(tc.to_responses_dict())
             return items
 
         if self.role == "tool":