What problem does it solve? Large project memory files like CLAUDE.md, todos, and preferences are loaded on every session start, repeatedly consuming input tokens. This Skill compresses those natural language files into a shorter caveman-style format while preserving code, URLs, paths, and structure, cutting repeated token costs. ## Core Features & Use Cases - Token-reducing compression: Rewrites prose by removing filler, hedging, and redundancy while preserving code blocks, inline code, URLs, file paths, headings, and tables exactly. - Validated in-place overwrite: Validates headings, code blocks, URLs, paths, and inline code after compression, applies targeted fixes with up to 2 retries, and restores the original on failure. - Safe backups: Stores the human-readable original as FILE.original.md in an out-of-tree data directory so skill auto-loaders never re-ingest it. - Use Case: Run it on a 1,000-token CLAUDE.md once, and every subsequent session loads the compressed version, saving roughly 46% of those repeated input tokens based on benchmarked fixtures. ## Quick Start Ask the assistant to run /caveman-compress on your CLAUDE.md file to compress it and create a backup of the original.
