Operations
Use this page to choose the right operational path. Avoid replaying writes merely because a request timed out; first inspect durable Work, Job, Run, and runtime state.
Normal checks
- Confirm the selected repository and checkout.
- Check Controller and repository readiness.
- Inspect active Work or Jobs before retrying an operation.
- Run the smallest focused validation first.
- Record exact evidence before finalizing or releasing.
Runtime incidents
- Primary MCP unavailable but Recovery reachable: use the independent Recovery channel to verify the Forge Runtime service, local endpoint, authenticated MCP flow, and external tunnel separately.
- External 502/503: distinguish tunnel transport, local Runtime health, and authenticated MCP failures before restarting anything.
- Release drift: stop activation and compare source commit, immutable whole-release manifest, service definition, release authority, SQLite backup binding, and running Runtime revision.
- Stuck work: use watchdog and maintenance diagnostics before deleting state or killing processes.