pre-release（预发布）

pre-release — Release Readiness & Changeset Generation

A structured pre-release workflow that runs automated checks, generates user-facing changesets from git history, and optionally spawns fresh-eyes README reviewers.

When to Use

• Before any npm publish or version bump
• When you need to generate CHANGELOG entries from recent work
• Before merging a release PR
• Any time you want a release-readiness report

Prerequisites

The target project must have @changesets/cli initialized:

npm install --save-dev @changesets/cli @changesets/changelog-github
npx changeset init

See Changesets Setup for first-time configuration.

The Workflow

Step 1: Pre-Flight Checks

Run these checks against the project root. Report each as ✅ / ❌ / ⚠️:

Blockers (must fix before release): checks 1-5, 8, 11-18, 20 (if applicable). Warnings (should fix): checks 6-7, 9-10, 19. Info: anything else notable.

Step 2: Generate Changesets

This is the core value — AI-written, user-facing release notes instead of mechanical commit scraping.

2a. Find the range

# Last release tag
LAST_TAG=$(git describe --tags --abbrev=0 2>/dev/null || echo "")

# If no tags, use first commit
if [ -z "$LAST_TAG" ]; then
  RANGE="$(git rev-list --max-parents=0 HEAD)..HEAD"
  echo "No previous tags found — covering entire history"
else
  RANGE="${LAST_TAG}..HEAD"
  echo "Changes since $LAST_TAG"
fi

2b. Gather the raw material

# Commits with files changed
git log $RANGE --pretty=format:'%h %s' --no-merges

# For more context on what changed
git log $RANGE --pretty=format:'### %h %s%n%b' --no-merges

# Files changed (to understand scope)
git diff --stat $LAST_TAG..HEAD 2>/dev/null || git diff --stat $(git rev-list --max-parents=0 HEAD)..HEAD

Also check for existing pending changesets — don't duplicate:

ls .changeset/*.md 2>/dev/null | grep -v README.md

If there are already changeset files, read them and account for what's already covered.

2c. Classify and write changesets

Group commits by impact and write changeset files. Each changeset is a markdown file in .changeset/:

File format (.changeset/<descriptive-name>.md):

---
"package-name": patch
---

Brief, user-facing description of what changed.

Semver classification:

Writing guidelines:

• Write for users, not developers. "Added --verbose flag" not "refactored logger module"
• Group related commits into a single changeset when they're part of one logical change
• Use present tense: "Add", "Fix", "Remove", not "Added", "Fixed", "Removed"
• If a commit is purely internal (CI, refactor with no behavior change), it can be omitted or grouped under a generic "Internal improvements" patch
• One changeset per logical change, not per commit

Name the files descriptively using kebab-case: add-verbose-flag.md, fix-auth-crash.md, initial-release.md.

2d. For initial releases (no previous tags)

Write a single changeset covering the initial release:

---
"package-name": minor
---

Initial release.

- Feature 1: brief description
- Feature 2: brief description
- Feature 3: brief description

Use minor (0.1.0) for initial releases unless the project is already at 1.x.

Step 3: Fresh-Eyes README Review (Optional)

Load the run-agents skill. Use pi-subagents to spawn 2 parallel agents on different models to review the README as first-time users:

{ "tasks": [
    { "agent": "_arch-reviewer", "task": "Read the README.md in <project-path>. You are a developer who has NEVER seen this project. Can you answer: (1) What does it do? (2) How to install? (3) How to use? (4) Prerequisites? (5) Where to get help? For each: quote relevant text or say MISSING. Then list anything confusing or that assumes prior knowledge.", "model": "google/gemini-3-pro" },
    { "agent": "_arch-reviewer", "task": "Read the README.md in <project-path>. You are a developer who has NEVER seen this project. Can you answer: (1) What does it do? (2) How to install? (3) How to use? (4) Prerequisites? (5) Where to get help? For each: quote relevant text or say MISSING. Then list anything confusing or that assumes prior knowledge.", "model": "github-copilot/gpt-5.3" }
]}

Read both outputs and note any gaps.

Step 4: Report

Present the final report in this structure:

# Pre-Release Report: <package-name>

## Version: <current> → <proposed>
## Date: <today>

### Checklist
| # | Check | Status | Notes |
|---|-------|--------|-------|
| 1 | README | ✅/❌ | ... |
| ... |

### Changesets Generated
| File | Bump | Summary |
|------|------|---------|
| `add-feature-x.md` | minor | Add feature X |
| ... |

### README Review
(If Step 3 was run)
- **Gaps**: ...
- **Confusing**: ...

### Blockers (must fix)
1. ...

### Suggestions (can wait)
1. ...

### Ready to Release?
**YES** / **NO — N blockers remain**

Step 5: Commit (if approved)

If the user approves, commit the changeset files:

git add .changeset/*.md
git commit -m "chore: add changesets for next release" -m "Co-Authored-By: Pi <noreply@pi.dev>"

Changesets Setup

For projects that don't have changesets yet. Run once:

1. Install

npm install --save-dev @changesets/cli @changesets/changelog-github
npx changeset init

2. Configure .changeset/config.json

{
  "$schema": "https://unpkg.com/@changesets/config@3.1.2/schema.json",
  "changelog": "@changesets/changelog-github",
  "commit": false,
  "fixed": [],
  "linked": [],
  "access": "public",
  "baseBranch": "main",
  "updateInternalDependencies": "patch",
  "ignore": []
}

Set "access": "public" for scoped packages (@scope/name). Set "baseBranch" to your default branch.

3. Add npm scripts to package.json

{
  "scripts": {
    "changeset": "changeset",
    "version-packages": "changeset version",
    "release": "changeset publish"
  }
}

4. GitHub Actions workflow (.github/workflows/release.yml)

name: Release

on:
  push:
    branches: [main]

concurrency: ${{ github.workflow }}-${{ github.ref }}

permissions:
  contents: write
  pull-requests: write
  id-token: write  # Required for npm Trusted Publishing (OIDC)

jobs:
  release:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 24  # npm >= 11.5.1 required for Trusted Publishing
          cache: npm
          registry-url: https://registry.npmjs.org

      - run: npm ci
      - run: npm run build
      - run: npm test

      - name: Create Release PR or Publish
        id: changesets
        uses: changesets/action@v1
        with:
          publish: npx changeset publish
          title: "chore: version packages"
          commit: "chore: version packages"
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

5. Configure Trusted Publishing (recommended — no tokens needed)

npm's Trusted Publishing uses OIDC to authenticate GitHub Actions with the npm registry. No NPM_TOKEN secret required.

1. Go to your package on npmjs.com → Settings → Trusted Publisher
2. Select GitHub Actions and configure:

- Organization or user: your GitHub username - Repository: your repo name - Workflow filename: release.yml

1. That's it. The id-token: write permission in the workflow enables OIDC. Provenance attestations are generated automatically.

After verifying Trusted Publishing works, go to package Settings → Publishing access → "Require two-factor authentication and disallow tokens" for maximum security.

⚠️ NPM_TOKEN is deprecated: Do not use NPM_TOKEN secrets for publishing from GitHub Actions. Trusted Publishers (OIDC) is the only supported method. If you encounter an existing workflow using NPM_TOKEN, migrate it to Trusted Publishers. See: https://docs.npmjs.com/trusted-publishers

Tips

• Run this skill early and often, not just before release. The checklist catches issues faster when run during development.
• Edit generated changesets freely. They're just markdown files — tweak wording, merge entries, split large ones.
• One changeset per PR is a good rhythm. The pre-release skill can also be run to generate changesets for a single PR's worth of work.
• For monorepos, changesets handles multi-package versioning natively. List multiple packages in the frontmatter.