| name | claude-review |
| description | Run Claude Code as a chat-only reviewer for a local diff or branch, including adversarial pressure-testing when requested. Use when the user explicitly wants a Claude second opinion without posting GitHub comments. |
Claude Review
Use this skill when the user asks Claude Code to review local changes or
pressure-test an implementation or design.
Contract
- Run Claude Code through
scripts/claude-review.mjs inside this skill.
- Resolve the helper relative to the absolute path of this loaded
SKILL.md.
- Use normal review mode by default:
node /absolute/path/to/claude-review/scripts/claude-review.mjs --mode review <user request>
- Use adversarial mode when the user asks to challenge assumptions, identify
reasons not to ship, or pressure-test the design:
node /absolute/path/to/claude-review/scripts/claude-review.mjs --mode adversarial <user request>
- Preserve user-supplied flags and focus text after the mode argument.
- If the user gives
--base <ref>, pass it through. Otherwise review the current
working tree.
- Keep adversarial feedback grounded in material risks, broken assumptions,
rollback hazards, failure paths, and real user impact.
- Do not edit files after receiving Claude's review unless the user explicitly
asks for follow-up implementation.
- Return Claude's review output to the user. Do not post GitHub comments.
Expected Output
Claude should report defensible findings first, with file and line references
where possible. If it finds no material issues, it should say so directly.
