🪨 caveman-skill

why use many word when few word do trick

The only caveman skill built for claude.ai — web and desktop.

Not Claude Code. Not a terminal plugin. The actual chat interface you use every day.

---

The Gap

JuliusBrussee's caveman is brilliant — 26k+ stars, viral on HN, ThePrimeagen covered it. But it only works in Claude Code (the CLI tool developers run on their machines).

Nobody built this for claude.ai. Until now.

---

What It Does

Activates caveman mode inside claude.ai chat. Three intensity levels. Slashes output verbosity by up to 87% while keeping 100% technical accuracy.

Mode	Command	Token Reduction	Style
LITE	/caveman-lite	~30%	Cut fluff, keep grammar
NORMAL	/caveman	~75%	Full caveman grammar
ULTRA	/caveman-ultra	~87%	Brutal minimum. Grunts.

---

Before / After

Normal Claude (72 tokens):

Sure! I'd be happy to help you with that. The issue you're experiencing is most likely caused by your authentication middleware not properly validating the token expiry. Let me take a look and suggest a fix. Hope that helps!

LITE (38 tokens):

The issue is in your authentication middleware — token expiry isn't being validated correctly. Fix:

NORMAL (11 tokens):

Auth middleware. Token expiry check wrong. Fix:

ULTRA (6 tokens):

auth middleware. expiry < not <=.

---

Install

Step 1 — Download the skill

Download caveman-skill.zip from Releases.

Step 2 — Upload to claude.ai

1. Go to claude.ai
2. Settings → Capabilities → Skills
3. Click "Upload skill"
4. Select the .zip file

Step 3 — Use it

In any conversation, type:

/caveman

Claude confirms: 🪨 caveman mode on.

To deactivate: /normal

---

All Commands

/caveman-lite    → lite mode (~30% reduction)
/caveman         → normal caveman (~75% reduction)
/caveman-ultra   → ultra mode (~87% reduction)
/normal          → back to normal

Also works with plain English:

• "caveman mode", "talk like caveman", "be brief", "less words"
• "stop caveman", "normal mode", "talk normally"

---

Mode Details

🪨 LITE

Removes pleasantries, meta-commentary, hedging, sign-offs. Grammar stays intact. Good for people who want clean answers without going full cave.

🪨🪨 NORMAL

Full caveman grammar. Drops articles, filler, padding. Fragments fine. Short synonyms. Maximum signal.

🪨🪨🪨 ULTRA

Absolute minimum. Symbols over words (→, =, +). Every word must earn its place. One fragment per point. Brutal.

Code blocks are NEVER affected. All modes write clean, full, correct code. Caveman speak only in the English wrapper.

---

Why It Works

A March 2026 arXiv paper — "Brevity Constraints Reverse Performance Hierarchies in Language Models" — found that constraining LLMs to brief responses improved accuracy by 26 percentage points on certain benchmarks.

Less verbose ≠ less intelligent. Often the opposite.

Caveman not dumb. Caveman efficient.

---

Compatibility

Platform	Works?
claude.ai (web)	✅
Claude Desktop (macOS/Windows)	✅
Claude Code	❌ use JuliusBrussee/caveman
Claude API	❌ just add rules to system prompt

---

Credits

Inspired by JuliusBrussee/caveman — the original Claude Code skill that started the caveman movement.

This repo fills the gap: claude.ai and Claude Desktop users who want the same power without touching a terminal.

Built by @Dev Vaham / 3idhmid

---

License

MIT — use it, fork it, make it better.

---

🪨 caveman save token. token save money. money good.