Aller au contenu principal
WORK

WORK-FLOW-RELEASE Agent

Complete workflow to prepare and publish a release.

Context

<arguments>

Objective

Execute the complete release cycle: branch, quality audit, changelog, semantic versioning, complete tests, production build, tag, deployment.

Workflow

  • BRANCH: Create release/vX.Y.Z branch from up-to-date main
  • AUDIT: Tests, lint, typecheck, npm audit, build (all must pass)
  • CHANGELOG: List changes since the last tag (Added, Changed, Fixed, Deprecated, Removed, Security)
  • VERSION: Semantic Versioning (breaking = MAJOR, features = MINOR, fixes = PATCH)
  • TESTS: Complete validation (unit, integration, E2E, manual)
  • BUILD: Production build, verify bundle size and assets
  • TAG: Annotated tag git tag -a vX.Y.Z, push, GitHub release with notes
  • DEPLOY: Production deployment with rollback plan ready

Performance notes

  • Use the project's parallel test runner when one exists. On claude-base specifically, bash scripts/test.sh runs ~4.5x faster than bats tests/*.bats directly (uses GNU parallel + 8 jobs; ~1 min for 455 tests vs ~4-5 min sequential). For other projects, prefer npm test / pytest -n auto / cargo test --jobs over single-threaded invocations.
  • Run the test suite ONCE per release, after all version bumps + CHANGELOG + generators are done. Doc-only changes (CHANGELOG, version banners, generated files) cannot break the test suite by construction; re-running tests after these changes wastes time.

Expected output

  1. Audit: Quality go/no-go report
  2. Changelog: CHANGELOG.md updated
  3. Release: Tag + release notes on GitHub
  4. Deploy: Application deployed, monitoring OK
AgentUsage
/qa:qa-auditQuality audit
/doc:doc-changelogChangelog
/dev:dev-testComplete tests
/ops:ops-releaseSimplified alternative
/ops:ops-monitoringPost-deployment

IMPORTANT: Never skip tests before a release.

YOU MUST have a rollback plan ready before deploying.

NEVER deploy on a Friday evening (except critical hotfix).

Think hard about the impact of each change on users.


See also