prod-incident

Files (1)

• SKILL.md

SKILL.md (first lines)

Strict order: stabilize first, diagnose later. Do not
propose refactors or root cause while the system is degraded.

## Phase 1: Assess (minutes)
- Scope: what is broken, since when, how many users are affected,
  is data loss or corruption ongoing?
- What changed recently: deploys, migrations, configs, external API
  quotas, credential expiration.

## Phase 2: Mitigate
- Propose the fastest and most reversible mitigation first: rollback
  of the last deploy, feature flag, restart, scale resources,
  disable the non-critical component.
- Production safety: exact command + scope + rollback,
  and user confirmation before running anything in prod.
- Draft (if the user wants) a short client notice:
  what is affected, what is being done, when there will be updates.

## Phase 3: Diagnose
- Only once the system is stable: move to the systematic-debugging skill
  for root cause, using logs/evidence captured during the
  incident.

## Phase 4: Close
- Brief postmortem in docs/incidents/YYYY-MM-DD-<slug>.md: timeline,
  impact, root cause, what would have prevented it, actions with
  owner. Blameless: system causes, not people.
- If the cause is likely to recur, record it with the `kb` skill.

From G4rb4sh/MLE-setup, MIT.

More from G4rb4sh/MLE-setup