Caveman

Caveman is a skill/plugin for AI coding agents that makes the agent talk like a caveman: it drops filler words and answers in tight, direct 'caveman-speak', while keeping code, commands, and errors byte-for-byte exact. According to the README, this produces the same answers with about 65% fewer output tokens.

It works with Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot, and 30+ other agents. The honest caveat is that Caveman only reduces output tokens; input and reasoning tokens are not reduced, and the skill itself adds roughly 1–1.5k input tokens per turn.

Who it's for

• Developers using AI coding agents who want shorter, more readable replies without losing technical accuracy.
• Anyone paying for output tokens and curious about reducing them, while understanding that the real win is readability and speed, not guaranteed cost savings.
• Users of Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot, and the 30+ other supported agents.

Before / After

Normal agent — 69 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.

Caveman agent — 19 tokens:

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

Normal agent:

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.

Caveman agent:

Bug in auth middleware. Token expiry check use < not <=. Fix:

Same fix. Third of the words. Nothing technical lost.

Core abilities

• Works with a large set of AI coding agents: Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot, and 30+ more.
• Compresses every reply into short caveman-speak while preserving code, commands, and errors byte-for-byte.
• Shrinks what the agent says, not what it knows — the brain stays big, the mouth gets small.
• Keeps your language: write Portuguese, Spanish, French, etc., and the agent responds in that language; it compresses style, not content. The wenyan level is the intentional exception, using classical Chinese because it packs the most meaning per token.
• Includes multiple levels: lite, full (default), ultra, and wenyan.
• Includes commands for commits, reviews, stats, memory-file compression, MCP middleware, and subagents.
• Can rewrite memory files (like CLAUDE.md) so future sessions load a smaller context, saving input tokens every session after.
• No telemetry, no analytics, no accounts, no backend. After install, zero network calls.

What you get

On Claude Code the statusline shows [CAVEMAN] ⛏ 12.4k — that is your lifetime tokens saved, updated on every /caveman-stats. Silence it with CAVEMAN_STATUSLINE_SAVINGS=0.

Levels

Switch anytime with /caveman <level>. The level sticks until you change it or the session ends.

Speak your tongue. Caveman keeps your language. Write Portuguese, caveman grunt Portuguese. Spanish, French, same. It compresses the style, never translates. wenyan mode is the exception on purpose: classical Chinese packs the most meaning per token.

Turn it on / off

• Turn it on: type /caveman or say 'talk like caveman'.
• Turn it off: say 'normal mode'.
• On Claude Code, Codex, and Gemini it is already on from message one. No command needed.

Compatibility and limitations

• Supported agents include Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot, and 30+ others.
• Caveman only compresses output tokens. Input and reasoning tokens are not saved.
• The skill itself adds roughly 1–1.5k input tokens per turn, so whole-session savings are smaller than the per-reply output reduction.
• On already-terse workloads, savings can go net-negative.
• The skill preserves your language; it compresses style, not content. The wenyan level is the intentional exception.
• The README also describes caveman-shrink and cavecrew-* as additional tools for MCP middleware and subagents.

Benchmarks

Real token counts from the Claude API. Average 65% output reduction across 10 prompts (range 22–87%), measured against default verbose replies. Output tokens only.

The README also cites a March 2026 paper, Brevity Constraints Reverse Performance Hierarchies in Language Models, which tested 31 models and found that constraining large models to brief answers improved accuracy by ~26 points on some benchmarks.

Honest number warning. Caveman only shrinks output tokens. Input and reasoning tokens are untouched, and the skill itself adds ~1–1.5k input tokens per turn. So whole-session savings run smaller than the output number, and on already-terse workloads they can go net-negative. The real win is readability and speed. Cost savings are the bonus. See docs/HONEST-NUMBERS.md for when Caveman wins, when it loses, and how to measure it yourself.

caveman-compress receipts

Real memory files, compressed once so every later session is cheaper:

Every session after, that file loads ~46% smaller. Input tokens saved forever, not just one reply.

How it works

1. The skill file is placed into your agent.
2. The skill tells the agent: drop filler, keep substance, use fragments — but never touch code, commands, or errors.
3. On Claude Code, a hook writes a tiny flag file each session, so the agent talks caveman from message one without /caveman.
4. /caveman-stats reads your session log, counts tokens saved, writes the number to your statusline.
5. /caveman-compress rewrites memory files (like CLAUDE.md) so every future session starts with a smaller context. Save tokens forever, not just once.

Privacy

Caveman no phone home. No telemetry, no analytics, no accounts, no backend. After install, zero network calls — the skill is a prompt, the hooks are local scripts, and /caveman-stats reads a log already on your disk. Install-time fetches (GitHub plus your agents' own registries) are spelled out in SECURITY.md.