Actions Template Repository

A batteries-included template for scaffolding new GitHub Actions.

A starting point for ESM, bundled GitHub Actions — ships with action.yml, an example handler, Jest (with manual @actions/* mocks), ESLint/Prettier, Changesets, Renovate, and issue/PR templates, plus a one-shot setup CLI that wires it all to your action.

Use this template

1. Click Use this template on GitHub and create your repository.
2. yarn install
3. yarn rename — an interactive CLI that collects the action name, description, owner, author, and your input/output variables, then rewrites action.yml, package.json, the README, and everything else before removing itself (the entire scripts/ directory).
   • Pass --dry-run to preview every change without writing anything.
   • Supply --name, --description, --owner, and/or --author to skip the matching prompts.
4. Edit src/main.js to implement your action's logic. yarn build bundles src/ locally so you can smoke-test it, but the bundle is never committed — CI only confirms it still compiles (see Architecture below).
5. Run yarn changeset to record any release-worthy change, then commit and push — Changesets handles versioning, the changelog, and tagging the release. The release workflow builds dist/ and commits it only to the release tag, so main stays source-only.

What the setup does

• Replaces the {{ ACTION_NAME }} / {{ ACTION_DESCRIPTION }} / {{ ACTION_AUTHOR }} / {{ ACTION_AUTHOR_EMAIL }} / {{ OWNER }} placeholders across the repository.
• Generates the action.yml inputs/outputs and the README tables from the variables you define.
• Installs the action-only package.json, jest.config.js, and README.md from scripts/template/ over the repository's own copies.
• Detects sensible defaults: the owner from the git remote, and the author from gh, the GitHub Actions context, or your commit email.

Architecture

The scaffolded action separates hand-written source from the generated bundle, and never commits the bundle to main:

src/
  index.js       # entry shell: reads the `token` input, builds the octokit
                  # client, calls main(), and maps a thrown error to
                  # core.setFailed()
  main.js         # the action's logic
  main.test.js    # jest coverage for main.js
dist/
  index.mjs       # generated by `yarn build` (@vercel/ncc) — the
                   # dependency-free bundle action.yml actually runs.
                   # Gitignored: built fresh at release time, not committed.

GitHub runs uses: owner/repo@vX by checking out that ref and executing dist/index.mjs directly, with no install step — so every consumer needs a dependency-free bundle at that ref. Rather than committing dist/ to main (which would put generated, node_modules-derived code in every PR diff and dependency bump), the generated repo's release workflow builds dist/ only when it publishes a release, commits it to a throwaway commit reachable solely from the release tag (vX.Y.Z and the floating vX), and pushes just that tag — main never sees it. The same bundle also ships in the npm tarball via package.json's files allowlist, which applies independently of .gitignore.

PR CI (linting.yml) still runs yarn build on every PR so a broken bundle is caught before it can reach a release, without needing to diff a committed artifact.

The bundle is ESM (dist/index.mjs, built with @vercel/ncc) because the package is "type": "module" — the entry source and configs are ESM, and ncc's ESM output correctly wires createRequire for the Node built-ins the bundled dependencies need. The .mjs extension matters: it forces ESM regardless of the nearest package.json's "type" field, so the bundle doesn't depend on (or get invalidated by) ncc's auxiliary dist/package.json, which the build step deletes.

Development

yarn format    # eslint
yarn lint      # eslint
yarn test      # jest
yarn coverage  # jest with the 80% coverage threshold

Built and maintained by Allons-y Studio — a US-based studio specializing in design systems, front-end architecture, and accessibility.