fix: use built SDK wheel for tutorial tests and fix tutorial agent bugs

scale-ballen · claude · scale-ballen · commit 8f007bca4af6 · 2026-03-23T16:15:42.000-04:00
- run_agent_test.sh: pass built wheel to pytest so tests use local SDK
  instead of PyPI version (fixes send_message NDJSON pydantic errors)
- agents.py: rewrite send_message (sync+async) to use streaming internally
  and handle NDJSON responses from FastACP server
- agent_rpc_response.py: make SendMessageStreamResponse.result Optional
  to handle null result in streaming done events
- hooks.py: fix execute_activity_method -&gt; execute_activity for
  function-based Temporal activities (fixes tool_request not appearing)
- Tutorial handlers: add missing return after no-API-key sorry messages
  to prevent fall-through to LLM calls (010_multiturn, 020_streaming,
  030_tracing, 040_other_sdks, 010_agent_chat, 050_agent_chat_guardrails)
- 010_agent_chat/workflow.py: fix gpt-5 -&gt; gpt-4o, remove invalid
  reasoning params (only valid for o-series models)
- 080_human_in_the_loop/workflow.py: guard span.output access against None
- test_agent.py (010_multiturn): add sleep for async state init race

Co-Authored-By: Claude Sonnet 4.6 &lt;noreply@anthropic.com&gt;
diff --git a/examples/tutorials/10_async/00_base/010_multiturn/project/acp.py b/examples/tutorials/10_async/00_base/010_multiturn/project/acp.py
@@ -94,6 +94,7 @@ async def handle_event_send(params: SendEventParams):
                 content="Hey, sorry I'm unable to respond to your message because you're running this example without an OpenAI API key. Please set the OPENAI_API_KEY environment variable to run this example. Do this by either by adding a .env file to the project/ directory or by setting the environment variable in your terminal.",
             ),
         )
+        return
 
     #########################################################
     # 5. (👋) Retrieve the task state.
diff --git a/examples/tutorials/10_async/00_base/010_multiturn/tests/test_agent.py b/examples/tutorials/10_async/00_base/010_multiturn/tests/test_agent.py
@@ -145,6 +145,7 @@ async def test_send_event_and_stream(self, client: AsyncAgentex, agent_id: str):
         task = task_response.result
         assert task is not None
 
+        await asyncio.sleep(1)  # wait for state to be initialized
         # Check initial state
         states = await client.states.list(agent_id=agent_id, task_id=task.id)
         assert len(states) == 1
diff --git a/examples/tutorials/10_async/00_base/020_streaming/project/acp.py b/examples/tutorials/10_async/00_base/020_streaming/project/acp.py
@@ -76,6 +76,7 @@ async def handle_event_send(params: SendEventParams):
                 content="Hey, sorry I'm unable to respond to your message because you're running this example without an OpenAI API key. Please set the OPENAI_API_KEY environment variable to run this example. Do this by either by adding a .env file to the project/ directory or by setting the environment variable in your terminal.",
             ),
         )
+        return
 
     #########################################################
     # 5. Retrieve the task state.
diff --git a/examples/tutorials/10_async/00_base/030_tracing/project/acp.py b/examples/tutorials/10_async/00_base/030_tracing/project/acp.py
@@ -117,6 +117,7 @@ async def handle_event_send(params: SendEventParams):
                 ),
                 parent_span_id=span.id if span else None,
             )
+            return
 
         #########################################################
         # 7. Call an LLM to respond to the user's message
diff --git a/examples/tutorials/10_async/00_base/040_other_sdks/project/acp.py b/examples/tutorials/10_async/00_base/040_other_sdks/project/acp.py
@@ -111,6 +111,7 @@ async def handle_event_send(params: SendEventParams):
                 ),
                 parent_span_id=span.id if span else None,
             )
+            return
 
         #########################################################
         # (👋) Call an LLM to respond to the user's message using custom streaming
diff --git a/examples/tutorials/10_async/10_temporal/010_agent_chat/project/workflow.py b/examples/tutorials/10_async/10_temporal/010_agent_chat/project/workflow.py
@@ -216,6 +216,7 @@ async def on_task_event_send(self, params: SendEventParams) -> None:
                     ),
                     parent_span_id=span.id if span else None,
                 )
+                return
 
             # Call an LLM to respond to the user's message
             # When send_as_agent_task_message=True, returns a TaskMessage
@@ -234,13 +235,7 @@ async def on_task_event_send(self, params: SendEventParams) -> None:
                     "to provide accurate and well-reasoned responses."
                 ),
                 parent_span_id=span.id if span else None,
-                model="gpt-5",
-                model_settings=ModelSettings(
-                    # Include reasoning items in the response (IDs, summaries)
-                    # response_include=["reasoning.encrypted_content"],
-                    # Ask the model to include a short reasoning summary
-                    reasoning=Reasoning(effort="medium", summary="detailed"),
-                ),
+                model="gpt-4o",
                 # tools=[CALCULATOR_TOOL],
             )
             if self._state:
diff --git a/examples/tutorials/10_async/10_temporal/050_agent_chat_guardrails/project/workflow.py b/examples/tutorials/10_async/10_temporal/050_agent_chat_guardrails/project/workflow.py
@@ -408,6 +408,7 @@ async def on_task_event_send(self, params: SendEventParams) -> None:
                     ),
                     parent_span_id=span.id if span else None,
                 )
+                return
 
             # Call an LLM to respond to the user's message
             # When send_as_agent_task_message=True, returns a TaskMessage
diff --git a/examples/tutorials/10_async/10_temporal/080_open_ai_agents_sdk_human_in_the_loop/project/workflow.py b/examples/tutorials/10_async/10_temporal/080_open_ai_agents_sdk_human_in_the_loop/project/workflow.py
@@ -204,7 +204,8 @@ async def on_task_event_send(self, params: SendEventParams) -> None:
                         self._state.input_list.append(msg)
 
             # Set span output for tracing - include full state
-            span.output = self._state.model_dump()
+            if span:
+                span.output = self._state.model_dump()
 
     @workflow.run
     async def on_task_create(self, params: CreateTaskParams) -> str:
diff --git a/examples/tutorials/run_agent_test.sh b/examples/tutorials/run_agent_test.sh
@@ -259,6 +259,19 @@ run_test() {
     cd "$tutorial_path" || return 1
 
 
+    # Determine pytest command - use built wheel if available (same wheel used to start agent)
+    local pytest_cmd="uv run pytest"
+    if [ "$BUILD_CLI" = true ]; then
+        local wheel_file=$(ls /home/runner/work/*/*/dist/agentex_sdk-*.whl 2>/dev/null | head -n1)
+        if [[ -z "$wheel_file" ]]; then
+            # Fallback for local development
+            wheel_file=$(ls "${SCRIPT_DIR}/../../dist/agentex_sdk-*.whl" 2>/dev/null | head -n1)
+        fi
+        if [[ -n "$wheel_file" ]]; then
+            pytest_cmd="uv run --with $wheel_file pytest"
+        fi
+    fi
+
     # Run the tests with retry mechanism
     local max_retries=5
     local retry_count=0
@@ -270,7 +283,7 @@ run_test() {
         fi
 
         # Stream pytest output directly in real-time
-        uv run pytest tests/test_agent.py -v -s
+        $pytest_cmd tests/test_agent.py -v -s
         exit_code=$?
 
         if [ $exit_code -eq 0 ]; then
diff --git a/src/agentex/lib/core/temporal/plugins/openai_agents/hooks/hooks.py b/src/agentex/lib/core/temporal/plugins/openai_agents/hooks/hooks.py
@@ -131,7 +131,7 @@ async def on_tool_start(self, context: RunContextWrapper, agent: Agent, tool: To
                 logger.warning(f"Failed to parse tool arguments: {tool_context.tool_arguments}")
                 tool_arguments = {}
 
-        await workflow.execute_activity_method(
+        await workflow.execute_activity(
             stream_lifecycle_content,
             args=[
                 self.task_id,
@@ -167,7 +167,7 @@ async def on_tool_end(
             else f"call_{id(tool)}"
         )
 
-        await workflow.execute_activity_method(
+        await workflow.execute_activity(
             stream_lifecycle_content,
             args=[
                 self.task_id,
@@ -195,7 +195,7 @@ async def on_handoff(
             from_agent: The agent transferring control
             to_agent: The agent receiving control
         """
-        await workflow.execute_activity_method(
+        await workflow.execute_activity(
             stream_lifecycle_content,
             args=[
                 self.task_id,
diff --git a/src/agentex/resources/agents.py b/src/agentex/resources/agents.py
@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 import json
-from typing import Union, Optional, Generator, AsyncGenerator
+from typing import Any, Union, Optional, Generator, AsyncGenerator
 from typing_extensions import Literal
 
 import httpx
@@ -453,38 +453,73 @@ def send_message(
     ) -> SendMessageResponse:
       if agent_id is not None and agent_name is not None:
         raise ValueError("Either agent_id or agent_name must be provided, but not both")
-      
+
       if "stream" in params and params["stream"] == True:
         raise ValueError("If stream is set to True, use send_message_stream() instead")
+
+      if agent_id is not None:
+        streaming_response = self.with_streaming_response.rpc(
+          agent_id=agent_id,
+          method="message/send",
+          params=params,
+          id=id,
+          jsonrpc=jsonrpc,
+          extra_headers=extra_headers,
+          extra_query=extra_query,
+          extra_body=extra_body,
+          timeout=timeout,
+        )
+      elif agent_name is not None:
+        streaming_response = self.with_streaming_response.rpc_by_name(
+          agent_name=agent_name,
+          method="message/send",
+          params=params,
+          id=id,
+          jsonrpc=jsonrpc,
+          extra_headers=extra_headers,
+          extra_query=extra_query,
+          extra_body=extra_body,
+          timeout=timeout,
+        )
       else:
-        if agent_id is not None:
-          raw_agent_rpc_response = self.rpc(
-            agent_id=agent_id,
-            method="message/send",
-            params=params,
-            id=id,
-            jsonrpc=jsonrpc,
-            extra_headers=extra_headers,
-            extra_query=extra_query,
-            extra_body=extra_body,
-            timeout=timeout,
-          )
-        elif agent_name is not None:
-          raw_agent_rpc_response = self.rpc_by_name(
-            agent_name=agent_name,
-            method="message/send",
-            params=params,
-            id=id,
-            jsonrpc=jsonrpc,
-            extra_headers=extra_headers,
-            extra_query=extra_query,
-            extra_body=extra_body,
-            timeout=timeout,
-          )
-        else:
-          raise ValueError("Either agent_id or agent_name must be provided")
-        
-        return SendMessageResponse.model_validate(raw_agent_rpc_response, from_attributes=True)
+        raise ValueError("Either agent_id or agent_name must be provided")
+
+      task_messages: list[Any] = []
+      response_meta: dict[str, Any] = {}
+
+      with streaming_response as response:
+        for _line in response.iter_lines():
+          if not _line:
+            continue
+          line = _line.strip()
+          if line.startswith("data:"):
+            line = line[len("data:"):].strip()
+          if not line:
+            continue
+          try:
+            chunk = json.loads(line)
+            if not response_meta:
+              response_meta = {"id": chunk.get("id"), "jsonrpc": chunk.get("jsonrpc")}
+            # If the server already aggregated into a list result, return directly
+            try:
+              return SendMessageResponse.model_validate(chunk)
+            except ValidationError:
+              pass
+            # Parse as a streaming event and collect parent_task_message
+            chunk_stream = SendMessageStreamResponse.model_validate(chunk, from_attributes=True)
+            result = chunk_stream.result
+            if result is not None and hasattr(result, 'type') and result.type == 'full':
+              parent = getattr(result, 'parent_task_message', None)
+              if parent is not None:
+                task_messages.append(parent)
+          except (json.JSONDecodeError, ValidationError):
+            continue
+
+      return SendMessageResponse(
+        id=response_meta.get("id"),
+        jsonrpc=response_meta.get("jsonrpc"),
+        result=task_messages,
+      )
     
     def send_message_stream(
       self,
@@ -552,10 +587,10 @@ def send_message_stream(
               from_attributes=True
             )
             yield chunk_rpc_response
-          except json.JSONDecodeError:
-            # Skip invalid JSON lines
+          except (json.JSONDecodeError, ValidationError):
+            # Skip invalid JSON lines or lines that can't be validated
             continue
-    
+
     def send_event(
       self,
       agent_id: str | None = None,
@@ -1021,38 +1056,73 @@ async def send_message(
     ) -> SendMessageResponse:
       if agent_id is not None and agent_name is not None:
         raise ValueError("Either agent_id or agent_name must be provided, but not both")
-      
+
       if "stream" in params and params["stream"] == True:
         raise ValueError("If stream is set to True, use send_message_stream() instead")
+
+      if agent_id is not None:
+        streaming_response = self.with_streaming_response.rpc(
+          agent_id=agent_id,
+          method="message/send",
+          params=params,
+          id=id,
+          jsonrpc=jsonrpc,
+          extra_headers=extra_headers,
+          extra_query=extra_query,
+          extra_body=extra_body,
+          timeout=timeout,
+        )
+      elif agent_name is not None:
+        streaming_response = self.with_streaming_response.rpc_by_name(
+          agent_name=agent_name,
+          method="message/send",
+          params=params,
+          id=id,
+          jsonrpc=jsonrpc,
+          extra_headers=extra_headers,
+          extra_query=extra_query,
+          extra_body=extra_body,
+          timeout=timeout,
+        )
       else:
-        if agent_id is not None:
-          raw_agent_rpc_response = await self.rpc(
-            agent_id=agent_id,
-            method="message/send",
-            params=params,
-            id=id,
-            jsonrpc=jsonrpc,
-            extra_headers=extra_headers,
-            extra_query=extra_query,
-            extra_body=extra_body,
-            timeout=timeout,
-          )
-        elif agent_name is not None:
-          raw_agent_rpc_response = await self.rpc_by_name(
-            agent_name=agent_name,
-            method="message/send",
-            params=params,
-            id=id,
-            jsonrpc=jsonrpc,
-            extra_headers=extra_headers,
-            extra_query=extra_query,
-            extra_body=extra_body,
-            timeout=timeout,
-          )
-        else:
-          raise ValueError("Either agent_id or agent_name must be provided")
-        
-        return SendMessageResponse.model_validate(raw_agent_rpc_response, from_attributes=True)
+        raise ValueError("Either agent_id or agent_name must be provided")
+
+      task_messages: list[Any] = []
+      response_meta: dict[str, Any] = {}
+
+      async with streaming_response as response:
+        async for _line in response.iter_lines():
+          if not _line:
+            continue
+          line = _line.strip()
+          if line.startswith("data:"):
+            line = line[len("data:"):].strip()
+          if not line:
+            continue
+          try:
+            chunk = json.loads(line)
+            if not response_meta:
+              response_meta = {"id": chunk.get("id"), "jsonrpc": chunk.get("jsonrpc")}
+            # If the server already aggregated into a list result, return directly
+            try:
+              return SendMessageResponse.model_validate(chunk)
+            except ValidationError:
+              pass
+            # Parse as a streaming event and collect parent_task_message
+            chunk_stream = SendMessageStreamResponse.model_validate(chunk, from_attributes=True)
+            result = chunk_stream.result
+            if result is not None and hasattr(result, 'type') and result.type == 'full':
+              parent = getattr(result, 'parent_task_message', None)
+              if parent is not None:
+                task_messages.append(parent)
+          except (json.JSONDecodeError, ValidationError):
+            continue
+
+      return SendMessageResponse(
+        id=response_meta.get("id"),
+        jsonrpc=response_meta.get("jsonrpc"),
+        result=task_messages,
+      )
     
     async def send_message_stream(
       self,
diff --git a/src/agentex/types/agent_rpc_response.py b/src/agentex/types/agent_rpc_response.py
@@ -40,7 +40,7 @@ class SendMessageResponse(BaseAgentRpcResponse):
     """The result of the message sending"""
 
 class SendMessageStreamResponse(BaseAgentRpcResponse):
-    result: TaskMessageUpdate
+    result: Optional[TaskMessageUpdate] = None
     """The result of the message sending"""
 
 

Original file line number	Diff line number	Diff line change
`@@ -94,6 +94,7 @@ async def handle_event_send(params: SendEventParams):`
`94`	`94`	`content="Hey, sorry I'm unable to respond to your message because you're running this example without an OpenAI API key. Please set the OPENAI_API_KEY environment variable to run this example. Do this by either by adding a .env file to the project/ directory or by setting the environment variable in your terminal.",`
`95`	`95`	`),`
`96`	`96`	`)`
	`97`	`+ return`
`97`	`98`
`98`	`99`	`#########################################################`
`99`	`100`	`# 5. (👋) Retrieve the task state.`
Original file line number	Diff line number	Diff line change
`@@ -76,6 +76,7 @@ async def handle_event_send(params: SendEventParams):`
`76`	`76`	`content="Hey, sorry I'm unable to respond to your message because you're running this example without an OpenAI API key. Please set the OPENAI_API_KEY environment variable to run this example. Do this by either by adding a .env file to the project/ directory or by setting the environment variable in your terminal.",`
`77`	`77`	`),`
`78`	`78`	`)`
	`79`	`+ return`
`79`	`80`
`80`	`81`	`#########################################################`
`81`	`82`	`# 5. Retrieve the task state.`
Original file line number	Diff line number	Diff line change
`@@ -117,6 +117,7 @@ async def handle_event_send(params: SendEventParams):`
`117`	`117`	`),`
`118`	`118`	`parent_span_id=span.id if span else None,`
`119`	`119`	`)`
	`120`	`+ return`
`120`	`121`
`121`	`122`	`#########################################################`
`122`	`123`	`# 7. Call an LLM to respond to the user's message`
Original file line number	Diff line number	Diff line change
`@@ -111,6 +111,7 @@ async def handle_event_send(params: SendEventParams):`
`111`	`111`	`),`
`112`	`112`	`parent_span_id=span.id if span else None,`
`113`	`113`	`)`
	`114`	`+ return`
`114`	`115`
`115`	`116`	`#########################################################`
`116`	`117`	`# (👋) Call an LLM to respond to the user's message using custom streaming`
Original file line number	Diff line number	Diff line change
`@@ -408,6 +408,7 @@ async def on_task_event_send(self, params: SendEventParams) -> None:`
`408`	`408`	`),`
`409`	`409`	`parent_span_id=span.id if span else None,`
`410`	`410`	`)`
	`411`	`+ return`
`411`	`412`
`412`	`413`	`# Call an LLM to respond to the user's message`
`413`	`414`	`# When send_as_agent_task_message=True, returns a TaskMessage`