Ship It

Replaces "merged the PR, now what?" with one workflow that gets the change live and verified. Pairs with merge-confidently (which ends at merge) — ship-it starts where that ends.

Auto-invocation triggers

• User says "ship to prod", "release this", "deploy to production", "cut a release"
• After merge-confidently returns successful merge if release/deploy is implied
• Scheduled releases (weekly/biweekly cadence)

Workflow

Phase 1 — Version + changelog (always)

• Invoke version-bump to determine semver bump (patch/minor/major from commit history)
• Invoke changelog-update to promote [Unreleased] to a versioned section
• Commit version + changelog, push to main

Phase 2 — Tag + GitHub release (always)

• Invoke ship to:
  • Create the version tag
  • Cut a GitHub release with changelog excerpt
  • Verify the tag pushed cleanly

Phase 3 — Deploy (always — pick the right deployer)

Detect deployment target from project:

• Vercel (vercel.json, Next.js): invoke vercel-deploy
• Cloudflare Workers/Pages (wrangler.toml): invoke cloudflare-deploy
• Docker on remote server: invoke prod-rebuild
• Generic CI/CD: invoke deployment-automation
• Multi-repo coordinated: invoke chain-release

Phase 4 — Post-deploy verify (always)

• Wait 60s for deploy to settle
• Invoke sentry to check for new issue events post-deploy
• Invoke ci-watch to verify any post-deploy smoke checks passed
• Hit a health endpoint if known (curl /health, /version to confirm new version is live)

If Sentry shows any new issue with frequency >0 in the post-deploy window, escalate to production-incident composite.

Phase 5 — Capture (conditional)

• If release contains breaking changes: invoke adr-write to record migration notes
• If release is significant (minor/major): invoke knowledge-loop to save the shipping summary for future reference

Reconciliation

SHIP IT — <repo> v<old> → v<new>
  Version:     <bump type>, commits since last tag: N
  Changelog:   M entries promoted from Unreleased
  Tag:         v<new> pushed
  Deploy:      <target>, took Xs
  Verify:      sentry clean, /health ok, version endpoint shows v<new>
  Captured:    ADR-NNNN (if breaking)

Outputs / Evidence

• New version + tag
• Changelog entries published
• Deploy target + URL
• Post-deploy verification proof (Sentry clean, health endpoint live)
• ADR if applicable

Failure / Stop Conditions

• Phase 1 reveals uncommitted changes → stop, commit first
• Phase 2 fails (tag conflict) → stop, investigate
• Phase 3 deploy fails → stop, surface error, do NOT auto-rollback (that's production-incident territory)
• Phase 4 finds new Sentry issues → STOP, escalate to production-incident
• Never use --force-with-lease or --admin to push past gates

Memory Hooks

• Read deployment history per-repo to detect cadence patterns (e.g., "Lucky ships Tuesdays")
• Write release outcome to memory for trend tracking (deploy time, issue count post-deploy, rollback rate)