AI-Friendly Project Initializer for Claude Code
Turn any blank or half-baked project into an AI-friendly development environment where Claude Code thrives.
Luoboa Init runs a 5-phase pipeline that transforms your project from "just code" into a Claude Code-optimized workspace:
Phase 1: Infrastructure Init (git / codegraph / .gitignore / openwolf)
↓
Phase 2: Generate CLAUDE.md (with Karpathy's 4 Principles)
↓
Phase 3: Run /init to analyze project structure
↓
Phase 4: Optimize CLAUDE.md (merge /init results + principles)
↓
Phase 5: Pick & install specialist agent skills (agency-agents + WeChat Mini Game)
Each phase completes before the next begins. No skipping.
Claude Code is powerful, but it makes predictable mistakes in new projects:
- ❌ Doesn't know platform-specific pitfalls (e.g., WeChat Mini Game's 19 hidden traps)
- ❌ Writes generic code that ignores version locks (
pixi.js@6→ accidentally upgrades to v7) - ❌ Over-engineers simple tasks, under-engineers critical constraints
- ❌ Has no context about your project's conventions and non-obvious rules
Luoboa Init fixes this by generating a CLAUDE.md that focuses on what AI gets wrong, not what the project looks like (AI can read code itself — it can't read your mind).
npx skills add opapa/luoboa-initAlternative: git clone
Global (all projects):
git clone https://github.com/opapa/luoboa-init.git ~/.claude/skills/luoboa-initProject-local (this project only):
git clone https://github.com/opapa/luoboa-init.git .claude/skills/luoboa-initIn any project directory, just type:
/luoboa-init
That's it. Follow the interactive prompts.
Runs scripts/init-project.sh which:
| Step | Action | Skip if already done |
|---|---|---|
| Git Init | git init |
✅ |
| .gitignore | Create with 20+ entries (node_modules, dist, .codegraph, .claude, etc.) | Merges, never overwrites |
| OpenWolf | openwolf init |
✅ |
| CodeGraph | codegraph init -i |
✅ |
All steps are fault-tolerant — missing tools are skipped gracefully.
Writes a CLAUDE.md that includes these behavioral guidelines (bilingual Chinese/English):
| # | Principle | Core Idea |
|---|---|---|
| 1 | Think Before Coding | Don't assume. Don't hide confusion. Surface tradeoffs. |
| 2 | Simplicity First | Minimum code that solves the problem. Nothing speculative. |
| 3 | Surgical Changes | Touch only what you must. Clean up only your own mess. |
| 4 | Goal-Driven Execution | Define success criteria. Loop until verified. |
These principles bias toward caution over speed — they prevent the most common AI mistakes.
Runs Claude Code's built-in /init to auto-detect:
- Dependencies and scripts (from
package.json) - Architecture and directory structure
- Configuration files (tsconfig, vite config, etc.)
- Existing documentation
If /init is unavailable, performs manual analysis.
Merges Phase 3 results into the CLAUDE.md while preserving the 4 principles. Key optimization rules:
- "What AI gets wrong" > "What the project looks like" — AI can read code, not your mind
- Constraints, not descriptions — write boundaries, not tours
- Version locks are critical —
pixi.js@6notpixi.js - Platform-specific pitfalls — must be explicitly stated
Choose from 50+ specialist agent roles across categories:
| Category | Examples |
|---|---|
| 🎮 Game Development | Game Designer, Level Designer, Technical Artist, Game Audio Engineer |
| 💻 Engineering | Frontend Developer, Backend Architect, WeChat Mini Program Developer |
| 🎨 Design | UI Designer, UX Researcher, Image Prompt Engineer |
| 📱 China Marketing | Xiaohongshu, Douyin, Bilibili, WeChat Official Account |
| 🔧 Specialized | MCP Builder, Document Generator, Compliance Auditor |
| 🆕 WeChat Mini Game | 19-pit checklist + platform cognitive rewrite + code templates |
Each selected skill is installed as a project-local skill (.claude/skills/) and activated with /skill-name.
luoboa-init/
├── SKILL.md # Main skill definition
├── README.md # This file (English)
├── README.zh.md # Chinese version
├── assets/
│ ├── qrcode.jpg # WeChat QR code
│ ├── agent-skill-template.md # Template for agent skill SKILL.md
│ └── wechat-minigame-skill.md # WeChat Mini Game skill template
├── references/
│ ├── agency-agents-catalog.md # Full agent role catalog (50+ roles)
│ ├── karpathy-principles.md # Karpathy's 4 coding principles
│ └── wechat-minigame-tips.md # WeChat Mini Game 19-pit checklist
└── scripts/
└── init-project.sh # Phase 1 infrastructure setup script
| Tool | Required? | Fallback |
|---|---|---|
| Git | Recommended | Skipped if unavailable |
| CodeGraph | Optional | Skipped if not installed |
| OpenWolf | Optional | Skipped if not installed |
Claude Code /init |
Optional | Manual analysis fallback |
- Existing CLAUDE.md: Merged, never overwritten — your content is preserved
- Existing .gitignore: Appended, never overwritten
- Partial run: Say "just Phase 2" or "skip Phase 5" — respects your choices
Found a bug or have a suggestion? Open an issue or PR at github.com/opapa.
萝卜啊 — AI practitioner, only writes about what I've personally verified.
🌐 Blog: www.luoboa.com
Latest AI tech articles, tool recommendations, curated Skills & MCP picks. Updated daily.
📱 WeChat Official Account: Follow for exclusive content
What you'll get:
- ✅ AI techniques personally verified — no rehashed content
- ✅ Agent interview insights from real interview scenarios
- ✅ Company AI adoption experience — practical, no fluff
📌 Reply with keywords to get resources (no forwarding tricks):
| Keyword | Resource |
|---|---|
| 技巧 | AI productivity techniques |
| Agent | Agent interview prep guide |
| 转型 | Company AI transformation plan |
📅 Update Schedule:
| Day | Content |
|---|---|
| Mon–Fri | Claude Code best practices + Best open-source Agent/Skill |
| Saturday | Weekly AI news + GitHub top charts |
| Sunday | AI Agent best practices |
| Platform | Link |
|---|---|
| 🐙 GitHub | github.com/opapa |
| 📘 Zhihu | zhihu.com/people/opapar |
| 📺 Bilibili | space.bilibili.com/316828422 |
| youtube.com/@ailuoboa |
MIT License — use freely, attribution appreciated.