The claude log contains a final result message with the error state of the agent cli. This flag is currently ignored by our plumbing code.
The generated code in the lock.yml file should scan the output log and fail the action if the agent cli is an error state.