Caveman

Tue, 09 Jun 2026 00:00:00 +0000

What It Is

Caveman makes AI coding assistants respond in “caveman style” — all filler words removed, only core technical content remains. Average 75% reduction in output tokens, ~3x faster responses. Four compression levels from light deduplication to telegram-style ultra-minimal. Also includes memory file compression that rewrites CLAUDE.md and similar files, cutting ~46% of input tokens.

What’s Actually Different

Most people add “be concise” to their system prompt, but models ignore it frequently. Caveman installs as a Skill and hard-controls output style with guaranteed results. Unique features include /caveman-stats for real-time token/dollar savings tracking, /caveman-compress for rewriting memory files, and cavecrew sub-agents that use ~60% fewer tokens.

Caveman: Cut AI Response Tokens by 75% Without Losing Accuracy

Fri, 05 Jun 2026 00:00:00 +0000

What Caveman Does

It makes AI coding assistants shut up — not by making them dumber, but by stripping filler.

Before Caveman, Claude Code explains a React re-render like this:

“The reason your React component is re-rendering is likely because you’re creating a new object reference on each render cycle. This causes React to see the prop as changed, triggering a re-render…”

After Caveman:

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

Token on Top AI Skills

Caveman

What It Is

What’s Actually Different

Caveman: Cut AI Response Tokens by 75% Without Losing Accuracy

What Caveman Does