all products
Compression
Caveman
Why use many token when few token do trick
The viral Claude Code skill (also Codex, Gemini, Cursor, Windsurf, Cline, Copilot and 30+ more) that makes your agent talk in terse, fragment-based language — cutting output tokens ~65% while keeping every byte of technical meaning. Code, commands and error strings stay verbatim; it compresses style, never substance.
curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bashGitHub72.8kCompress anything, live
caveman
And update user authentication module it validates email address it saves record to database log errors process
tokens
2856
saved
50%
illustrative estimate · ~4 chars/token
What it can do
- ~65% output-token reduction with no loss of accuracy (10-test avg, range 22–87%)
- Four intensity levels: lite, full, ultra, and wenyan (classical Chinese)
- Works across 30+ coding agents from one skill
- Commands: /caveman, /caveman-commit, /caveman-review, /caveman-stats, /caveman-compress
- Code, paths and identifiers preserved exactly — compresses prose only
- language
- Claude Code skill
- license
- MIT