Skip to main content
skill CX Codex

pull-request-review — Claude Code skill: staff-level PR review

A skill that reviews a pull request or branch diff: checks the change against the issue, hunts correctness and security defects, and returns prioritized comments.

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add autohandai/code-cli --skill pull-request-review -a claude-code
Open on GitHub

What pull-request-review does

The skill gives the agent a fixed order of work. It pulls the diff first — git_diff for the working tree, git_diff_range against the base branch — and reads all of it before saying anything. Then it opens the linked issue, PR description or commit messages and writes one line about what the change claims to do. Files around the diff get opened too, since a change is only correct in context.

Review order is spelled out: does it do what it says, then correctness (edge cases, error paths, concurrency, ordering, off-by-one, null and empty inputs, resource cleanup, timeouts), then security (injection, path traversal, secrets, permission checks, unsafe deserialization, shell execution with user input), then compatibility (public API, config, storage formats, migrations, older callers), then tests, then maintainability. Where the environment allows, the skill runs tests and lint itself.

Output starts with a one-line verdict: approve, approve with nits, or request changes. Findings follow by severity, each with file:line, what is wrong, why it matters and a concrete fix; blocking items are kept apart from optional ones. The agent also states what it did not review. Everything runs through git and file reads, so a repository with no commit history gives the diff check nothing to work from.

What it helps with

  • confirm a PR meets the issue requirements instead of drifting in scope
  • catch correctness bugs and cite the exact file and line
  • run a security pass over the change before merge
  • check whether the edit breaks public API or migrations
  • sort blocking findings from optional nits

Example prompts for the agent

  • Review the current branch against main and give me a verdict
  • Check this PR: what the issue asks for versus what it changes
  • Review my changes and flag security issues with line numbers

How to install pull-request-review

1Skills CLI (Vercel)

npx skills add autohandai/code-cli --skill pull-request-review -a claude-code

2Manually: put the files in place

~/.claude/skills/pull-request-review/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/pull-request-review/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about pull-request-review

How do I install pull-request-review in Claude Code?

Run npx skills add autohandai/code-cli --skill pull-request-review -a claude-code. The skill lands at ~/.claude/skills/pull-request-review/SKILL.md.

What does pull-request-review do?

It reads a pull request or branch diff and returns ranked comments: whether the issue requirements are met, correctness and security defects, test coverage and compatibility.

Which platforms support the skill?

Claude, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode, each with its own install path.

Source text

SKILL.md · Apache-2.0
---
name: pull-request-review
description: Review a pull request or branch diff like a staff engineer. Verify it does what the issue asks, find correctness and security defects with file and line evidence, check tests and compatibility, and deliver prioritized, actionable comments. Use for PR review, branch review, and "review my changes".
allowed-tools: read_file find_grep fff_find list_tree git_status git_diff git_diff_range git_log run_command code_review
---

# Review the change, not the style guide

A useful review finds the defect that would have shipped and explains it so the
author can fix it in one pass.

## Establish scope

1. Get the diff (`git_diff` for the working tree, `git_diff_range` against the
   base branch). Read the whole diff before commenting on any part.
2. Read the linked issue, PR description, or commit messages. Write one line
   stating what the change claims to do.
3. Open the files around the diff; a change is only correct in context.

## Check in this order

1. **Does it do what it says**: every requirement from the issue is met, no
   silent scope changes, no leftover debug code or TODOs.
2. **Correctness**: edge cases, error paths, concurrency, ordering, off-by-one,
   null and empty inputs, resource cleanup, timeouts.
3. **Security**: injection, path traversal, secrets, permission checks,
   unsafe deserialization, shell execution with user input.
4. **Compatibility**: public API, config, storage formats, migrations, feature
   flags, older callers.
5. **Tests**: a test fails before and passes after for each fix; new behavior
   has coverage including edge cases; flaky patterns (sleeps, real network).
6. **Maintainability**: naming, duplication, module boundaries, comments only
   where logic is non-obvious.

Run the tests and lint yourself when the environment allows; a review with
evidence beats one with opinions.

## Deliver

- Verdict first: approve, approve with nits, or request changes, in one line.
- Findings ordered by severity. Each: file:line, what is wrong, why it matters,
  a concrete fix. Mark confidence when you could not verify.
- Separate blocking from optional. Keep nits short or drop them.
- Say what you did not review.

© autohandai · Apache-2.0 · autohandai/code-cli

Similar skills

All skills for Code review and refactoring →

From the same set5

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor