Operate a run

You finish this when you can say why a run stopped and whether a retry moved it.

1. Watch, then read history

Monitoring and debugging is the order: identify yourself, watch the live run, read the history of a FAILED run, retry, and force-terminate only as the last resort. Retry re-runs the failed step. Completed steps keep their outputs.

The same tools exist for an assistant: get_workflow_status, get_workflow_history, retry_workflow. A run you started from chat is the same run.

2. Turn on Lumen when you need the logs

Status and history answer "what state is it in". Logs, error groups, and traces are Lumen. Enable it before you expect a line to be there. Sending data and the query API are the next pages on that section.

3. Make a ticket only when a person should change code

When the failure should become a pull request, Tickets and software delivery is the path from the error to a governed change. The actor still does not hold git credentials.