Diagnose and recover
Establish causality, inspect native authorities and preserve accepted work.
For a reported software defect, diagnose before authorizing a fix. A scout establishes a falsifiable cause, controls alternatives and returns evidence. Project implementation begins only after that report supports a bounded ship outcome.
For Fleet trouble, first decide which authority can prove or disprove the symptom:
- PostgreSQL for identity, ownership and durable communication;
- Kubernetes for workload and effective authorization;
- Herdr for the native terminal and harness session;
- PVC state for home and unfinished work;
- Git and provider state for delivered code.
A stale status row is not live-process evidence. A running Pod is not proof the correct harness session is working. A terminal label is not Agent identity.
Recovery reuses the same Agent identity, home and Assignment. It never creates a duplicate Agent to avoid diagnosis or destroys unlanded work. Inspect narrowly, preserve contradictory evidence and stop at the first unverified boundary. The runtime and supervision Skills own the current recovery primitive for the selected release.
Last updated on