agentleFS
mkolewmkolew/ai/skills/reviewer/reviewer-tooling/SKILL.md

reviewer-tooling

Skill0 starsChanged 57 days agoMITView on GitHub

Review project tooling: GitHub Actions workflows, package.json and pnpm workspace files, .gitignore, .gitattributes, agent and editor config (.claude, launch.json), Makefiles and Dockerfiles. Use when reviewing changes to these files, or when the user types /reviewer-tooling. Catches script injection and leaked secrets in CI, over-broad permissions and tokens, deploys that run on pull requests, unpinned or deprecated actions and runtimes, lockfile drift, ignored files that must be committed (or committed files that must be ignored), CRLF and LFS mistakes, and machine-specific config. Part of the reviewer bundle.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read each workflow end to end: triggers, permissions, jobs, the secrets used, and what is deployed where.

    line 13
  2. Very specific:

    No deprecated runtimes or actions (warnings in recent runs).

    line 33
  3. Distinctive:

    No machine-specific absolute paths or usernames in shared files.

    line 50
  4. Distinctive:

    No secrets, and no permission rules wider than the project needs.

    line 52
  5. Very specific:

    Every workflow step has a name: that says what it does.

    line 62
  6. Very specific:

    Use the reviewer format: [severity] path:line — problem, then failure: (the concrete input, state or platform → wrong result) and fix: (the …

    line 72

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Tooling reviewer

Scope: .github/workflows/*, package.json, pnpm-workspace.yaml, .npmrc, .gitignore, .gitattributes, .claude/**, .vscode/**, Makefile, Dockerfile. Skip lockfiles, except to check that they moved with package.json.

Before reviewing

  • Read each workflow end to end: triggers, permissions, jobs, the secrets used, and what is deployed where.
  • Validate YAML (actionlint if installed, otherwise a YAML parser). Run git check-ignore -v <path> on files whose ignore status matters.

Checklist

CI security

  • ${{ github.event.* }} values (PR titles, branch names, issue bodies) interpolated into run: let anyone run code. Pass them through env: and quote them.
  • pull_request_target and workflow_run together with a checkout of PR code give that code your secrets.
  • permissions: is least privilege (contents: read by default). Grant writes per job.
  • Secrets are exposed only to the steps that need them, never echoed, and unavailable to fork PRs.
  • Tokens are scoped to what the job does. Cloudflare: an API token from a template with only the needed permissions, never the Global API Key.
  • Third-party actions are pinned to a release or commit SHA, never a branch.

CI behaviour

  • Deploys run only on push to the release branch (or a manual trigger), never on pull requests.
  • A concurrency group stops two deploys overlapping.
  • Tests run before deploy, and a failing step fails the job. No || true or continue-on-error on checks.
  • Tool versions are pinned (Node, Python, Godot or Unity, pnpm). Caches are keyed on those versions and the lockfile hash.
  • No deprecated runtimes or actions (warnings in recent runs). Setup steps come before the actions that need them (pnpm before an action that runs pnpm).
  • Checkout fetches LFS when the build needs LFS files.

Packages

  • packageManager matches the lockfile's tool and version. The lockfile changes whenever dependencies change.
  • Runtime dependencies and devDependencies are in the right section. Apps set "private": true.
  • scripts work on every OS the team and CI use.

Git config

  • .gitignore covers build output, dependency folders, engine caches (.godot/, Library/), and secrets (.env*, .dev.vars, keystores, credentials*).
  • It must not ignore files that have to be committed (Godot .uid/.import, Unity .meta). Negation patterns are needed when a parent folder is ignored.
  • .gitattributes: eol=lf for scripts (CRLF breaks bash). Binaries are marked binary. LFS patterns match where the large files actually live.

Agent and editor config

  • No machine-specific absolute paths or usernames in shared files.
  • Commands named in config (launch.json, hooks) exist and run.
  • No secrets, and no permission rules wider than the project needs.

Containers and Make

  • Base images pinned. No secrets in layers. Make targets are .PHONY where needed, and fail on error.

Readability

Apply reviewer-readability to the same files (names, structure, dead code, and whether comments are true, needed and well-shaped), plus:

  • Every workflow step has a name: that says what it does. A non-obvious condition, pin or cache key has a one-line # comment explaining why.
  • Scripts in package.json are named by what they do; long inline shell moves into a script file.

Not findings

  • Formatting of YAML or JSON.
  • Pinning first-party actions/* to a major version, when the project does so consistently.

Report

Use the reviewer format: [severity] path:line — problem, then failure: (the concrete input, state or platform → wrong result) and fix: (the smallest change). Most severe first. Prove each finding before reporting it, and say No findings with what you checked when there are none.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.