caveman-compress

To reduce input tokens in markdown memory files, you can compress natural language prose into a caveman format. This process preserves code blocks, URLs, and headings while overwriting the original file and saving a human-readable backup.

Caveman compression is an automated token reduction technique for memory files like CLAUDE.md and todos. It compresses natural language prose while skipping code and configuration sections, validating results and retrying to ensure content fidelity.

To compress a memory file, run the command caveman-compress <filepath> in your CLI. The script enforces a 500KB size limit, compresses the prose, saves a backup as FILE.original.md, and overwrites the original file.

Yes, memory file compression preserves code blocks, URLs, headings, and other critical structural elements. It specifically targets natural language prose for compression and skips code or configuration files to maintain technical integrity.

When compressing markdown memory files, the CLI enforces a strict 500KB size limit. A human-readable backup is automatically generated as FILE.original.md before the original file is overwritten with the compressed version.