compress

Compress CLAUDE.md memory files to save input tokens by converting natural language into caveman format while preserving technical substance, code, and URLs. The compressed version overwrites the original file, and a human-readable backup is saved as <filename>.original.md.

Caveman format compression for markdown memory files is a deterministic token-optimization process that strips natural language verbosity while retaining all technical substance, code, and structure. It targets files like CLAUDE.md, todos, and preferences to reduce input token consumption.

The compress Skill keeps a backup of original markdown files by saving the human-readable content as <filename>.original.md before overwriting the original file with the compressed caveman format version, ensuring you retain full readability.

The maximum file size limit for compressing memory files is 500KB. The compression process enforces this safety boundary along with sensitive-path checks to prevent processing restricted files, validating output with up to two retries if needed.

You can use the caveman CLI to compress todos and preferences files by running the command with the file path. It applies deterministic caveman compression to natural language content, preserving technical substance while saving input tokens.

Token optimization for memory files matters because reducing input tokens in files like CLAUDE.md lowers processing overhead and cost when using Claude. Compressing natural language into caveman format preserves technical substance while shrinking token count.