| name | core-ship-launch-checklist |
| description | Use as the final gate before anything goes live — the pre-launch checklist across code/security/performance/a11y/infrastructure/docs, and the first-hour post-launch verification. |
Launch Checklist
The last thing checked before anything goes live, and the first hour after. Adapted from
Addy Osmani's shipping-and-launch (source).
Rollout mechanics live in core-ship-deployment-strategies; the instrumentation this
checklist verifies live in core-operate-observability.
Areas under consideration
Skill
Pre-launch: every section green
- Code — all tests pass; build clean with no warnings; lint + types pass; reviewed;
no unresolved TODOs or debug logging; error handling covers expected failure modes.
- Security — the pre-ship pass in
core-build-secure-coding is complete: no secrets in
code/history/logs/artifacts; audit clean of reachable critical/high issues or documented
exception; boundary validation on user-facing endpoints; authn and authz checked; new
permissions/data flows reviewed; security headers; auth rate limiting; CORS pinned to origins.
- Performance — Core Web Vitals in "Good"; no N+1 on critical paths; images
optimized; bundle within budget; indexes on hot queries; caching for static assets and
repeated queries.
- Accessibility — full keyboard navigation; screen-reader-conveyable structure; WCAG
AA contrast; correct focus management for modals/dynamic content; descriptive errors
associated with fields; clean axe/Lighthouse run.
- Infrastructure — production env vars set; migrations applied or staged; DNS/SSL;
CDN for static assets; logging and error reporting confirmed configured; a health-check
endpoint that responds.
- Documentation — README, API docs, ADRs, changelog, user-facing docs current.
Plus, from the deployment side: feature flag configured, rollback plan documented and
exercised, monitoring dashboards up, team notified. No one available to watch the first
hour — including "Friday afternoon" — is a reason not to ship yet.
Post-launch: the first hour
- Health endpoint returns 200.
- Error monitoring shows no new error types.
- Latency dashboards show no regression.
- Walk the critical user flow manually in production.
- Logs are flowing and readable.
- Rollback mechanism confirmed ready (dry-run if possible).
After the release window, compare production defects, performance, and real usage with the
claims the pre-launch checks were meant to support. Feed missed scenarios, weak evidence, and
misleading green signals back into core-setup-scoping-and-requirements and
core-verify-testing-strategy.
"It works in staging" is not evidence — production has different data, traffic, and edge
cases; the monitors are the evidence.
