caveman-compress

To reduce input token costs for markdown memory files, you can compress natural-language artifacts like CLAUDE.md into shorter phrasing while preserving structure. This compression lowers recurring token consumption on every session start without losing critical context.

The safest way to compress markdown context is using validated compression that strictly preserves content inside fenced code blocks, inline backticks, URLs, headings, and file paths. It converts natural language to shorter phrasing while leaving structural elements untouched.

To compress a CLAUDE.md memory file for Claude context optimization, trigger the compression skill and specify the file. The process creates a backup as FILE.original.md, compresses the text, and runs local validation to ensure no structural elements were lost.

Markdown compression does not work on code files and configuration files. The process explicitly skips code and config files, applying only to natural-language memory artifacts like preferences and todos to prevent accidental syntax corruption.

You can restore the original content because the compression process automatically writes a human-readable backup as FILE.original.md before making any changes. This backup safety mechanism ensures you never lose your initial memory file content.