Karpathy-Inspired Skills
Fourteen skills with working habits distilled from Karpathy's public posts, plus a router skill, for Claude Code and Codex
- 124 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- Skills give instructions to the agent and run nothing themselves apart from `install.sh` at install time.
- Some skills advise running experiments and tests, which the agent executes with your permissions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/karpathy-inspired-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- Skills give instructions to the agent and run nothing themselves apart from `install.sh` at install time.
- Some skills advise running experiments and tests, which the agent executes with your permissions.
This is third-party code. Review the repository files before installing.
What it does
A set of 14 skills and one router, `karpathy-methodology`, in which the author distils working habits from Andrej Karpathy's public X posts and Gist. Each skill gives trigger scenarios, steps and acceptance criteria: `agentic-engineering` asks for one reviewable change per round, `minimalism` doubts every new dependency, `llm-wiki` maintains a Markdown wiki with a model. Four workflows are assembled from the skills and are invoked with a phrase like "use the Karpathy workflow...". Codex, Claude Code, OpenCode, OpenClaw and Hermes are supported. The README and skill texts are in Chinese.
Who it is for. Developers who want an agent bound to small verifiable steps and can read Chinese documentation.
Good fit when
- An agent writes oversized changes and needs a one-verifiable-step loop.
- Before adding an npm package the team checks whether the dependency is justified.
- A vibe-coded prototype is moved into a ship-ready phase.
- Notes and sources accumulate in a Markdown wiki maintained by a model.
Not a fit when
- You need Karpathy's literal method: the set is the author's reinterpretation, not his work.
- The team does not read Chinese and will not translate instructions.
- One narrow skill is enough and the full set would bloat the agent's skill list.
Example request
Use the Karpathy workflow: first a vibe-coded prototype, then bring it to release in small verifiable steps.Limitations
This is a study interpretation of Karpathy's public material; the rights belong to him, and he has not endorsed the set. The README and texts are in Chinese. The last commit is dated July 2026. Some skills fire on phrases like "how would Karpathy do it", which can cause false triggers; skip skills you do not need. Check paths and the agent list in `install.sh`: the script accepts `--target` and `--dry-run`.
How to disable. Delete the `karpathy-*` folders from the agent skills directory, for example `~/.claude/skills/`. Codex and other agents use their own paths.
FAQ
How do I install the Karpathy skill set for Claude Code?
Run `npx skills add LearnPrompt/andrej-karpathy-skills` or `bash install.sh --target claude`. The script supports `--dry-run` to preview without writing.
Where do I start if I do not know which skill fits?
With `karpathy-methodology`: it acts as a router and picks a suitable one of the 14 skills for your task.
Which agents are supported?
Codex, Claude Code, OpenCode, OpenClaw and Hermes. `install.sh` has a `--target` value for each.
The skills are in Chinese: do they suit a non-Chinese team?
Texts and the README are in Chinese, so instructions need translating. An agent that handles Chinese executes them as written.
How do I preview what the script installs without writing files?
Run `bash install.sh --dry-run`. The `--target` option picks the agent: `codex`, `claude`, `openclaw`, `hermes`, `opencode` or `all`.
Discussion
No questions or reviews yet.
Related
SkillsA hundred ready loops for Claude Code, Cursor and Codex: the agent acts, runs a check and fixes the cause until the goal is met.
A working folder for Claude Code where the agent asks questions and writes a new skill by Anthropic's guidance, not from memory.
Five skills by Dan McAteer: Claude Code mods, supervised Codex workflows, task handoff and deeper prompts.
A set of ready-to-copy Claude Code skill folders: PDF, slides, Next.js, image and video generation and social publishing.
A catalog of nearly two hundred skills: code delivery, review, debugging, AI governance, product work and agent coordination.
11 skills for Claude Code: SKILL.md authoring, prompt tuning, refactoring, change review and planning in files.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Karpathy-Inspired Skills (this card) | no | 124 | |
| Karpathy-inspired Claude Code guidelines | no | 232,688 | that is a compact charter of rules for CLAUDE.md; here 14 separate skills and four workflows |
| Karpathy LLM Wiki | no | 2,732 | that card installs a single LLM wiki with MCP; here the wiki is only one of 14 skills |
README in short
The skill table links to the source posts and Gist. Four workflows and the trigger phrases to call them are described alongside.
README badge
Original SKILL.md
--- name: karpathy-methodology description: 'Apply Andrej Karpathy AI methodology and principles from his 2023-2026 insights. Use this skill when the user wants to apply Karpathy-style thinking, needs guidance on agentic engineering, LLM knowledge bases, minimalist coding, understanding-first principles, or any of the 14 core methodologies distilled from his work. Also triggers on: "karpathy method", "karpathy approach", "karpathy way", "AK style", "like karpathy does", "karpathy principles", or when user asks how Karpathy would approach any programming or AI problem.' disable-model-invocation: false user-invocable: true related_skills: - karpathy-agentic-engineering - karpathy-llm-simulator - karpathy-llm-wiki - karpathy-autoresearch - karpathy-meta-reflection --- # Karpathy Methodology — 14 Core Skills > Distilled from Andrej Karpathy's most-liked posts on X (2023–2026), his LLM Wiki Gist, and the multica-ai/andrej-karpathy-skills repo. > Source: https://x.com/karpathy | https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f This skill is the **master index**. When invoked, either: 1. Apply the specific sub-methodology relevant to the user's current task, or 2. Guide the user to the right one from the 14 below. --- ## The 14 Methodologies | # | Skill Name | One-line Essence | |---|-----------|-----------------| | 1 | `karpathy-agentic-engineering` | Orchestrate agents with clear tasks and verifiable success criteria | | 2 | `karpathy-llm-wiki` | Let LLM maintain your knowledge base; you do the exploring | | 3 | `karpathy-llm-simulator` | Ask LLM to simulate expert debate, not give one opinion | | 4 | `karpathy-minimalism` | 200 lines of pure Python beats 50 npm packages | | 5 | `karpathy-vibe-to-agentic` | Vibe coding raises the floor; agentic engineering raises the ceiling | | 6 | `karpathy-autoresearch` | Agent loops on git branches; you only change the prompt | | 7 | `karpathy-output-evolution` | Text → Markdown → HTML → rendered; always demand structure | | 8 | `karpathy-understanding-first` | Outsource thinking, never understanding | | 9 | `karpathy-idea-files` | Ship the Gist (abstract + criteria), not the code | | 10 | `karpathy-meta-reflection` | Monthly audit: what's atrophying, what's exploding | | 11 | `karpathy-supply-chain-hygiene` | Every dependency is an attack surface | | 12 | `karpathy-education-first` | Make everything teachable; nano-projects over monoliths | | 13 | `karpathy-system-prompt-learning` | Write strategy into the system prompt like a textbook | | 14 | `karpathy-practice-environments` | Build gyms where agents can try, fail, and learn | --- ## Meta-Principle (the one that unifies all 14) > "You can outsource your thinking but you cannot outsource your understanding." > — Karpathy, ~46k likes, 2026