deployment-workflow

Run this Skill before opening the PR. It writes a rollback document with trigger conditions, pre-rollback steps, platform-specific commands for Kubernetes, blue-green, or canary rollbacks, and a five-minute post-rollback verification checklist.

Expand-contract splits a breaking schema change into three deployments: expand adds the new nullable structure while code writes to both old and new, migrate backfills data in small batches, and contract removes the old structure only after code stops reading it.

Choose based on change risk: rolling for stateless services, blue-green for stateful or database-heavy changes needing instant router-level rollback, and canary for high-traffic features needing gradual validation from 1% to 100% traffic.

Yes. The Skill requires an analysis definition enforced by the rollout controller, such as an Argo Rollouts AnalysisTemplate or Flagger MetricTemplate, which queries SLO-based SLIs at each traffic step and auto-aborts on threshold breaches.

Skip it for documentation-only changes and config-only changes with no runtime effect. Any change touching user-facing behavior, database schema, endpoints, or runtime dependencies requires the full artifact set.

The release is blocked. Only P0 or security fixes ship with explicit reliability approval until the budget recovers, following the Google SRE error-budget policy that prevents feature velocity from silently degrading reliability.