multica-ai/andrej-karpathy-skills
multica-ai/andrej-karpathy-skills provides a single CLAUDE.md guideline file designed to improve Claude Code behavior, derived from Andrej Karpathy's observations on large language model coding pitfalls. The repository targets software developers working with Claude Code and Cursor. It addresses common problems where models make unverified assumptions, conceal confusion, omit tradeoffs, overcomplicate code architectures, and modify or delete unrelated code during implementation. The file introduces four concrete operational principles to guide coding interactions: Think Before Coding, Simplicity First, Surgical Changes, and Goal-Driven Execution. These principles instruct models to state assumptions explicitly, implement the minimum viable code, limit modifications strictly to requested edits, and transform tasks into verifiable criteria with iterative test loops.
The README directs Claude Code plugin installation to the separate `forrestchang/andrej-karpathy-skills` repository, using "/plugin marketplace add forrestchang/andrej-karpathy-skills" followed by "/plugin install andrej-karpathy-skills@karpathy-skills". For a new project, the README gives a download from the forrestchang repository; this command overwrites an existing CLAUDE.md: "curl -o CLAUDE.md https://raw.githubusercontent.com/forrestchang/andrej-karpathy-skills/main/CLAUDE.md" or append it to an existing CLAUDE.md file. Cursor users can apply the committed rule at ".cursor/rules/karpathy-guidelines.mdc". The GitHub snapshot shows the repository is not archived. GitHub metadata does not identify a license; the README names MIT. Check the license file before use. The forrestchang repository linked by the README is another reference; note that it is a separate repository.
Data as of : 228687 stars · 22803 forks.