A Codex-native packaging of Andrej Karpathy-inspired coding-agent guidelines.

• AGENTS.md is the repository-level operating context.
• .codex-plugin/plugin.json is the plugin manifest.
• skills/andrej-karpathy-skill/SKILL.md is the reusable skill.
• instruction.md is the copy-paste version for Codex Custom Instructions.
• EXAMPLES.md shows Codex-style examples.

English | Simplified Chinese

Coding agents often fail in predictable ways:

• they assume too much,
• they overbuild simple requests,
• they edit unrelated code,
• they call work done without a clear success check.

This package turns the original Karpathy-style guidelines into a Codex-ready skill, repo instruction file, and Custom Instructions block.

Use this repository as a local Codex plugin source.

1. Add the repository to your Codex plugin sources.
2. Codex reads .codex-plugin/plugin.json.
3. Codex loads skills/andrej-karpathy-skill/SKILL.md.

Plugin name:

andrej-karpathy-skills

Skill name:

andrej-karpathy-skill

If someone does not want to install the plugin or add an AGENTS.md, they can use the standalone Custom Instructions version:

1. Open instruction.md.
2. Copy the block under "Paste the block below".
3. Paste it into Codex Settings -> Custom Instructions.

That version is self-contained, so Codex can apply the same behavior without reading any repo files.

.codex-plugin/plugin.json
AGENTS.md
EXAMPLES.md
instruction.md
LICENSE
README.md
README.zh.md
skills/andrej-karpathy-skill/SKILL.md

Inspired by Andrej Karpathy's public comments on LLM coding behavior and by the original community repo that packaged those ideas as reusable agent guidance.

Original repo credit:

multica-ai/andrej-karpathy-skills

MIT