glebis/release
>- Cut a software release and maintain a tiered compatibility policy. Use when the user wants to release, ship a version, bump the version, tag a release, write a changelog, or update COMPATIBILITY. Config-driven via release.config.json; bumps version files, runs a readiness gate, updates COMPATIBILITY.md tiers and deprecations, tags (→ release workflow), and reports closed issues. Teaches the underlying standards as it runs.
npx skills add https://github.com/glebis/claude-skills --skill release
For the Cull repository, use $cull-release; Cull's repository-enforced state,
artifact, updater, and Homebrew gates are stricter than this generic workflow.
A config-driven release orchestrator. Mechanics live in scripts/release.py
(unit-tested, stdlib-only); this file is the workflow you (or an agent) follow.
Read reference/config-schema.md for release.config.json and
reference/standards.md for *why* each step exists.
> Scope honesty: reference-tested on a Tauri + Rust + SvelteKit repo (Cull).
> Other stacks are supported *by config*, not yet validated. Treat first runs on
> a new stack as a dry run (see "Dry run" below) until you trust it.
The user says "release", "ship it", "cut a version", "bump version", "tag a
release", "update the changelog/COMPATIBILITY". Requires a release.config.json
at the repo root (scaffold from templates/release.config.json.tmpl).
A version only means something once you declare *what you promise to keep
working*. release.config.json → surfaces[] is that declaration; each surface
has a tier (experimental → preview → stable) and a compatibility mode.
Only stable surfaces carry the promise. Breaking a stable surface forces a
major bump — the engine enforces this. (Standards: see reference/standards.md.)
/release <patch|minor|major> — run the steps below. On an unfamiliar repo, do a
dry run first (see below). When the user asks to "explain", expand each step's
*why* into a short lesson from reference/standards.md.
releaseBranch (default main, inworktree if set) is checked out, clean, and synced with origin. Abort
clearly otherwise. — *why: a release tag must point at a known-good, pushed tree.*
python3 scripts/release.py --config <cfg> plan <kind> prints thenew version + tag. It asserts the version files currently agree. Show old → new.
— *why: SemVer math; 0.x lets minors break (pre-1.0).* [[Semantic Versioning (SemVer)]]
cfg.gate then each cfg.extraGate[]. All must exit 0(fmt, clippy, tests, license audit, prod build, golden contract tests). Block on
failure. — *why: this is a Production-Readiness Review.* [[Production Readiness Review]]
(git log <lastTag>..HEAD --format=%s) and draft a section — the engine's
draft_changelog buckets them into Added/Changed/Fixed (Keep a Changelog).
Insert under the top of CHANGELOG.md; hand-curate the user-facing lines.
— *why: humans read changelogs; conventional commits seed them.* [[Keep a Changelog]]
cfg.compatibility.path (COMPATIBILITY.md). Ask:stable surface? If yes, the required bump ismajor — re-run with major or the release is invalid. (Enforced by
enforce_bump.) Stamp "Last updated: <new version> (<date>)".
— *why: tiers + deprecation windows are how you evolve without lying.* [[Kubernetes API Deprecation Policy]]
python3 scripts/release.py --config <cfg> bump <kind> writesevery version file; refresh cfg.lockfiles (e.g. cargo update -p <crate> or a
build). Commit chore(release): v<new> including CHANGELOG + COMPATIBILITY.
git tag v<new> and push the tag (→ the repo's releaseworkflow) and the branch. Confirm the tag trigger exists *before the first
release* (grep -A3 '^on:' .github/workflows/*.yml).
last tag (if cfg.issueTracker is set, e.g. bd) — those are the release notes.
There is no --dry-run flag — a dry run is steps 1–5 done *without mutating*:
run python3 scripts/release.py --config <cfg> plan <kind> (pure: prints the
version/tag, writes nothing) and optionally run cfg.gate to check readiness.
Do NOT run bump, commit, or tag. The bump subcommand is the only engine
command that writes (version files only); commit/tag/push are git steps you take
in step 6–7, never the engine.
plan → run gate → edit CHANGELOG + COMPATIBILITY → bump → commit → tag → push.
The engine is just scripts/release.py; everything else is git.
The readiness gate runs *golden/contract tests* (cfg.extraGate). Start with one
(a DB round-trip), then add export and API contract tests. See the consuming
repo's docs/CONTRACTS.md and reference/standards.md. [[Pact — Consumer-Driven Contract Testing]]
Take glebis/release from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.