Skip to content
Get packs

Guide · Agent skill

How to use the Code Review and Release Check agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The check skill is a Waza agent skill that makes an AI coding agent review code changes, pull requests and release readiness in a structured way. It is written for developers and maintainers who want a second pass over a diff before merging or releasing. After this guide you will know what the skill does, how to install it, how to trigger it with plain requests, and what it needs to run safely.

Code Review and Release Checktw93/Waza · MIT

Quick start

  1. 1Download the ZIP. Download the check skill ZIP from this page.
  2. 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
  3. 3Check the requirements. Confirm Python 3, git, the GitHub CLI (gh) and repository access are available.
  4. 4Ask for a review in plain language. Describe the diff, PR or release you want checked, for example before a merge.
  5. 5Read the findings first. Review the concise findings, then the verification commands and their results.
  6. 6Authorize any write action explicitly. If you want a fix, commit, push or comment, name that action and goal in your request.

What is an agent skill and what does check make the agent do?

An agent skill is a folder of instructions and optional scripts that an AI agent loads when a task matches the skill's description. The check skill is scoped to review and release work: it reads the diff, the project's public docs, CI workflows and manifests, then reports concise findings first, followed by the verification it actually ran.

It can also triage issues and PRs, judge whether a change is worth a release, and run a project-wide code-quality audit that produces a scorecard. It is report-only by default and will not edit, commit, push, publish or comment unless you explicitly authorize that action.

  • Reviews diffs and pull requests before merging or releasing
  • Triages issues and PRs
  • Judges release worthiness and release readiness
  • Runs a project-wide code-quality audit with a scorecard
  • Does not debug root causes of a specific bug, and is not for prose review

Install the check skill

Download the ZIP from this page. Then choose one of two options: unzip it into your personal skills folder at ~/.claude/skills/, or unzip it into a project's .claude/skills/ folder so that only that project uses it.

In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

  • Personal install: unzip into ~/.claude/skills/
  • Project install: unzip into the project's .claude/skills/
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what you need before running it

The skill expects Python 3 for its bundled audit and incident ledger scripts, which use only the standard library. It also uses git and the GitHub CLI (gh) for pull request and remote-state inspection, and it needs access to the repository.

Publishing or commenting actions need your own credentials and your explicit authorization. The skill runs scripts and uses the network, so read the safety notes below before using it on a repository you have not reviewed.

  • Python 3 (standard library only)
  • git
  • GitHub CLI (gh)
  • Repository access
  • Your own credentials, only if you ask it to publish or comment

How do you trigger the skill with natural requests?

You do not need a special command. Ask in plain language and the agent matches your request to the skill. Requests such as reviewing a diff, triaging issues or PRs, asking whether a change is ready to ship, or requesting a project audit all fit.

The skill also recognises the /check command, and the alias code-review. The skill notes that /review is a separate built-in command for PR review, and that Waza uses /check instead.

  • "Review this diff before I merge."
  • "Look at the open PRs and triage them."
  • "Is this worth a release?"
  • "Run a project audit and give me a scorecard."
  • "/check" or "code-review"

What does a typical review look like?

When you ask for a review, the agent first reads the current worktree state. It then reads the diff and the project context from public sources such as manifests, CI workflows and release docs. The report lists findings first, then the verification commands and their results from that session.

For release or publishing work, the agent closes with a numbered completion ledger marking each item as done, not applicable or remaining, so you can see what is still open.

  • Findings are listed first, ordered by importance
  • Verification shows the commands actually run and their output
  • Multi-step or ship-action runs end with a numbered done / not applicable / remaining ledger
git status --short --branch -uall

What are the safety notes for scripts and network use?

Audit mode runs bundled Python scripts (scripts/audit_signals.py and scripts/incident_ledger.py) that read your project tree. Only use it on repositories whose scripts you have reviewed. The skill also uses the network for remote state, such as pull request data through gh.

The skill protects your uncommitted work. It will not stash, switch branches, clean, reset or check out a PR without explicit approval in the current turn. It also does not move untracked or generated files into /tmp.

  • Read the bundled scripts before running audit mode on an unfamiliar repository
  • Publishing and commenting need your credentials and explicit authorization
  • Authorization for a write action covers only the action and goal you named
  • Approval of a draft is not approval to commit, push, publish or delete

Tips for getting useful results

Name the scope you want reviewed, such as one diff, one PR number or the whole backlog. Say explicitly if you want the agent to make a change, because by default it only reports. Keep your worktree in a known state so the review reflects what you intend to ship.

Remember that the agent treats the current diff, CI and remote state as more reliable than any remembered context.

  • Give a PR number or diff range to keep the scope narrow
  • Authorize writes explicitly, naming the action and the goal
  • Expect a report, not edits, unless you ask for fixes

What people use it for

  • Getting a structured second pass over a diff before merging
  • Triaging a backlog of issues or pull requests across a repository
  • Deciding whether a set of changes is worth a release
  • Checking release readiness against real CI and package state
  • Running a project-wide code-quality audit that produces a scorecard

Frequently asked questions

How do I install the check agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills instead.

What do I need to run the check skill?

You need Python 3 for the bundled scripts, git, the GitHub CLI (gh) for pull request and remote-state inspection, and repository access. Publishing or commenting also needs your own credentials and explicit authorization.

Is the check skill free?

The source material provided for this page does not state pricing terms. The skill is listed under the MIT licence, which is stated on this page, and you should check the licence text in the original repository for the details.

Will the check skill change my code or publish anything?

Not by default. It is report-only and will not edit, commit, push, publish or comment unless you explicitly authorize that action in your request.

Is check the same as the built-in /review command?

No. The skill notes that /review is a built-in Anthropic plugin command for PR review, and that Waza uses /check, or the alias code-review, instead. The skill also says not to re-trigger /review from within itself.

Is it safe to run on any repository?

Only run audit mode on repositories whose scripts you have reviewed. Audit mode runs bundled Python scripts that read your project tree, and the skill uses the network for remote state.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use check if you want a structured, report-first review of diffs, a clear triage of issues and PRs, or a release checklist tied to real CI and package state. Pick something else if you need help debugging the root cause of a specific bug or reviewing prose and documents, since this skill is scoped away from both.

Download the skill