Karpathy 风格代码审查规则生成器

You are an expert in AI-assisted software development best practices, deeply familiar with Andrej Karpathy's observations on LLM coding pitfalls.

Generate a comprehensive CLAUDE.md (or AGENTS.md) rules file for my project that addresses these common LLM coding failures:

1. Over-engineering prevention: Rules against adding unnecessary abstractions, premature optimization, or gold-plating
2. Deletion discipline: Prefer deleting dead code over commenting it out; resist accumulating unused imports/variables
3. Simplicity enforcement: One obvious way to do things; no clever tricks; readable > compact
4. Testing sanity: Write tests that test behavior, not implementation; no mocking everything; integration tests over unit tests for I/O
5. Error handling: No swallowing errors; no generic catch-all; specific error types
6. Naming conventions: Descriptive names; no abbreviations; consistency throughout
7. Context boundaries: What the AI should and should not modify; protected files list
8. Commit hygiene: Small, focused commits; meaningful messages; no fix stuff commits

Project context:

• Language/Framework: [YOUR STACK]
• Project type: [WEB APP / CLI / LIBRARY / etc.]
• Team size: [SOLO / SMALL / LARGE]
• Key constraints: [ANY SPECIFIC RULES]

Output the complete file in markdown, ready to drop into my project root.