# Andrej Karpathy Skills

Source: https://trendhub.net/skills/multica-ai-andrej-karpathy-skills
Repository: https://github.com/multica-ai/andrej-karpathy-skills
Author: Multica (https://github.com/multica-ai)
License: MIT
GitHub stars: 233,682

Andrej Karpathy Skills is a single agent skill, also shipped as a CLAUDE.md file, that turns Andrej Karpathy's observations on LLM coding mistakes into four working rules: think before coding, keep it simple, make surgical changes and define verifiable goals. It suits developers whose agent guesses, over-engineers or rewrites code nobody asked it to touch.

## What problems it solves

- **The agent guesses what you meant**: Think Before Coding makes it state assumptions, list competing readings and ask before writing any code.
- **Simple tasks come back over-engineered**: Simplicity First rules out unrequested features, single-use abstractions and error handling for cases that cannot happen.
- **Diffs touch code you never mentioned**: Surgical Changes limits edits to the request and leaves unrelated code, comments and formatting alone.
- **"Make it work" never ends**: Goal-Driven Execution turns each task into a test or check the agent can loop on until it passes.

## Install

### Claude Code (plugin)

Makes the skill available in every project. The marketplace uses the repository's original name, which GitHub redirects

```
/plugin marketplace add forrestchang/andrej-karpathy-skills
/plugin install andrej-karpathy-skills@karpathy-skills
```

### Claude Code (CLAUDE.md, one project)

Writes the rules to CLAUDE.md in the current folder and replaces an existing file; to keep yours, append with \>\> instead

```
curl -o CLAUDE.md https://raw.githubusercontent.com/forrestchang/andrej-karpathy-skills/main/CLAUDE.md
```

Pick one route per project: the plugin loads the skill across all your Claude Code projects, while CLAUDE.md applies to a single repository and can sit next to your own project rules. Cursor users copy the repository's .cursor/rules/karpathy-guidelines.mdc rule into their project instead.

## Skills

Skill names are shown as /name. Depending on the agent, a skill runs when it is relevant or when you call it by name; plugin installs may add a prefix.

- `/karpathy-guidelines`: Applies four rules against common LLM coding mistakes: ask before guessing, keep code simple, change only what was asked, verify. (https://trendhub.net/skills/multica-ai-andrej-karpathy-skills/karpathy-guidelines.markdown)

## Questions

### Did Andrej Karpathy write these skills?

No. The repository turns points from a post Karpathy published on X into rules. It was written by developer Jiayuan (forrestchang) and is maintained under the Multica organization.

### Is it free?

Yes. The README and the skill file both state the MIT license.

### Plugin or CLAUDE.md: which should I use?

The plugin loads the skill when a coding task matches, in every project. CLAUDE.md is always in context for one project and is easy to merge with your own project rules.

### Will it slow down simple tasks?

The rules favour caution over speed. For typo fixes and one-line changes the agent is told to use judgment rather than the full process.

## Before you install

The skill is plain instructions with no scripts; it runs no commands and makes no network requests. The CLAUDE.md install command overwrites an existing CLAUDE.md in the folder, so append to it if you already have one.
