fix(codex): poll for response in async live tests; collab_tool_call errors

- The async-base and temporal live tests used the sync send_message pattern,
  which returns an empty result for async (event-driven) agents (assert 0 >= 1).
  Rewrite to create_task + send_event + poll task messages, matching how these
  agents actually handle events (on_task_event_send / RECEIVE_EVENT signal).
- _tool_output_for: route collab_tool_call through the same error/result
  handling as mcp_tool_call so a failed collab call is reported with
  is_error=True instead of a generic success dump [greptile].

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: declan-scale

examples/tutorials/10_async/00_base/harness_codex/tests/test_agent.py

@@ -145,19 +145,44 @@ def client(self):
 
         return Agentex(base_url=AGENTEX_API_BASE_URL)
 
-    def test_send_simple_message(self, client):
+    @pytest.fixture
+    def agent_id(self, client):
+        for agent in client.agents.list():
+            if agent.name == AGENT_NAME:
+                return agent.id
+        raise ValueError(f"Agent {AGENT_NAME!r} not found.")
+
+    def test_send_simple_message(self, client, agent_id: str):
+        """Async agents process events out of band, so create a task, send an
+        event, and poll the task's messages for the agent's response."""
+        import time
+        import uuid
+
         from agentex.types import TextContentParam
-        from agentex.types.agent_rpc_params import ParamsSendMessageRequest
+        from agentex.types.agent_rpc_params import ParamsSendEventRequest, ParamsCreateTaskRequest
+
+        task = client.agents.create_task(agent_id, params=ParamsCreateTaskRequest(name=uuid.uuid1().hex)).result
+        assert task is not None
 
-        response = client.agents.send_message(
-            agent_name=AGENT_NAME,
-            params=ParamsSendMessageRequest(
+        client.agents.send_event(
+            agent_id=agent_id,
+            params=ParamsSendEventRequest(
+                task_id=task.id,
                 content=TextContentParam(
                     author="user",
                     content="What is 3+3? Reply with just the number.",
                     type="text",
-                )
+                ),
             ),
         )
-        assert response.result is not None
-        assert len(response.result) >= 1
+
+        deadline = time.monotonic() + 60
+        while time.monotonic() < deadline:
+            msgs = client.messages.list(task_id=task.id)
+            agent_msgs = [m for m in msgs if getattr(m.content, "author", None) == "agent"]
+            if agent_msgs:
+                assert len(agent_msgs) >= 1
+                return
+            time.sleep(2)
+
+        raise AssertionError("No agent response received within 60 s")

examples/tutorials/10_async/10_temporal/harness_codex/tests/test_agent.py

@@ -199,19 +199,47 @@ def client(self):
 
         return Agentex(base_url=AGENTEX_API_BASE_URL)
 
-    def test_send_simple_message(self, client):
+    @pytest.fixture
+    def agent_id(self, client):
+        for agent in client.agents.list():
+            if agent.name == AGENT_NAME:
+                return agent.id
+        raise ValueError(f"Agent {AGENT_NAME!r} not found.")
+
+    def test_send_simple_message(self, client, agent_id: str):
+        """Temporal agents process events out of band, so create a task, send an
+        event, and poll the task's messages for the agent's response."""
+        import time
+        import uuid
+
         from agentex.types import TextContentParam
-        from agentex.types.agent_rpc_params import ParamsSendMessageRequest
+        from agentex.types.agent_rpc_params import ParamsSendEventRequest, ParamsCreateTaskRequest
 
-        response = client.agents.send_message(
-            agent_name=AGENT_NAME,
-            params=ParamsSendMessageRequest(
+        task = client.agents.create_task(agent_id, params=ParamsCreateTaskRequest(name=uuid.uuid1().hex)).result
+        assert task is not None
+
+        client.agents.send_event(
+            agent_id=agent_id,
+            params=ParamsSendEventRequest(
+                task_id=task.id,
                 content=TextContentParam(
                     author="user",
                     content="What is 5+5? Reply with just the number.",
                     type="text",
-                )
+                ),
             ),
         )
-        assert response.result is not None
-        assert len(response.result) >= 1
+
+        deadline = time.monotonic() + 90
+        while time.monotonic() < deadline:
+            msgs = client.messages.list(task_id=task.id)
+            agent_msgs = [m for m in msgs if getattr(m.content, "author", None) == "agent"]
+            response_msgs = [
+                m for m in agent_msgs if "Task initialized" not in str(getattr(m.content, "content", ""))
+            ]
+            if response_msgs:
+                assert len(response_msgs) >= 1
+                return
+            time.sleep(3)
+
+        raise AssertionError("No agent response received within 90 s")

src/agentex/lib/adk/_modules/_codex_sync.py

@@ -156,7 +156,10 @@ def _tool_output_for(item_type: str, payload: dict[str, Any]) -> tuple[str, bool
         exit_code = payload.get("exit_code")
         is_error = exit_code is not None and exit_code != 0
         return _truncate(out), is_error
-    if item_type == "mcp_tool_call":
+    if item_type in ("mcp_tool_call", "collab_tool_call"):
+        # collab_tool_call mirrors mcp_tool_call's error/result convention
+        # (see _tool_args_for); without this branch a failed collab call would
+        # fall through to the generic path and be reported as a success.
         err = payload.get("error")
         if err:
             msg = err.get("message", "") if isinstance(err, dict) else str(err)