A single CLAUDE.md file distilling Andrej Karpathy's observations on LLM coding pitfalls into 4 hard rules: think before coding, prefer simplicity over abstraction, make targeted changes only, and verify by outcome not intent. Packaged as a proper Agent Skill with .cursor/rules, plugin.json, and multi-tool support. 140K stars — the fastest validation that developers already knew AI was doing these things wrong.
选择你使用的 Agent
方法一：命令行安装（推荐）
推荐（无需提前安装 clawhub）
npx clawhub@latest --dir ~/.claude/skills install multica-karpathy-guidelines或使用 clawhub CLI（需提前安装）
clawhub --dir ~/.claude/skills install multica-karpathy-guidelines⚠️ 需要 Node.js 18+，没有 Node？请使用下方方法二直接下载 ZIP。 安装 Node.js →
方法二：手动下载安装（无需 Node）
下载 ZIP，解压后将文件夹放到以下路径，重启 Agent 即可：
安装路径
~/.claude/skills/multica-karpathy-guidelines/💡解压后将文件夹放到上方路径，重启 Agent 即可生效
Karpathy Guidelines (multica-ai) 是一款 AI 智能体 技能，为你的 AI 助手赋予强大的新能力。A single CLAUDE.md file distilling Andrej Karpathy's observations on LLM coding pitfalls into 4 hard rules: think before coding, prefer simplicity over abstraction, make targeted changes only, and verify by outcome not intent. Packaged as a proper Agent Skill with .cursor/rules, plugin.json, and multi-tool support. 140K stars — the fastest validation that developers already knew AI was doing these things wrong.
安装只需几秒，即可在 AI 工作流中立即使用。
Karpathy Guidelines (multica-ai) 通过模型上下文协议（MCP）直接与 Cursor、Claude Code、Codex CLI 集成，让你的 AI 智能体获得默认情况下没有的专业工具。
安装后，你可以在对话中自然地调用其功能——只需描述你想做的事，AI 助手就会使用 Karpathy Guidelines (multica-ai) 来完成。
安装 Karpathy Guidelines (multica-ai) 只需几秒。将技能文件夹放到对应目录：
# 个人级（所有项目可用）
~/.claude/skills/multica-karpathy-guidelines/
# 项目级（仅当前项目）
.claude/skills/multica-karpathy-guidelines/放置好技能文件夹后，重启你的 AI 客户端（Claude Code、Cursor、Gemini CLI、OpenClaw 等）。之后可以用 /multica-karpathy-guidelines 主动调用，或让 AI 根据上下文自动发现并使用。
Karpathy Guidelines (multica-ai) 兼容 Cursor、Claude Code、Codex CLI。它遵循开放的 MCP（模型上下文协议）标准，无需任何修改即可在所有兼容 MCP 的 AI 客户端上运行。
安装 Karpathy Guidelines (multica-ai) 后，可以对 AI 说这些话来触发它
Help me get started with Karpathy Guidelines (multica-ai)
Explains what Karpathy Guidelines (multica-ai) does, walks through the setup, and runs a quick demo based on your current project
Use Karpathy Guidelines (multica-ai) to a single CLAUDE
Invokes Karpathy Guidelines (multica-ai) with the right parameters and returns the result directly in the conversation
What can I do with Karpathy Guidelines (multica-ai) in my ai agent workflow?
Lists the top use cases for Karpathy Guidelines (multica-ai), with example commands for each scenario
将技能文件夹放到 ~/.claude/skills/multica-karpathy-guidelines/ 目录（个人级，所有项目可用），或 .claude/skills/multica-karpathy-guidelines/（项目级）。重启 AI 客户端后，用 /multica-karpathy-guidelines 主动调用，或让 AI 根据上下文自动发现并使用。
Karpathy Guidelines (multica-ai) 支持 Cursor、Claude Code、Codex CLI，可与这些 AI 平台无缝集成，扩展其能力。
Karpathy Guidelines (multica-ai) 可免费安装使用。请查阅仓库了解许可证信息。
A single CLAUDE.md file distilling Andrej Karpathy's observations on LLM coding pitfalls into 4 hard rules: think before coding, prefer simplicity over abstraction, make targeted changes only, and verify by outcome not intent. Packaged as a proper Agent Skill with .cursor/rules, plugin.json, and multi-tool support. 140K stars — the fastest validation that developers already knew AI was doing these things wrong.
I need to work with karpathy and best-practices
Recognizes the task, selects Karpathy Guidelines (multica-ai) as the right tool, and completes it without leaving the chat
Automate my ai agent tasks using Karpathy Guidelines (multica-ai)
Identifies repetitive steps in your workflow and sets up Karpathy Guidelines (multica-ai) to handle them automatically