LLM Skills
~/catalogue/full-stack//generate-pr-review
Full-stacksource GitHub

Générer un projet

/generate-pr-review

Tout d'abord, découvrez les caractéristiques du projet à partir de plusieurs sources :

citypaulcitypaul
714
20 mars 2026
MIT License
// contenu du skill

description: Generate project-specific PR review automation that combines global rules with project conventions

allowed-tools: Read, Glob, Grep, Write, Bash(git:), Bash(npm:), Bash(cat:*)


Generate Project-Specific PR Review Automation

You are generating a customized PR review configuration for this project. This combines:

  1. Global rules from CLAUDE.md (TDD, TypeScript strict, functional patterns)
  2. Project-specific rules discovered from codebase analysis

Step 1: Analyze the Project

First, discover the project's characteristics from multiple sources:

AI/LLM Configuration Files

Critical: Check for existing AI assistant configurations that define project rules:

bash
# Claude Code
.claude/CLAUDE.md
.claude/settings.json
.claude/agents/*.md
.claude/skills/*/SKILL.md
CLAUDE.md (root level)

# Cursor
.cursorrules
.cursor/rules/*.md

# GitHub Copilot
.github/copilot-instructions.md

# Aider
.aider.conf.yml
.aiderignore

# Codeium
.codeium/config.json

# Other common patterns
.ai-rules
.llm-config
AGENTS.md
CONTRIBUTING.md (often contains coding standards)

Extract rules, patterns, and conventions from these files - they represent explicit project decisions.

Project Documentation

Check for documented conventions:

bash
# Architecture Decision Records
docs/adr/*.md
docs/decisions/*.md
adr/*.md
architecture/decisions/*.md

# General documentation
docs/*.md
README.md
CONTRIBUTING.md
DEVELOPMENT.md
CODING_STANDARDS.md
STYLE_GUIDE.md

# API documentation
docs/api/*.md
API.md

Parse ADRs for architectural decisions that affect code review (e.g., "We use Zod for all validation").

CI Pipeline Detection

Check for CI configuration files and document the pipeline:

bash
# GitHub Actions
.github/workflows/*.yml

# Forgejo Actions
.forgejo/workflows/*.yml

# Woodpecker
.woodpecker/*.yml

# Other
Jenkinsfile
.circleci/config.yml
.gitlab-ci.yml

Extract: pipeline steps, execution order, Node/runtime versions, environment variables, and any known differences from local development.

Tech Stack Detection

Check for:

  • package.json - Dependencies, scripts, type of project
  • tsconfig.json - TypeScript configuration (note strict, noUncheckedIndexedAccess, and other strict flags)
  • .eslintrc.* or eslint.config.* - Linting rules
  • jest.config.* or vitest.config.* - Testing setup
  • biome.json - Biome configuration
  • .prettierrc* - Formatting rules

Framework/Library Detection

Look for:

  • React/Vue/Angular/Svelte in dependencies
  • Express/Fastify/Hono for backend
  • Testing libraries (Jest, Vitest, Testing Library)
  • State management (Redux, Zustand, etc.)
  • Schema libraries (Zod, io-ts, Yup, etc.)

Existing Code Conventions

Search for:

  • Existing code patterns in src/
  • Test file organization
  • Naming conventions
  • Import patterns

Step 2: Create Project Review Configuration

Based on analysis, create .claude/agents/pr-reviewer.md in the project:

markdown
---
name: pr-reviewer
description: >
  Project-specific PR review combining global standards with [PROJECT_NAME] conventions.
  Use proactively for review guidance or reactively to analyze PRs.
tools: Read, Grep, Glob, Bash
model: sonnet
color: cyan
---

# [PROJECT_NAME] PR Review

This reviewer enforces:
1. **Global standards** - TDD, TypeScript strict, functional patterns
2. **Project conventions** - [Discovered patterns]

## Global Rules (Non-Negotiable)

### TDD Compliance
- Every production code change needs corresponding tests
- Tests come BEFORE implementation (test-first)
- Tests verify behavior, not implementation

### Testing Quality
- Test through public API only
- No `let`/`beforeEach` - use factory functions
- Factory functions validate with real schemas (don't redefine)
- No spying on internal methods
- No 1:1 mapping between test files and implementation files

### TypeScript Strictness
- No `any` types - ever
- No type assertions without justification
- `type` for data structures, `interface` for behavior contracts
- Schema-first at trust boundaries (Zod/Standard Schema)
- `readonly` on immutable data
[IF noUncheckedIndexedAccess IS ENABLED: - All indexed access returns `T | undefined` — use optional chaining or explicit guards, never non-null assertions]
[ADD ANY OTHER STRICT FLAGS DETECTED FROM tsconfig.json]

### Functional Patterns
- No data mutation (no `.push()`, `.splice()`, property assignment)
- Pure functions (no side effects)
- Early returns (no nested if/else)
- Array methods over loops
- Options objects over positional parameters
- No comments (self-documenting code)

### General Quality
- No `console.log` or debug statements
- No TODO comments without linked issues
- No hardcoded secrets
- Small, focused changes

---

## Project-Specific Rules

[GENERATED BASED ON PROJECT ANALYSIS]

### Rules from Existing Configuration

[Extract from .cursorrules, CLAUDE.md, .github/copilot-instructions.md, CONTRIBUTING.md, etc.]

**Source files found:**
- [L
// source originale publique
citypaul/.dotfiles
/claude/.claude/commands/generate-pr-review.md
Licence : MIT License
Projet indépendant, non affilié à Anthropic. Ce skill reste la propriété de son auteur original.
// installer ce skill
Collez cette commande dans votre terminal à la racine de votre projet :
mkdir -p .claude/commands && curl -o ".claude/commands/generate-pr-review.md" "https://raw.githubusercontent.com/citypaul/.dotfiles/main/claude/.claude/commands/generate-pr-review.md"
Ensuite dans Claude Code, tapez /generate-pr-review pour l'activer.
open_in_newVoir la source originale
// sauvegarder
Sauvegarde disponible après connexion.
loginSe connecter pour sauvegarder
// informations
Créateurcitypaul
Étoiles 714
CatégorieFull-stack
LicenceMIT License
Mis à jour20 mars 2026
Format.md
AccèsGratuit
// similaires

Skills Full-stack

Voir toutarrow_forward