What problem does it solve? Large project memory files like CLAUDE.md, todo lists, and preference notes are loaded into every AI session, repeatedly consuming input tokens. This Skill rewrites supported natural-language files into a terse caveman style while preserving code blocks, URLs, headings, and file paths, and keeps a human-readable backup for editing. ## Core Features & Use Cases - Token Reduction: Compresses .md, .txt, .rst, .typ, .typst, and .tex files by removing filler words, hedging, and redundant phrasing, averaging about 46% fewer tokens on benchmark fixtures. - Structural Validation: Verifies that headings, fenced and indented code blocks, inline code, URLs, and file paths survive compression exactly, with up to two targeted fix retries before aborting safely. - Safe Backups: Stores the original as a byte-for-byte .original.md backup in an out-of-tree data directory so skill auto-loaders never re-ingest it, and refuses sensitive filenames like credentials or secrets. - Use Case: A developer with a 1,000-token CLAUDE.md runs the skill once; every subsequent session loads the compressed version, cutting recurring input token cost while the editable original stays in the backup directory. ## Quick Start Ask the AI to run /caveman-compress on your CLAUDE.md file to shrink it and create a readable backup.
