Check Skill from tw93 Waza: Review Diffs and PRs Before You Ship

Makes Claude review a diff or pull request against your project's own rules, triage issues, audit a repo and check release readiness.

Installs: 16.2K; on skills.sh
Stars: 7.5K; on GitHub
Files: 16; besides SKILL.md
Updated: Nov 2026; last repo commit

Check is one of the eight Waza skills by Tw93. It makes Claude read your diff or PR, learn your project from its README, manifests and CI, and report findings with proof. Other modes cover issue and PR triage, release readiness and publishing follow-through, and a project-wide audit scorecard.

Review, audit and triage are report-only. Claude edits files, commits, pushes, publishes, comments or closes issues only when your current request allows that action. Check is not for debugging root causes or for prose.

Good for

Developers who want a careful review before they merge
Maintainers triaging issues and pull requests across repos
Anyone preparing a release who wants the gates checked with evidence

Install check

Claude Code

Copy this prompt.

Install the check skill in Claude Code. Run this command:

/plugin marketplace add tw93/Waza
/plugin install waza@waza

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

check works on its own. SKILL.md only mentions /write, /hunt and /think as optional hand-offs. The whole-set routes install all 8 skills (think, ui, check, hunt, write, learn, read, health). Plugin skills are namespaced, for example /waza:check. Pick one route, not both. The per-skill plugin line needs the marketplace added first (the first command in the all-skills plugin route).

From the project README. Source.

What's in the skill

When it loads

Reviews diffs, PRs, release readiness, and publishing follow-through. Use when asked to review, triage issues or PRs, or ship. Not for debugging root causes or prose.

Sections in SKILL.md

Outcome Contract
Durable Context Preflight
Worktree Safety Preflight
Mode Picker
Project Context Extraction
Plan Execution Mode
Get the Diff
Scope
Did We Build What Was Asked?
Question the Approach, Not Just the Diff
Pattern-Fix Completeness
Command And Install Surfaces
Hard Rules
Hard Stops (fix before merging)
Finding Quality Gate
Knowledge Sync

Key facts

Author: Tw93 (Waza)
License: MIT
Installs on skills.sh: 16,156 (Nov 2, 2026)
GitHub stars: 7,488 for the whole tw93/waza repo (Nov 2, 2026)
Last commit: Nov 1, 2026
What is inside: 17 files including SKILL.md: 2 reviewer personas, 10 reference guides and 4 scripts (3 Python, 1 shell).
Part of: Waza, a set of 8 skills: think, ui, check, hunt, write, learn, read, health
Works in: Claude Code, Codex, Cursor and other agents, Claude Desktop (ZIP) and Pi (Waza README)

What it includes

Reads the diff, sets a depth and checks it against the goal and project rules
Quick, Standard or Deep depth by lines and files. It checks scope drift, hard stops such as destructive auto-execution and injection, and ends with a sign-off.
Try: “/check review my uncommitted changes before I merge”
Runs code · Default review
Implements a /think plan and then runs the project check
Starts when your message begins with "Implement the following plan" or links a /think output.
Runs code · Plan execution
Looks through issues and PRs, including across repos
Loaded from references/mode-triage.md when you ask to look at issues, review PRs or triage.
Network · Triage mode
Commit, push, publish, close issues and release reactions with gates
Loaded from references/mode-ship.md. The release gate uses scripts/release_gate.py, which the script says is read-only with no network.
Runs code · Ship and release mode
Project-wide scorecard using scripts/audit_signals.py
For "audit", "scorecard" or "项目体检". The script walks the project and emits PASS, WARN or FAIL blocks.
Runs code · Audit mode
Persona agents for architecture and security on larger diffs
Standard and Deep reviews add specialists from references/persona-catalog.md, and Deep adds an adversarial pass.
Read · Specialist reviewers
Checks whether review prompts still catch past bugs
scripts/incident_ledger.py validates a table of past defects and prints a replay plan.
Runs code · Incident ledger replay
scripts/run-tests.sh detects the project type and runs its checks and tests
Used only when the project does not name its own verification command. It runs cargo check and cargo test, npx tsc --noEmit and npm test, npm test, make test or pytest, whichever matches the files it finds.
Runs code · Verification

From the official docs

Permissions

Medium risk

Check reads your repo and can run your tests, and with your approval it can commit, push, publish, comment on issues or close them. Say "review only" when you only want findings.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Ship mode can push, publish and comment
SKILL.md lists commit, push, publish, release, close issue and release reactions as actions. Each needs authorization from your request. Do not give a standing "ship it" unless you mean it. github.com
It runs your project's tests
run-tests.sh picks one command from the files it finds: cargo check and cargo test, npx tsc --noEmit and npm test, npm test, make test, or pytest. That runs your project code and, for TypeScript, npx, so use it only in repos you trust. github.com
It avoids switching your branch
The worktree preflight says not to run git switch, checkout, reset --hard, clean or stash by default, and to review PRs with gh pr view and gh pr diff. github.com
Its other scripts read git state
release_gate.py says it is read-only with no network, and incident_ledger.py and audit_signals.py run git queries. The three scripts import no network library. github.com

Technical details

Talks to	GitHub or your git host when reviewing PRs, issues and CI
Scopes	Reads the worktree, diff, CI and remote state (git, and gh for PRs) Runs your project tests through run-tests.sh when no project command is named Writes, commits, pushes, publishes or posts only with authorization from the current request

Questions

What does the check skill do?

It makes Claude review a diff or pull request with evidence, triage issues and PRs, audit a whole project and check release readiness. It learns your project's rules from its README, manifests and CI.

How do I install check?

Use the install box at the top of this page. The Waza README also documents npx skills add tw93/Waza -a claude-code codex cursor -g -y, which installs all eight skills, and a ZIP upload for Claude Desktop.

Will it change my code or push without asking?

Not by default. SKILL.md says review, audit and triage are report-only, and edits, commits, pushes and publishing need authorization from your current request.

Is it safe to run?

It can run your tests and, if you allow it, push, publish and post to issues. Keep it on "review only" in repos you do not fully trust, and read each action before you approve.

How is it different from the built-in /review?

SKILL.md notes that /review is Anthropic's built-in plugin command for PR review and that Waza uses /check instead. Check also covers triage, release gates and project audits.

What is the difference from the hunt skill?

The Waza README describes hunt as systematic debugging that confirms the root cause before any fix. Check reviews changes and release readiness, and SKILL.md says it is not for debugging root causes.

Does it need an API key?

None is named in SKILL.md. It uses git, and gh when you review GitHub PRs and issues, with your existing logins.

Information

updated Nov 2, 2026

Repository: tw93/waza
Folder: skills/check

tw93/waza skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5

github.com · github.com
marketplace.json · github.com
github.com · github.com
run-tests.sh · github.com
release_gate.py · github.com