Coding in "Natural Language" — Prevent vibe coding with structured AI prompts
NatLangDev (Natural Language Developer Guide) is a collection of curated guidelines that help developers communicate effectively with AI coding assistants like Claude Code.
"Vibe coding" happens when developers prompt AI assistants without clear constraints or standards, resulting in:
- Inconsistent code quality
- Security vulnerabilities
- Poor architectural decisions
- Code that "works" but violates best practices
Instead of vague prompts like "write me a web server", NatLangDev provides explicit rules that AI assistants follow. By adding these guidelines to your AI's context, you get code that adheres to proven standards — whether that's NASA's safety-critical rules, idiomatic language patterns, or framework-specific conventions.
natlangdev/
├── languages/
│ ├── rust/ # Rust guidelines
│ ├── python/ # Python guidelines
│ └── typescript/ # TypeScript guidelines
├── frameworks/
│ ├── nextjs/ # NextJS (TSX) guidelines
│ ├── axum/ # Rust Axum guidelines
│ ├── fastapi/ # FastAPI (Python) guidelines
│ └── pytorch/ # PyTorch (Python) guidelines
├── patterns/ # Cross-language design patterns
├── workflows/ # Development workflows
├── stacks/ # Recommended tech stacks
└── vscode/ # VS Code configurations
- Power of 10 Rules for Rust — NASA JPL's safety-critical coding guidelines adapted for Rust
- Power of 10 Rules for Python — NASA JPL's safety-critical coding guidelines adapted for Python
- Power of 10 Rules for TypeScript — NASA JPL's safety-critical coding guidelines adapted for TypeScript
- Power of 10 Rules for NextJS — NASA JPL's safety-critical coding guidelines adapted for NextJS
- Monorepo Preferences — Bun + Turborepo + Turbopack + RSC guidelines
- Power of 10 Rules for Axum — NASA JPL's safety-critical coding guidelines adapted for Axum
- Power of 10 Rules for FastAPI — NASA JPL's safety-critical coding guidelines adapted for FastAPI
- Power of 10 Rules for PyTorch — NASA JPL's safety-critical coding guidelines adapted for PyTorch
- Feature-Sliced Hexagonal Architecture — The definitive pattern for parallel AI agent swarm development in a monorepo
- Minimal Dependencies — Philosophy on when to use vs. avoid libraries
- Code Comments — Inline, concise, and descriptive commenting guidelines
- Lisa Flow — Structured AI development with automatic TDD and self-healing tests
- FastMVP — Vercel + NextJS + Convex/Supabase + Clerk + Stripe for rapid MVP development
- Review & Orchestration Mode — VS Code settings optimized for reviewing AI-generated code (disables autocomplete, strips UI bloat, enhances diff view)
- Claude Code — Run Claude Code directly in VS Code
- Error Lens — Inline error/warning highlighting for quick review
Other extensions vary by project — ask Claude Code for recommendations based on your stack.
Contributions welcome! Feel free to submit guidelines for additional languages or improvements to existing ones.
CC0 1.0 Universal — Public Domain
