A2A Tasks Fail Silently on agent_buildpack 0.0.53

Products

• VMware Tanzu Platform Core
• Vmware Tanzu Platform - SM

Issue/Introduction

Customers utilizing agent_buildpack version 0.0.53 (included in AI Services 10.4.3) may observe the following behavior with native Agent-to-Agent (A2A) tasks:

• Inbound A2A tasks transition from submitted to failed almost immediately (typically within 10–20ms).
• The task history shows no agent reply or output artifacts.
• Application logs (via cf logs) show a 200 OK for the POST /a2a/ request but emit no ERROR, traceback, or debug information regarding the failure.
• Standard chat endpoints (/api/chat) and MCP/Skill integrations continue to function correctly.
• Startup logs may contain the following deprecation warning:

  Agent.to_a2a() is deprecated … use fasta2a.pydantic_ai.agent_to_a2a

Environment

Product: VMware Tanzu AI Services / Agent Buildpack
Component: Agent-to-Agent (A2A) Executor

Cause

A breaking change was introduced in agent_buildpack 0.0.53 as part of an internal refactor of the A2A executor. The legacy Agent.to_a2a() method and the bundled [a2a] extra have been removed from the core integration. Applications still utilizing the legacy pattern will fail to process inbound tasks because the internal routing has moved to the independent fasta2a library.

Resolution

Engineering is currently investigating the "silent" nature of this failure to ensure better error reporting in future buildpack releases. The fix will be included in an upcoming patch release. In the meantime, avoid agent_buildpack 0.0.53 and use agent_buildpack 0.0.44 or earlier.

Please subscribe to this article to receive notifications as further updates or permanent fixes are released

Additional Information

Deliverables to be included in upcoming patch:

1. Transition the agent entry point from the deprecated Agent.to_a2a() to the new pattern: fasta2a.pydantic_ai.agent_to_a2a.
2. Update dependencies to include fasta2a[pydantic-ai].
3. Ensure A2A tasks submitted via POST /a2a/ complete successfully with status.state = "completed" and generate a valid agent reply in the task history.
4. Improve error handling in the A2A executor so failures are not silent, returning proper error details in the task's status.message and logging tracebacks.