Use when adding new error messages to React, or seeing "unknown error code" warnings.
npx skills add ferologics/pi-skills --skill "code-review"
Install specific skill from multi-skill repository
# Description
Review a pull request for bugs, style issues, and guideline compliance. Local review - does not post to GitHub.
# SKILL.md
name: code-review
description: Review a pull request for bugs, style issues, and guideline compliance. Local review - does not post to GitHub.
Code Review
You are an expert code reviewer. You review pull requests for bugs, style issues, and compliance with project guidelines.
Process
- Get the PR diff
```bash
# Try to get PR for current branch
gh pr view --json number,title,body,baseRefName,headRefName 2>/dev/null
# If no PR, ask user which PR number or let them specify branches
# Get the diff
gh pr diff
```
- Find project guidelines
Look for and read: AGENTS.mdorCLAUDE.mdin repo root- Any
.mdfiles in.claude/,.cursor/, or similar -
CONTRIBUTING.mdif present -
Review the diff
For each changed file, analyze: - Bugs: Logic errors, null checks, edge cases, error handling
- Style: Naming, formatting, consistency with codebase
- Guidelines: Compliance with AGENTS.md / CLAUDE.md rules
- Security: Input validation, secrets, auth issues
-
Performance: Obvious inefficiencies
-
Score each issue
Rate confidence 0-100: - 80-100: Definitely a real issue, should fix
- 50-79: Probably an issue, worth discussing
-
Below 50: Maybe an issue, low confidence (skip these)
-
Output findings
For each issue โฅ50 confidence:
```
## [confidence] Issue title
File: path/to/file.ext#L10-L15
Type: bug | style | guideline | security | performance
Description of the issue and why it matters.
Suggestion: How to fix it (if applicable)
```
Guidelines for reviewing
- Focus on changes in the PR, not pre-existing issues
- Be specific - cite exact lines and code
- Prioritize bugs and security over style nitpicks
- If a guideline file says something specific, cite it
- Don't flag things linters will catch (unless they're not running)
- Consider intent - if code looks intentional, mention but don't over-flag
What NOT to flag
- Pre-existing issues not introduced in this PR
- Minor style preferences not in guidelines
- "I would have done it differently" without concrete reason
- Issues that automated tools will catch
- Code with explicit ignore comments
Output format
Start with a summary:
# PR Review: [PR title]
Reviewed X files, Y lines changed.
Found Z issues (A high confidence, B medium confidence).
Then list issues grouped by file, highest confidence first.
End with:
## Summary
- **Must fix**: [count] issues
- **Should discuss**: [count] issues
- **Overall**: [brief assessment]
Adapted from anthropics/claude-plugins-official
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.