What problem does it solve? Project memory files like CLAUDE.md, todos, and preference notes are loaded into context on every session start, so large files repeatedly consume input tokens. This Skill rewrites those files into a terse caveman style that preserves all technical substance while cutting token count by roughly 46% on average. ## Core Features & Use Cases - Token-reducing compression: Rewrites prose by removing articles, filler, hedging, and pleasantries while preserving code blocks, inline code, URLs, file paths, headings, and commands exactly. - Validation with targeted repair: After compression, a local validator checks headings, code blocks, URLs, paths, bullets, and inline code; failures trigger a targeted fix pass (up to 2 retries) rather than full recompression, and the original is restored if validation still fails. - Safe backup handling: The original is saved as FILE.original.md in an out-of-tree data directory ($XDG_DATA_HOME/caveman-compress/backups or %LOCALAPPDATA%) so auto-loaders never re-ingest it, with atomic writes and backup readback verification. - Use Case: Run it on a 700-token CLAUDE.md preferences file and get a ~285-token version that Claude reads every session, with the human-readable original kept safely in the backup directory for future edits. ## Quick Start Ask the assistant to run /caveman-compress on your CLAUDE.md file to shrink it and save tokens in every future session.
