Unlock the Guide

Gimme rock.Me give guide.

Drop your name and email. We'll send you the install command plus a few other Claude skills worth trying.

No spam. Unsubscribe whenever. We use your email to send you the guide and occasional AI automation insights.

Caveman Skill

ME TALK CAVEMAN.CLAUDE SAVE TOKEN.

A Claude Code skill that strips the fluff from Claude's responses. 25K stars on GitHub. One line install. Same answer, way fewer tokens.

Real Token Comparison

Normal Claude

Caveman Claude

SAVE 0 TOKENS

What it does

Strip the fluff. Keep the answer.

The Caveman skill forces Claude to drop filler words, hedging, and pleasantries. No "I'd be happy to help." No "let me summarise." Just the answer. Same technical accuracy, way fewer tokens. See the difference for yourself...

Normal Claude

"I'd be happy to help you with that. Let me summarise what I just did. I executed the file search and identified three matching results for your query."

38 tokens

Caveman Claude

"Found 3 files. Done."

5 tokens

Try It

The Caveman Translator

Tap a level. See how Claude would answer at that intensity.

Claude's Response

"New object ref each render. Inline object prop = new ref. Wrap in useMemo."

Tokens

72%

Saved

100%

Accuracy

Install

One line. Done.

STEP 01

Install via Claude Code plugin (recommended)

Open your terminal. Run these two commands. Auto-activation is built in.

claude plugin marketplace add JuliusBrussee/caveman

claude plugin install caveman@caveman

STEP 02

Or one-line install for any agent

Works with Cursor, Windsurf, Codex, Copilot, Cline, Gemini CLI, and 40+ more. Auto-detects your agent.

npx skills add JuliusBrussee/caveman

For Cursor specifically: npx skills add JuliusBrussee/caveman -a cursor. Same pattern for windsurf, copilot, cline, etc.

STEP 03

Pick your intensity level

Six levels available. Default is Full. Switch anytime with /caveman ultra in Claude Code.

Lite

Light trim. Pleasantries gone.

Full (default)

Telegraphic. Sentence fragments.

Ultra

Maximum compression. Caveman speech.

Wenyan

Classical Chinese mode. Three flavours.

WORKS WITH

Most coding agents on the market

Claude Code

Codex

Cursor

Windsurf

Copilot

Gemini CLI

The Science

Less padding. Fewer mistakes.

arXiv 2604.00025

Brevity Constraints Reverse Performance Hierarchies in Language Models

MD Azizul Hakim · March 11, 2026

Tested 31 LLMs across 1,485 problems. Forcing large models to give brief responses improved accuracy by 26 percentage points on certain benchmarks. Mechanism: large models talk themselves into wrong answers via overelaboration. Brevity strips that out.

Read the paper →

GO INSTALL CAVEMAN

Open the GitHub repo. Copy the install command. You'll be running caveman mode in under a minute.

Open the GitHub repo Read the brevity research paper