Skip to content

A2aAgentExecutor sends the raw exception text to the A2A caller when the run fails #7385

Description

@lamberttraccard

🔴 Required Information

Describe the Bug:

When the agent run raises, A2aAgentExecutor.execute ends the task with a TASK_STATE_FAILED status whose message is str(e) (make_text_part(str(e)) in a2a/executor/a2a_agent_executor.py, and the same line in a2a_agent_executor_impl.py). Whatever the exception says goes to the A2A caller verbatim.

Some exceptions say a lot. LiteLLM's APIConnectionError, for one, embeds the Python traceback in its message, so the caller receives the server's file paths:

litellm.APIConnectionError: litellm.APIError: LiteLLM: provider returned a response with no 'choices'. ...
Traceback (most recent call last):
  File "<server venv>/lib/python3.12/site-packages/litellm/llms/openai/openai.py", line 900, in acompletion
    final_response_obj = convert_to_model_response_object(
...

When the caller is an ADK RemoteA2aAgent, that text then reaches the calling agent's event stream as a plain text part (#6708), so its end user reads it as the agent's answer. That is how we found it: a user of one of our agents got the traceback above as a reply.

The server already logs the exception with exc_info=True on the line before, so the caller does not need the detail.

Steps to Reproduce:

  1. pip install "google-adk[a2a]==2.11.0" "a2a-sdk[http-server]==1.2.1"
  2. Run the script below. It serves an ADK A2aAgentExecutor in process over httpx.ASGITransport, with a model that raises, and sends one SendMessage. It needs no network or credentials.
repro.py
"""What an A2A caller receives when the served ADK agent's run raises."""

import asyncio
import json
import logging
import warnings

import httpx
from a2a.server.request_handlers import DefaultRequestHandler
from a2a.server.routes import add_a2a_routes_to_fastapi, create_jsonrpc_routes
from a2a.server.tasks import InMemoryTaskStore
from a2a.types import AgentCapabilities, AgentCard, AgentInterface
from fastapi import FastAPI
from google.adk.a2a.executor.a2a_agent_executor import A2aAgentExecutor
from google.adk.agents import LlmAgent
from google.adk.models.base_llm import BaseLlm
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService

warnings.filterwarnings("ignore")
logging.disable(logging.CRITICAL)


class FailingModel(BaseLlm):  # stands in for a provider error
    async def generate_content_async(self, llm_request, stream=False):
        raise RuntimeError(
            "provider rejected the call: Traceback (most recent call last):\n"
            '  File "/srv/agent/.venv/lib/python3.12/site-packages/provider/client.py", line 900'
        )
        yield


async def main():
    card = AgentCard(
        name="agent",
        description="agent",
        version="1",
        capabilities=AgentCapabilities(streaming=True),
        default_input_modes=["text/plain"],
        default_output_modes=["text/plain"],
        supported_interfaces=[
            AgentInterface(url="http://agent.test/a2a", protocol_binding="JSONRPC")
        ],
    )
    agent = LlmAgent(name="agent", model=FailingModel(model="fake"), instruction="x")
    executor = A2aAgentExecutor(
        runner=Runner(
            app_name="app",
            agent=agent,
            session_service=InMemorySessionService(),
            auto_create_session=True,
        )
    )
    app = FastAPI()
    add_a2a_routes_to_fastapi(
        app,
        jsonrpc_routes=create_jsonrpc_routes(
            DefaultRequestHandler(executor, InMemoryTaskStore(), card), rpc_url="/a2a"
        ),
    )
    async with httpx.AsyncClient(
        transport=httpx.ASGITransport(app=app), base_url="http://agent.test"
    ) as client:
        response = await client.post(
            "/a2a",
            headers={"A2A-Version": "1.0"},
            json={
                "jsonrpc": "2.0",
                "id": 1,
                "method": "SendMessage",
                "params": {
                    "message": {
                        "messageId": "m1",
                        "role": "ROLE_USER",
                        "parts": [{"text": "hello"}],
                    }
                },
            },
        )
    status = response.json()["result"]["task"]["status"]
    print(json.dumps(status, indent=2))


asyncio.run(main())

Expected Behavior:

The failed status carries a generic message, or one the server chooses, and the exception's details stay in the server's log. ADK already has a path for a chosen message: convert_event_to_a2a_events turns an event with error_code into a TASK_STATE_FAILED status carrying the event's error_message, or DEFAULT_ERROR_MESSAGE. Only an exception that escapes the run takes the str(e) path.

Observed Behavior:

{
  "state": "TASK_STATE_FAILED",
  "message": {
    "messageId": "5f89d64d-db72-4846-a968-c014801f3870",
    "role": "ROLE_AGENT",
    "parts": [
      {
        "text": "provider rejected the call: Traceback (most recent call last):\n  File \"/srv/agent/.venv/lib/python3.12/site-packages/provider/client.py\", line 900"
      }
    ]
  },
  "timestamp": "2026-10-02T20:45:29.502125Z"
}

Environment Details:

  • ADK Library Version (pip show google-adk): 2.11.0, with a2a-sdk 1.2.1. Both make_text_part(str(e)) lines are unchanged on main.
  • Desktop OS: macOS
  • Python Version (python -V): 3.12.12

Model Information:

  • Are you using LiteLLM: Yes in production, where the exception came from. No in the reproduction, which uses a BaseLlm that raises.
  • Which model is being used: Gemini through a LiteLLM proxy in production, N/A for the reproduction.

🟡 Optional Information

Regression:

Not known.

Related:

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions