---
title: "The Caveman skill"
description: "A free, open-source skill that makes your coding agent answer in fewer words, with code, commands and errors kept exact. Works in Claude Code, Codex and 30+ agents. Free and open source (Apache-2.0), and it needs no account."
canonical: https://caveman.so/skill
last-updated: 2026-11-21
---

# The Caveman skill

A free, open-source skill that makes your coding agent answer in fewer words, with code, commands and errors kept
exact. Works in Claude Code, Codex and 30+ agents. Free and open source (Apache-2.0), and it needs no account.

## Install

```sh
npx skills add JuliusBrussee/caveman -g
```

- GitHub: https://github.com/JuliusBrussee/caveman (more than 120,000 stars)

## Sample

Why is my React component re-rendering?

- Normal reply (about 72 tokens): The reason your React component is re-rendering is likely because you’re creating
  a new object reference on each render cycle. When you pass an inline object as a prop, React’s shallow comparison
  sees it as a different object every time, which triggers a re-render. I’d recommend using useMemo to memoize the
  object.
- With the Caveman skill (about 21 tokens): New object ref each render. Inline object prop = new ref = re-render.
  Wrap in useMemo.

## How many tokens it saves

Replies get shorter. Most of an agent's bill is what it reads, and the skill doesn't change that.

- **Shorter answers than asking for brevity.** Against a plain "Answer concisely." prompt: 3% shorter at the median,
  35% at the ultra level. 10 prompts, one run.
- **Code stays exact.** The skill tells your agent to leave code, commands and error text exactly as they are. Only
  the words around them get shorter.

More: [Does the skill save money?](https://caveman.so/news/does-the-caveman-skill-save-money)

## Contributors

It reached #1 on Hacker News with 904 points, and the work on it happens in the open on GitHub. Engineers at these
companies have filed issues and opened pull requests on the Caveman skill: OpenAI, Microsoft, Amazon, Vercel,
Supabase, Intel, LinkedIn, Cisco, SAP, Red Hat, Tencent, Booking.com, Akamai, Comcast, Yandex, Canonical, Coursera
and Two Sigma.

- Pull requests: https://github.com/JuliusBrussee/caveman/pulls

## Questions

### Is the skill free?

Yes. It is free and open source under Apache-2.0, and it needs no account.

### Which agents does it work with?

Claude Code, Codex, Gemini, Cursor and 30+ more, all from the one install command.

### Will it cut my bill?

Not by much on its own. It makes replies shorter, but an agent's bill is mostly what it reads: input and cache re-reads, which a reply style can't touch, and the rules add a few input tokens of their own. If your own before-and-after comes out higher, turn it off for that work.

### Does it change my code?

No. Code, commands and errors stay exact, byte for byte.

### Can I make it stronger or lighter?

Yes, with four levels: lite, full, ultra and wenyan, which answers in classical Chinese.

### How is it different from Caveman Code?

The skill is a free tool. Caveman Code is our product: it needs a Caveman account and is free on your own subscriptions and keys.

## Install the Caveman skill

```sh
npx skills add JuliusBrussee/caveman -g
```

[View on GitHub](https://github.com/JuliusBrussee/caveman).
[Caveman Code](https://caveman.so/code) is the coding agent we built after the skill.
