tailwindlabs / tailwindcss
github.com·21h ago
Developer Tools
GitHub Trending·May 20·5 min read
multica-ai / andrej-karpathy-skills
TL;DR
Large language models (LLMs) like Claude often make silent assumptions, overcomplicate code, and edit unrelated sections when assisting with programming tasks. A GitHub repository created a single CLAUDE.md configuration file based on Andrej Karpathy's observations about LLM coding failures, implementing four principles: think before coding, simplicity first, surgical changes, and goal-driven execution.
✦ Why It Matters
Engineers can use this CLAUDE.md template to reduce costly mistakes and overengineering when using AI coding assistants.
Key Takeaways
- Adopt explicit reasoning to avoid wrong assumptions.
- Simplify code to prevent overengineering.
- Limit changes to only what is necessary.
- Define clear success criteria for tasks.
How It Works
The repository's principles guide developers in using LLMs effectively. 'Think Before Coding' emphasizes clarifying assumptions and seeking clarification when confused.
'Simplicity First' advocates for minimalistic code solutions, while 'Surgical Changes' restricts edits to only necessary areas. 'Goal-Driven Execution' transforms vague tasks into specific, verifiable goals, allowing LLMs to work more independently and efficiently.