| name | operations |
| description | Use when a live service needs observability, incident response, SLOs, runbooks, or continuity work. |
| targets | ["*"] |
| claudecode | {"context":"fork","agent":"sre"} |
This skill defines the operations phase. Use the `sre` agent when bounded production work can proceed independently; otherwise work locally. Produce operational evidence, recovery guidance, and only material follow-up work.
Use lifecycle-documentation only when the result needs a durable artifact or handoff.
Keep the shipped software healthy in the real world. Stand up observability, respond to incidents, and feed what production teaches you back to maintenance and requirements.
Shipping is the start of operations, not the end of the work. Hope is not a strategy — instrument first.
What to run
- Observability — logs, metrics, and traces, with SLOs that define "healthy" and alerts that fire before users notice.
- Incident management — detect, triage by impact, mitigate, resolve, and write a blameless postmortem. Every incident yields a prevention item.
- Runbooks — a written procedure for each routine operation and recovery path, so response doesn't depend on one person's memory.
- Continuity — backup, failover, and rollback that have actually been tested, not assumed.
- Close the loop — route defects to
debugging-and-error-recovery, recurring toil to maintenance, and new needs to requirements-gathering.
What to produce
An operations plan and runbooks, the monitoring and SLOs in place, incident records with postmortems, and the feedback items routed back into the lifecycle.
