AGENT SKILL

release

Use when cutting a release — tagging a version, pushing tags, publishing a GitHub release, or shipping this toolkit's plugins. Detects the repo's release machinery (AGENTS.md release section, changesets, claude plugin tag, plain git tag + gh release), always shows a dry-run plan first, and executes only on one confirm. Triggers include "cut a release", "release v1.2.3", "tag and publish", "ship a new version", "whats in the next release".

License 许可
MIT
最近更新
2026-11-03
Stars
★ 2
Forks
0

01 前言 Front Matter

name
release
description
Use when cutting a release — tagging a version, pushing tags, publishing a GitHub release, or shipping this toolkit's plugins. Detects the repo's release machinery (AGENTS.md release section, changesets, claude plugin tag, plain git tag + gh release), always shows a dry-run plan first, and executes only on one confirm. Triggers include "cut a release", "release v1.2.3", "tag and publish", "ship a new version", "whats in the next release".
when_to_use
Cutting the release itself. Not for deciding a plugin or skill's semver bump beforehand — that's toolkit-change-control; in jabworks/agentic-toolkit, run that gate first and release the version it picks. Not for everyday undo/recovery git work — that's git-operations.

02 资源文件 2

03 SKILL.md 正文

/release

One dry-run, one confirm, one release. Never a force-push, never a guess.

Usage

/release                 # detect machinery, propose the next version
/release v1.4.0          # explicit version
/release minor           # bump hint

When NOT to use

  • Committing the work → git-commit (it stops before push; this skill starts after the commits exist).
  • Undo / discard / stash / recovery → git-operations.
  • "Is this toolkit change ready to publish?" → toolkit-change-control gates; this skill executes.
  • Deploying to servers or infrastructure — out of scope.

Guards — all must pass before anything else

Each failure names its fix and ends the attempt. --force is banned. History rewrites are banned.

□ Working tree clean              → commit (git-commit) or park it (git-operations)
□ HEAD is an ancestor of main    → git merge-base --is-ancestor HEAD origin/main
                                    (merge first — tags on dead branches haunt forever)
□ Test gate green                 → AGENTS.md test/finalize command; node --test here
□ Version consistent              → tag must equal the manifest/package version
                                    (claude plugin tag validates this natively)
□ Tag doesn't already exist       → git tag -l <tag>; git ls-remote --tags origin
□ gh authenticated                → gh auth status

The Router — first match wins

  1. AGENTS.md release: / ## Release section → follow the project's own process verbatim. The project override always wins.
  2. Changesets (.changeset/ exists) → changesets owns versioning and the CHANGELOG — never hand-tag on top of it:
    • npx changeset status to see what's pending.
    • If a changesets GitHub Action manages "Version Packages" PRs, the release is merging that PR — guide the user there and stop.
    • Otherwise: npx changeset version → show the diff → confirm → npx changeset publish → git push --follow-tags.
  3. This toolkit (.claude-plugin/marketplace.json exists) — plugin releases are automated; do not hand-tag:
    • Merging to main runs .github/workflows/plugin-release.yml, which calls scripts/release-plugins.mjs --since <push-base> --execute: every plugin version introduced by that push gets its <name>--v<version> tag and a GitHub release, with notes from the commits that touched its dist tree. CI never backfills — a gap stays a gap until someone runs --initial.
    • The release is merging the PR. Confirm the version bump landed and stop.
    • Before committing a bump: node scripts/release-plugins.mjs to see the plan, and --write-changelog to regenerate CHANGELOG.md — a test fails when a shipped version has no entry.
    • Versions on an unmerged branch are held back by design; the script refuses to release a commit that is not on the default branch.
    • If a tag lands but its release does not (they are two calls, and the second can fail alone), --repair creates the missing releases. A tag with no release is otherwise stranded forever.
    • Hand-tagging remains the fallback if the workflow is broken: claude plugin tag dist/plugins/<name> validates plugin.json against the marketplace entry, then git push origin <tag> and gh release create <tag> --generate-notes --title "<name> v<version>".
    • Remind: installed copies refresh only on a version change — users must update the plugin.
  4. Generic GitHub project:
    • Propose the version from conventional commits since git describe --tags --abbrev=0 (feat → minor, fix → patch, BREAKING CHANGE → major).
    • Annotated tag vX.Y.Z → git push origin vX.Y.Z → gh release create vX.Y.Z --generate-notes.
    • npm publish never happens on this branch — only via branch 1 or 2.

GitLab/Bitbucket: no route yet — say so and stop rather than improvise.

Dry-Run Plan — always, before touching anything

## Release plan: <repo> — <vX.Y.Z | "merge Version Packages PR">
Machinery   AGENTS.md | changesets | toolkit | generic-github
Version     <current> → <proposed>   (N feats, M fixes since <last tag>)
Commits     <git log --oneline <last-tag>..HEAD>
Notes       <gh api repos/{owner}/{repo}/releases/generate-notes -f tag_name=<tag>>
Will run    1. <exact command>  2. …  3. …
Rollback    per step — references/rollback.md

Then ask once: "Execute this release? [y/n]" One yes runs the whole sequence. Any step failure → stop, report what completed and what's reversible, with the rollback command for the completed steps. Never retry-loop.

Evidence

Every step prints its verification (tag exists locally and on the remote, release URL, publish output). Final line: the release URL plus the rollback one-liner.

Common traps

Trap Reality
"Just force-push the fixed tag" Never. Delete + re-tag via the rollback path, or fix forward with the next patch version.
Hand-tagging in a changesets repo Changesets owns versions — you'll fork the changelog. Use the changesets route.
Releasing from a feature branch The ancestor-of-main guard exists because published tags on unmerged branches haunt forever.
Skipping the dry-run "because it's just a patch" The dry-run IS the safety, and it costs one screen.
Toolkit release without a version bump Installed caches keep serving the old copy.

Related skills

git-commit (gets the work committed — stops before push by design, hands off here), git-operations (undo/recovery, including tag-deletion mechanics), toolkit-change-control (publish-readiness gate for this repo — it gates, this executes), condux workflow CP-3 (routes here after everything is green).

Provenance and maintenance

Re-verify volatile claims with:

  • claude plugin tag --help — the toolkit tag+validate primitive
  • gh api repos/{owner}/{repo}/releases/generate-notes -f tag_name=<tag> — notes preview
  • npx changeset status — changesets detection behavior

Last generated: 2026-07-08 Known uncertainty:

  • The guard set is adopted from claude-sdlc-wizard's release post-mortems (ancestry + version-match); not yet exercised by a real failure here.
  • GitHub-only: gh paths assume github.com remotes; GitLab/Bitbucket routes are reserved, unbuilt.

04 同仓库更多 Skills

jabworks
jabworks/agentic-toolkit

orphan

暂无简介

★ 2
jabworks
jabworks/agentic-toolkit

no-blurb

暂无简介

★ 2
jabworks
jabworks/agentic-toolkit

workflow

Routes any dev task into the right execution tier (Small/Medium/Large) and carries condux's operating rules — proportional effort, lazy skill loading, soft gates, implement-yourself-by-default.

★ 2
jabworks
jabworks/agentic-toolkit

usage-trace

Checks and manages durable, local usage capture — ingest, health status, and the opt-in limit snapshot — for Claude Code and Codex. Capture itself runs automatically at session start.

★ 2
jabworks
jabworks/agentic-toolkit

toolkit-skill-standards

Use when writing or reviewing a SKILL.md for jabworks/agentic-toolkit — frontmatter budgets, the trigger contract ("Use when…" description or when_to_use field), progressive disclosure into references/, and keeping a new description out of existing skills' trigger space. Triggers include "review this skill description", "write the frontmatter", "does this collide with an existing skill", "where should this content live". Not for scaffold/sync mechanics; use toolkit-foundry.

★ 2
jabworks
jabworks/agentic-toolkit

toolkit-research-frontier

Use when planning improvement work on jabworks/agentic-toolkit itself — open problems (trigger-collision automation, YAML-strict frontmatter checks, docs-catalog enforcement), existing assets, the next three concrete steps, and the four-front library-health campaign with commands and success criteria. Triggers include "let's improve our toolkit", "let's improve this repo", "what should we improve next", "toolkit roadmap", "open problems", "run the health campaign".

★ 2