Review Before PR Skill

Review your code before you open a PR. Get a structured, AI-powered code review (design, security, performance, tests) as a single markdown report—no GitHub, no API keys, no PR required. Works on any git repo, offline. Use it with any AI agent that supports skills (Cursor, Claude Code, Windsurf, Gemini, etc.).

Why this exists

Most code-review tools assume you already have a pull request. This skill is for the step before that: you want feedback on your branch, your staged changes, or your local edits so you can fix issues and polish the diff before raising a PR. No GitHub dependency, no external services—just your repo and your AI agent.

Use it when you:

Want a review of your branch vs main (or any base) before opening a PR
Want to double-check what you're about to commit (--staged)
Want feedback on all uncommitted work (--local)

You get one markdown file with Critical, High, Medium, and Positive notes, plus suggested code snippets in the doc only—your source files stay untouched unless you choose to apply changes.

What makes it different

	This skill	Typical PR review tools
When	Before you open a PR	After you open a PR
Needs GitHub?	No	Yes (PR number, `gh` CLI, etc.)
Needs API keys?	No	Often (GitHub token, service account)
Works offline?	Yes	Usually no
Modes	Branch vs branch, staged only, all local	PR-only
Output	One markdown report (git-ignored)	Inline comments on the PR

So: review first, PR when you're ready.

Repository layout

review-before-pr/
├── README.md
├── SKILL.md
├── references/
│   └── REVIEW_CHECKLIST.md
└── scripts/
    └── generate_diff.sh

Installation

Clone this repo directly into your AI agent's skills directory. From your project root (the repo you want to review), run the command for your agent.

Agent	Skills directory	Install command
Cursor	`.cursor/skills/review-before-pr`	`git clone https://github.com/amanblog/review-before-pr.git .cursor/skills/review-before-pr`
Claude Code	`.claude/skills/review-before-pr`	`git clone https://github.com/amanblog/review-before-pr.git .claude/skills/review-before-pr`
Windsurf	`.windsurf/skills/review-before-pr`	`git clone https://github.com/amanblog/review-before-pr.git .windsurf/skills/review-before-pr`
Continue	`.continue/skills/review-before-pr`	`git clone https://github.com/amanblog/review-before-pr.git .continue/skills/review-before-pr`
Other	Check your agent's docs for "skills" path	Clone into that path as `review-before-pr`.

Ensure your repo has **.review/** in .gitignore so diff and review output aren't committed. The script will add it automatically if missing.

Usage

You can use the skill in three ways. How you invoke it depends on your AI agent (see How agents invoke skills below).

1. Ask the agent to review your branch (recommended)

In chat, ask the agent to review your current branch against a base branch (e.g. main or develop) using this skill. The agent will run the diff script and then produce the review.

Example prompts:

"Review my current branch against main using the review-before-pr skill"
"Review my branch against develop using review-before-pr"
"Review my staged changes using the review-before-pr skill"
"Review my local (uncommitted) changes using review-before-pr"

The agent runs the script, writes the diff to .review/diff.txt, analyzes it, and writes the report to e.g. .review/review-<branch>.md.

2. Use a slash command or pick the skill (when your agent supports it)

In Cursor: type **/** in Agent chat, then choose review-before-pr from the list (or type the skill name). When prompted, say which base branch to compare against (e.g. main, develop) or that you want staged/local changes.

In Claude Code: type **/review-before-pr. In Windsurf: type **@review-before-pr in Cascade. See the table below for other agents.

3. Run the diff yourself, then ask for a review

Generate the diff from the CLI (use the path where you installed the skill, e.g. .cursor/skills/review-before-pr), then ask the agent to review .review/:

# From project root — <SKILL_DIR> = where you installed the skill (e.g. .cursor/skills/review-before-pr)
<SKILL_DIR>/scripts/generate_diff.sh main my-feature-branch
# Or: --staged for staged only, --local for all local changes

Examples:

Cursor: .cursor/skills/review-before-pr/scripts/generate_diff.sh main my-branch
Claude Code: .claude/skills/review-before-pr/scripts/generate_diff.sh main my-branch

Then in chat: "Review the diff in .review/diff.txt using the review-before-pr skill". The agent will read the diff and write the report under .review/.

How agents invoke skills

Agent	How to invoke this skill
Cursor	Type `/` in Agent chat, then select review-before-pr. You can also type the skill name. Cursor skills docs
Claude Code	Type `/review-before-pr`. Claude Code skills
Windsurf	Type `@review-before-pr` in Cascade. Windsurf Cascade skills
Continue	Use natural language in Agent mode, e.g. "Review my current branch against main using the review-before-pr skill".
Other	Check your agent's docs; invoke by name or via the UI.

What you get

The review document is organized into Critical, High priority, Medium / good-to-have, and Positive notes, with suggested fixes as code blocks in the doc. For very large diffs (e.g. 5,000+ lines), the agent summarizes by file/module.

Example: See examples/sample-review.md for a complete review output.

✨ New Features (v2.0)

Smart Filtering

Automatically excludes noise files (lock files, SVGs, build artifacts) to keep reviews focused on actual code. Saves AI tokens and improves review quality.

Secret Detection

Scans for hardcoded secrets (API keys, tokens, passwords) before generating diffs. Prevents accidental exposure to AI APIs.

Cross-Platform Support

Works on Mac, Linux, and Windows. Python version available for full Windows compatibility.

Configuration

Customize ignore patterns, secret detection, and review focus areas with .reviewrc file. See docs/CONFIGURATION.md.

Context-Aware Reviews

AI proactively searches your codebase for related code, reducing false positives and improving accuracy.

Metrics Tracking

Track code quality trends over time with automatic metrics logging.

Configuration

Create a .reviewrc file in your project root to customize behavior:

{
  "baseBranch": "develop",
  "ignorePatterns": ["*.generated.ts", "migrations/*"],
  "review": {
    "categories": ["security", "performance"],
    "contextGathering": true
  }
}

Full documentation: docs/CONFIGURATION.md

Advanced Features

Safe Fix Application (Coming Soon)

After reviewing, apply suggested fixes safely via patch files:

# Generate patch for Critical fixes
python scripts/apply_fixes.py --priority critical --preview

# Review the patch, then apply
python scripts/apply_fixes.py --priority critical --apply

# Rollback if needed
git apply -R .review/patches/critical-fixes.patch

Metrics Tracking

View your code quality trends:

python scripts/analyze_metrics.py

Pre-commit Integration

Automatically review staged changes before committing. See docs/PRECOMMIT_SETUP.md.

Troubleshooting

Windows: "bash: command not found"

Solution 1: Install Git Bash (comes with Git for Windows)

winget install Git.Git

Solution 2: Use Python version (works on all platforms)

python .cursor/skills/review-before-pr/scripts/generate_diff.py --staged

"Secret detected" error

If you need to review code containing test API keys:

# Temporarily disable secret detection
./scripts/generate_diff.sh --allow-secrets main feature

Large diff timeout

For diffs >5000 lines, the AI will summarize. To review specific parts:

# Split into smaller reviews by directory
./scripts/generate_diff.sh main feature .review/frontend.txt -- src/frontend/
./scripts/generate_diff.sh main feature .review/backend.txt -- src/backend/

Full troubleshooting guide: docs/TROUBLESHOOTING.md

Comparison with Other Tools

Feature	review-before-pr	CodeRabbit	Qodo	GitHub Copilot	SonarQube
Pre-PR reviews	✅ Core focus	⚠️ Supports	⚠️ Supports	❌ PR-only	⚠️ CI-only
Local execution	✅ Yes	⚠️ CLI only	❌ No	❌ No	✅ Yes
No API keys needed	✅ Yes	❌ Requires account	❌ Requires account	❌ Requires license	✅ Yes
Works offline	✅ Yes	❌ No	❌ No	❌ No	✅ Yes
AI-powered	✅ Yes (your agent)	✅ Yes	✅ Yes	✅ Yes	⚠️ Rules-based
Context-aware	✅ Yes (v2.0)	✅ Yes	✅ Yes	✅ Yes	⚠️ Limited
Secret detection	✅ Yes (v2.0)	✅ Yes	⚠️ Limited	❌ No	✅ Yes
Metrics tracking	✅ Yes (v2.0)	✅ Yes	✅ Yes	❌ No	✅ Yes
Cost	Free	$12-48/user/mo	$19-39/user/mo	$10-39/user/mo	Free (OSS)

Requirements

A git repository
An AI agent that supports skills (Cursor, Claude Code, Windsurf, Continue, etc.)
Python 3.7+ (for cross-platform support and secret detection)

Documentation

Configuration Guide - Customize for your project
Troubleshooting - Common issues and solutions
Pre-commit Setup - Automatic reviews before commits
Sample Review Output - See what you'll get

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

License

MIT

Navigation Menu

Repository files navigation

Review Before PR Skill

Why this exists

What makes it different

Repository layout

Installation

Usage

1. Ask the agent to review your branch (recommended)

2. Use a slash command or pick the skill (when your agent supports it)

3. Run the diff yourself, then ask for a review

How agents invoke skills

What you get

✨ New Features (v2.0)

Smart Filtering

Secret Detection

Cross-Platform Support

Configuration

Context-Aware Reviews

Metrics Tracking

Configuration

Advanced Features

Safe Fix Application (Coming Soon)

Metrics Tracking

Pre-commit Integration

Troubleshooting

Windows: "bash: command not found"

"Secret detected" error

Large diff timeout

Comparison with Other Tools

Requirements

Documentation

Contributing

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
docs		docs
examples		examples
references		references
scripts		scripts
.gitignore		.gitignore
.reviewrc.example		.reviewrc.example
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
QUICKSTART.md		QUICKSTART.md
README.md		README.md
SKILL.md		SKILL.md
UPCOMING_FEATURES.md		UPCOMING_FEATURES.md

Latest commit

History

Folders and files

Repository files navigation

Review Before PR Skill

Why this exists

What makes it different

Repository layout

Installation

Usage

1. Ask the agent to review your branch (recommended)

2. Use a slash command or pick the skill (when your agent supports it)

3. Run the diff yourself, then ask for a review

How agents invoke skills

What you get

✨ New Features (v2.0)

Smart Filtering

Secret Detection

Cross-Platform Support

Configuration

Context-Aware Reviews

Metrics Tracking

Configuration

Advanced Features

Safe Fix Application (Coming Soon)

Metrics Tracking

Pre-commit Integration

Troubleshooting

Windows: "bash: command not found"

"Secret detected" error

Large diff timeout

Comparison with Other Tools

Requirements

Documentation

Contributing

License

About

Resources

Contributing

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages