caveman-compress

To reduce token consumption for CLAUDE.md memory files, compress the natural language content to shrink file size by an average of 46% while preserving code blocks, URLs, and document structure.

Yes, you can compress markdown and text files without losing code blocks or file paths. The compression process automatically skips code and configuration files, preserving all commands and document structure exactly.

Compressing project memory files automatically saves a human-readable backup of the original content as FILE.original.md before overwriting, ensuring you can edit the original and re-run compression at any time.

The best way to cut recurring input token overhead for preference documents is to compress their natural language text. A 1000-token memory file saves nearly half its tokens per session, accumulating 100,000+ tokens saved across 100 sessions.

You should not use natural language file compression for code or configuration files, as the compression process is designed to automatically skip these formats and only targets human-readable project memory documents.