ClawPatch Report-Only Skill

A public Agent Skill for running ClawPatch in report-only mode.

This skill helps coding agents use ClawPatch to map a repository, review bounded feature slices, generate a Markdown findings report, inspect findings, and revalidate or triage findings without changing source code.

Install

Install with skills.sh:

npx skills add fabriqaai/clawpatch-report-only-skill

After installation, ask your agent to use the clawpatch-report-only skill when you want a ClawPatch review.

Requirements

ClawPatch requires Node.js 22+, Git 2.x, and a local Codex CLI provider. You can use ClawPatch through npx without a global install:

npx --yes clawpatch@latest --help

You can also install ClawPatch directly:

npm install -g clawpatch

See the upstream ClawPatch site for the latest project documentation: https://clawpatch.ai/.

What This Skill Covers

This skill covers the safe review path:

• doctor: check the local ClawPatch environment.
• init: initialize ClawPatch state in .clawpatch/.
• map: build ClawPatch's semantic feature map.
• review: run a bounded first-pass review, usually --limit 3 --jobs 1.
• report: write a Markdown findings report, usually .clawpatch/review.md.
• next: identify the next highest-priority finding.
• show: inspect one finding's evidence before recommending action.
• revalidate: re-check a finding without editing source files.
• triage: mark findings without changing source code.
• fix --dry-run: preview a fix only after the user explicitly asks about one specific finding.

What This Skill Intentionally Does Not Cover

This skill is intentionally not an autofix or shipping workflow. By default it does not:

• Run non-dry-run clawpatch fix.
• Edit source files.
• Commit changes.
• Push changes.
• Create pull requests.
• Create or switch branches.
• Create worktrees.
• Run destructive git commands.

If a user explicitly asks to fix a specific finding, the skill requires a dry-run first. A non-dry-run fix should only happen after the user approves the dry-run result.

Default Workflow

From the repository you want to review:

STATE_DIR=".clawpatch"

npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" doctor --no-input
npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" init --no-input
npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" map
npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" review --limit 3 --jobs 1 --provider codex --reasoning-effort high --no-input
npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" report -o "$STATE_DIR/review.md"
npx --yes clawpatch@latest --root . --state-dir "$STATE_DIR" next

Keep .clawpatch/ ignored by git. The directory contains review state, findings, reports, locks, and patch-attempt metadata.

Example Prompts

Use clawpatch-report-only to review this repo and summarize the top findings.

Run ClawPatch report-only with a limit of 3, then show me the next finding.

Inspect finding fnd_123 with ClawPatch and tell me whether it looks actionable.

Dry-run a ClawPatch fix for finding fnd_123, but do not apply it yet.

Skill Structure

This repo is a single-skill package:

.
├── SKILL.md
├── README.md
├── LICENSE
└── .gitignore

SKILL.md follows the Agent Skills convention: YAML frontmatter with name and description, followed by concise Markdown instructions.

License

MIT