[Fix BUG] 400 Reasoning Engine Execution failed on LLM Auditor #560

[marktech0813]

I updated the llm_auditor Agent Engine deployment to use modern, compatible dependency versions for the managed runtime, which should prevent the “Cannot send a request, as the client has been closed” crash that’s commonly caused by older google-adk/google-genai/aiplatform combos inside the managed container. Now you’ll get google-adk >= 1.6.0, google-cloud-aiplatform[adk,agent_engines] >= 1.120.0, and google-genai >= 1.6.0.

What to do next Redeploy the LLM Auditor Agent Engine so the new requirements take effect: Rebuild/recreate the remote agent via your existing deploy command for python/agents/llm-auditor/deployment/deploy.py with the “create” option. Then run your create_session(user_id="new_user") again.

If it still fails, check these quickly:

• Ensure vertexai.init(project, location, staging_bucket) matches the region of your Agent Engine.
• Verify the runtime service account has at least: roles/aiplatform.user roles/logging.logWriter roles/storage.objectAdmin (for staging bucket)
• If you use any genai.Client() directly, prefer genai.Client(vertexai=True, project=PROJECT, location=LOCATION) and don’t reuse a client after a context (with ...:) exits.

Contribution by Gittensor, learn more at https://gittensor.io/

[marktech0813]

Please check PR. thanks.