跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
部署与云服务

ship-release

Cut a new semver release for this project — analyze commits since the last tag, recommend a major/minor/patch bump, draft English release notes with AI, get user confirmation, then bump `package.json`, commit, create the GitHub release with `gh release create`, and let the existing `.github/workflows/release.yml` pipeline handle Docker + npm publishing. Use this skill whenever the user says "发个 release", "发新版本", "ship a release", "bump version", "cut v1.2.0", "tag a release", "/ship-release", or otherwise asks to publish a new version. Also use to preview what the next release would contain ("what's in the next release?", "show release diff since last tag").

camtrik · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add camtrik/agent-trail --skill ship-release

技能说明


name: ship-release description: Cut a new semver release for this project — analyze commits since the last tag, recommend a major/minor/patch bump, draft English release notes with AI, get user confirmation, then bump package.json, commit, create the GitHub release with gh release create, and let the existing .github/workflows/release.yml pipeline handle Docker + npm publishing. Use this skill whenever the user says "发个 release", "发新版本", "ship a release", "bump version", "cut v1.2.0", "tag a release", "/ship-release", or otherwise asks to publish a new version. Also use to preview what the next release would contain ("what's in the next release?", "show release diff since last tag").

Release Workflow

This project publishes via tag-triggered GitHub Actions. When a tag v* is pushed (or a release is created with gh release create), .github/workflows/release.yml runs Docker build → npm publish in parallel. Release notes are written by AI (this skill), not auto-generated by GitHub.

What this skill does and doesn't do

Does:

  • Reads commits since last tag, recommends semver bump per conventional commits
  • Drafts English release notes, grouped by type, with the user able to edit before publishing
  • Bumps package.json version, commits as chore(release): vX.Y.Z
  • Creates the GitHub Release via gh release create with the AI-written notes (this also creates and pushes the tag)
  • Reports the workflow run URL for the user to monitor

Doesn't:

  • Publish to npm directly — the GitHub Actions workflow does that on tag push
  • Build Docker images — same, workflow handles it
  • Create release notes by hand without AI input — the whole point is AI drafts them

Two modes

  • Preview — just analyze and report; no writes. Use when the user asks "what's in the next release?" or "show me what would go into a release". No version bump, no commit, no tag.
  • Publish — the full flow below. Default when the user asks to release / ship / bump version.

If unclear, ask: "Preview only, or actually publish?"


Pre-flight checks (always, both modes)

Run in parallel:

git rev-parse --show-toplevel       # confirm we're in the repo
git branch --show-current           # must be main
git status --porcelain              # must be empty (for publish mode)
git fetch origin main --tags        # sync tags and main
git describe --tags --abbrev=0      # find last tag (may fail if no tags)
gh auth status                      # gh CLI must be authenticated (publish mode)

Failure modes:

  • Not on main — in publish mode, refuse. In preview mode, warn but proceed.
  • Dirty working tree — in publish mode, refuse and show git status output. Don't auto-stash.
  • Behind origin/main — refuse and tell the user to git pull --ff-only.
  • No previous tag — that's the initial release case. Start the version at v1.0.0 unless package.json says otherwise; analyze all commits.
  • gh not installed or not authenticated — in publish mode, refuse with install instructions: brew install gh && gh auth login.

Step 1 — Analyze commits since last tag

git log <last-tag>..HEAD --pretty=format:'%H%x09%s%x09%b%x1e' --no-merges

Parse each commit as conventional-commits:

  • feat(scope)?: subject → category Features, signal minor
  • fix(scope)?: subject → category Bug Fixes, signal patch
  • refactor(scope)?: subject → category Refactor, signal patch
  • perf(scope)?: subject → category Performance, signal patch
  • docs(scope)?: subject → category Docs, signal patch
  • chore(scope)?: subject → category Chores, signal patch
  • test(scope)?: subject → category Tests, signal patch
  • build|ci(scope)?: subject → category Build/CI, signal patch
  • Non-conforming subject → category Other, signal patch

Breaking change detection (any of the following → signal major):

  • Subject contains ! before the colon, e.g. feat!: or feat(api)!:
  • Body or footer contains a line starting with BREAKING CHANGE: or BREAKING-CHANGE:

Step 2 — Recommend bump

Aggregate signals: major > minor > patch. The highest signal wins.

Compute next version from the last tag (strip leading v):

  • major — (X+1).0.0
  • minor — X.(Y+1).0
  • patch — X.Y.(Z+1)

If package.json version disagrees with the last tag, mention it; usually the tag is the source of truth.

Step 3 — Collect closed backlog items

Closed items can live in two places by the time you reach this step:

  1. Pre-archived in docs/backlog/_done/vX.Y.Z/*.md — the user (or the backlog skill) may have already moved items into the folder matching the new version before running ship-release. Check this folder first using the recommended next version from Step 2. If it exists and has .md files, those are the canonical "shipped" set for this release.
  2. Still active in docs/backlog/*.md with status: done — items finished since the last release but not yet archived.

Read both, then union them (dedupe by slug — pre-archived wins). The combined set is what goes into the "Closed items" section.

For each item, capture: title, type, slug (filename minus .md), and branch (for the compare link). Also remember whether it was pre-archived or still active — Step 7 only needs to move the still-active ones.

If both sources are empty, skip the "Closed items" section entirely.

Why check the version folder first: the user frequently pre-archives items via the backlog skill (commits like docs(backlog): reorganize done items into vX.Y.Z archive). If you only scan top-level, you'll publish notes with an empty "Closed items" section even though the work is sitting right there in _done/vX.Y.Z/.

Step 4 — Draft release notes

Backlog items are the primary source of truth for release content. Closed backlog items (from Step 3) map directly into Features / Bug Fixes / Refactor by their type f

阅读完整来源 ↗