Release Operator

Release Operator

A release is not done when the commit is made. It is done when a fresh consumer can install/update and the remote host agrees.

Third-order stance: a release is a trust event with an exit door. Verify not just the happy path, but leakage, rollback, stale consumers, and the support email this version would generate if it went sideways.

When to Use

• The user says release, deploy, publish, cpd, or "do that release"
• A packaged CLI/app needs version bump, changelog, tag, and verification
• A remote host such as vesta should validate the install/update path

When NOT to Use

• Pure local prototype work with no publish surface
• The repo has its own release automation and the user only asked for a small code fix
• There are unreviewed failing tests that need debugging first

---

Phase 1: Establish Release Facts

Read the repo instructions and existing release docs/scripts. Determine:

• current version and target version
• package manager / build tool
• release target: npm, GitHub release, app notarization, static deploy, etc.
• required changelog format
• whether install/update must be verified on a remote host

Run:

git status --short
git tag --list --sort=-version:refname | head

Phase 2: Quality Gates

Run the repo's standard checks. For Node CLIs, default to:

npm run check
npm pack --dry-run

For packaged CLIs, inspect the dry-run tarball contents and confirm support scripts advertised by doctor/live checks are included.

Run a leak audit against the artifact list: no .env, tokens, local logs, private notes, nested .git, test-only fixtures advertised as live features, or generated files that encode local paths.

Phase 3: Version and Changelog

• Move changelog entries from Unreleased to the target version/date.
• Bump package metadata and lockfiles together.
• Keep changelog user-facing, not just commit-shaped. Include changed trust semantics explicitly: install path, migration, telemetry/logging, permissions, reversibility, and support impact.

Phase 4: Commit, Tag, Push, Publish

Use the repo's release script if present. Otherwise:

git add 
git commit -m "Release vX.Y.Z"
git tag vX.Y.Z
git push
git push origin vX.Y.Z

Then publish using the configured channel. Do not invent a publish command; inspect docs/scripts first.

Phase 5: Post-release Verification

Verify the published artifact, not just the local checkout:

 --version
 doctor --smoke
 update

If a remote host was named:

ssh vesta ' update;  --version;  doctor --smoke'

Classify any warnings. Release blockers are install failure, wrong version, missing packaged files, broken smoke test, or stale remote config pointing at the wrong binary.

Do a rollback drill when feasible: confirm the previous version/tag can be reinstalled or the deployment can be rolled back. A release is only safe if the exit door opens.

Phase 6: Report

Return concise evidence:

• version/tag published
• checks run
• package/publish verification
• local install/update result
• remote install/update result
• rollback/exit-door evidence, or why it was not applicable
• any follow-up issues filed

• Use remote-host-verifier for local/remote command comparison.
• Use status-copy-trust-audit before release when CLI status text changed.
• Use changelog-writer when changelog wording needs user-facing translation.
• Use support-storm or status-copy-trust-audit when the release changes setup, billing, permissions, status output, or failure modes.
• Use remote-host-verifier after release to catch stale package sources and hidden host drift.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: carlkibler
• Source: carlkibler/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.