caveman-compress

Compressing markdown memory files into caveman format reduces per-session token usage when interacting with Claude. It preserves code blocks, URLs, and file paths while overwriting the original with a human-readable .original.md backup.

Compressing CLAUDE.md preserves code blocks, URLs, file paths, and headings exactly while applying deterministic caveman compression to natural language. The process creates a compressed file and saves a backup as FILE.original.md.

Run caveman:compress <filepath> to compress a natural-language markdown file into caveman format. This reduces memory load and token consumption for long-running projects where token efficiency is critical.

Original files are preserved by creating a human-readable backup saved as FILE.original.md before the compressed version overwrites the original. The workflow also skips sensitive filenames and validates results to ensure safety.

The compression workflow enforces safety by validating results, retrying failed compressions, and preserving the original file when necessary. Sensitive filenames are automatically skipped to prevent unintended modifications.

You can compress any natural-language markdown file including CLAUDE.md, todos, and memory notes. The process targets adjacent markdown files and applies caveman compression while maintaining technical substance and structure.