karpathy-guidelines

multica-ai/andrej-karpathy-skills · Agent Skill

Provides behavioral guidelines to minimize common LLM coding mistakes, promoting clarity and efficiency.

Jump to install ↓

Overview

Publishermultica-aiOrganization
Repositorymultica-ai/andrej-karpathy-skills ↗
Skill folderkarpathy-guidelines
Stars210,607
Bundled files1
Licensenot specified — check with the author
Last updatedJanuary 27, 2026
LinksSkill source ↗ · Repository ↗

Mozgi check

What we found in the skill files. Details come with a subscription.

Security
low risk · no obvious risks found — the full report shows the details
Compatibility
instructions only — works with most agents
Similar skills
144 similar skills found
Also with subscription

Installation

A skill is a folder with a SKILL.md file. Pick your agent and run the commands in a terminal (macOS, Linux or Git Bash on Windows) from your project root.

Copy the prompt into Claude, Codex or another assistant: it will review the skill and install it for you.

Install the “karpathy-guidelines” skill from mozgi.io (https://mozgi.io/skills/karpathy-guidelines/) for me.

1. Download the archive: https://mozgi.io/skills/karpathy-guidelines/download
   If it fails, take the skill folder from the source: https://github.com/multica-ai/andrej-karpathy-skills/tree/main/skills/karpathy-guidelines
2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me.
3. Put the “karpathy-guidelines” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.).
4. Make sure <skills folder>/karpathy-guidelines/SKILL.md exists.
5. Briefly tell me what this skill does and offer to try it on my task.

If you cannot do a step yourself, tell me what to click or type.

Review the source files

1 file Download Zip

Read SKILL.md and any companion files before deciding whether to install. Click a file to preview it.

SKILL.md contents

Original author text · read-only

Karpathy Guidelines

Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations ↗ on LLM coding pitfalls.

Tradeoff: These guidelines bias toward caution over speed. For trivial tasks, use judgment.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

  • State your assumptions explicitly. If uncertain, ask.
  • If multiple interpretations exist, present them - don't pick silently.
  • If a simpler approach exists, say so. Push back when warranted.
  • If something is unclear, stop. Name what's confusing. Ask.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

  • No features beyond what was asked.
  • No abstractions for single-use code.
  • No "flexibility" or "configurability" that wasn't requested.
  • No error handling for impossible scenarios.
  • If you write 200 lines and it could be 50, rewrite it.

Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

  • Don't "improve" adjacent code, comments, or formatting.
  • Don't refactor things that aren't broken.
  • Match existing style, even if you'd do it differently.
  • If you notice unrelated dead code, mention it - don't delete it.

When your changes create orphans:

  • Remove imports/variables/functions that YOUR changes made unused.
  • Don't remove pre-existing dead code unless asked.

The test: Every changed line should trace directly to the user's request.

4. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

  • "Add validation" → "Write tests for invalid inputs, then make them pass"
  • "Fix the bug" → "Write a test that reproduces it, then make it pass"
  • "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a brief plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.

Frequently asked questions

What does the karpathy-guidelines skill do?

Provides behavioral guidelines to minimize common LLM coding mistakes, promoting clarity and efficiency.

How do I install karpathy-guidelines?

Open the “Installation” block on this page and pick your agent — Claude Code, Codex, Cursor or Gemini CLI — for ready-made commands that copy the skill folder to the right place. Or copy the prompt from the “Via assistant” tab and let the assistant review and install the skill.

Which AI models work with karpathy-guidelines?

Skills are plain Markdown instructions (the Agent Skills format), so karpathy-guidelines works with any agent that reads SKILL.md: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and more. The model loads the instructions only when a task matches the skill's description.

Is karpathy-guidelines free?

The skill is published openly by multica-ai. The skill itself costs nothing — you only pay your AI provider for the tokens you use. No license is specified in the repository — check the terms with the author before commercial use.

Is it safe to install skills?

A skill is a set of instructions your agent will follow, so read SKILL.md and its companion files before installing. Scripts in the skill folder only run if your agent runs them — keep an eye on what it does and do not grant unnecessary permissions.

How is a skill different from an MCP server or a plugin?

An MCP server or plugin gives the model new tools — code that runs and returns a result. A skill gives it knowledge and process: how to approach a task, which steps to follow and what good output looks like. It needs no server or API key.

Related skills

All skills →
agent-transcript Adds a redacted agent transcript to GitHub PR or issue bodies within OpenClaw agent-created workflows. openclaw389k autoreview Automates code review using various AI models, providing structured reports and recommendations. openclaw389k to-issues Breaks a plan or PRD into independently actionable issues. mattpocock304k improve-codebase-architecture Scans a codebase for deepening opportunities and turns them into a prioritized visual report. mattpocock304k triage Moves issues through a structured triage state machine. mattpocock304k grill-with-docs Sharpens the domain model and updates supporting docs while grilling through design choices. mattpocock304k