json.JSONDecodeError inherits from ValueError, but in run_agent.py the error classification treats ValueError as a local validation error (non-retryable). When a provider returns malformed JSON (transient parse failure), the agent fails immediately instead of retrying, causing unnecessary session failures on provider hiccups. JSONDecodeError should be excluded from the local validation error classification and treated as a retryable transient error.

TL;DR

Update the error classification in run_agent.py to treat json.JSONDecodeError as a retryable transient error instead of a local validation error.

Guidance

• Review the error handling logic in run_agent.py to identify where ValueError is classified as a non-retryable error.
• Update the classification to exclude json.JSONDecodeError from ValueError or create a separate handling path for json.JSONDecodeError to allow retries.
• Verify that the agent now retries requests when encountering malformed JSON responses from providers.
• Test the updated error handling with a provider that occasionally returns malformed JSON to ensure the agent behaves as expected.

Example

except json.JSONDecodeError as e:
    # Treat as retryable transient error
    retry_request()
except ValueError as e:
    # Handle other ValueError cases as non-retryable
    handle_local_validation_error(e)

Notes

This fix assumes that the json.JSONDecodeError exception is properly caught and handled in the run_agent.py script. If the issue persists, further debugging may be required to identify other potential causes.

Recommendation

Apply workaround: Update the error classification in run_agent.py to treat json.JSONDecodeError as a retryable transient error, as this directly addresses the identified issue and allows for a more robust handling of transient errors.