Smart token optimization for your Claude workspace files. Compresses verbose CLAUDE.md files, todos, and preferences into caveman-speak while preserving every code block, URL, and technical term exactly as written. Creates a .original.md backup before overwriting, so you can always recover the human-readable version. The compression is aggressive on natural language but treats anything in backticks as sacred. Worth running on bloated memory files that are eating your context window. Works through a Python CLI that validates output and retries on failures.
npx skills add https://github.com/juliusbrussee/caveman --skill compress