What Is andrej-karpathy-skills?
andrej-karpathy-skills is a single CLAUDE.md file derived from Andrej Karpathy's publicly shared observations on working effectively with LLMs for software development. It distills his philosophy into four actionable principles that shape how Claude approaches every coding task.
The Four Principles
Don't Assume — Surface Tradeoffs
Claude under this skill never silently fills in ambiguities. If a requirement can be interpreted multiple ways, it names the interpretations and asks which is intended. If an implementation choice involves a tradeoff (performance vs. simplicity, flexibility vs. safety), it surfaces the tradeoff before proceeding.
This principle alone eliminates a large class of AI coding frustrations: code that technically works but made the wrong call on something the developer would have cared about.
Minimum Viable Code
The skill enforces a preference for the simplest implementation that meets the stated requirements. No extra abstractions, no premature generalization, no framework where a function will do. This keeps codebases readable, maintainable, and auditable.
Push Back When Warranted
Claude is instructed to express disagreement or concern before complying with a request it considers misguided, risky, or likely to cause problems. This isn't refusal — it's professional candor. After flagging the concern, Claude proceeds if the user confirms.
Goal-Driven Execution
The skill keeps Claude anchored to the actual objective, not just the literal task. If following an instruction verbatim would undermine the goal, Claude says so. This is especially valuable in long agentic sessions where task drift can cause real damage.
How to Install andrej-karpathy-skills?
Clone the repo and copy CLAUDE.md to your project root git clone https://github.com/multica-ai/andrej-karpathy-skills cp andrej-karpathy-skills/CLAUDE.md ~/your-project/CLAUDE.md
Or use the skills CLI npx skills add multica-ai/andrej-karpathy-skills