openclaw - 💡(How to fix) Fix Sub-agent reports 'completed successfully' when provider returns stopReason: error [1 participants]

When a sub-agent session receives a provider-level error (stopReason: error, finish_reason: error, 0 tokens), the parent agent is notified with status: completed successfully and (no output). This masks the failure and prevents automatic retry/fallback logic. 2. If the upstream provider crashes or returns an error during processing, the sub-agent session ends with:

• stopReason: error
• errorMessage: Provider finish_reason: error When stopReason is error, the completion event should propagate the error state:
• status: completed with error (or failed) status: completed successfully with empty result. The parent has no indication that an error occurred unless it independently verifies the sub-agent's output. "stopReason": "error", "errorMessage": "Provider finish_reason: error",
• Parent agents cannot distinguish between "sub-agent completed with empty output" (legitimate) and "provider crashed before executing" (error)
• stopReason: error → completed with error + forward errorMessage

TL;DR

Modify the sub-agent completion handler to propagate error states to the parent agent based on the stopReason value.

Guidance

• Check the stopReason in the sub-agent completion handler and update the status accordingly:
  • stopReason: error should return status: completed with error along with the errorMessage.
  • stopReason: end_turn or tool_use can return status: completed successfully.
  • Consider handling stopReason: max_tokens to return status: completed (truncated) for better error handling.
• Verify the fix by reproducing the error scenario and checking if the parent agent receives the correct status and error message.
• Review the sub-agent completion handler code to ensure it correctly handles different stopReason values and propagates the error state to the parent agent.

Example

if (stopReason === "error") {
  return {
    status: "completed with error",
    errorMessage: errorMessage
  };
} else if (stopReason === "end_turn" || stopReason === "tool_use") {
  return {
    status: "completed successfully"
  };
}

Notes

The suggested fix assumes that the sub-agent completion handler has access to the stopReason and errorMessage values. If this is not the case, additional modifications may be needed to pass these values to the completion handler.

Recommendation

Apply the workaround by modifying the sub-agent completion handler to correctly propagate error states to the parent agent, as this will allow for better error handling and automated retry/fallback logic.