caveman-compress

Markdown memory files like CLAUDE.md are compressed into a caveman format to reduce loaded input tokens during Claude sessions. The process preserves code blocks, URLs, and file structure while backing up the original file.

Compressing CLAUDE.md while preserving code blocks and URLs requires a deterministic workflow that detects file type, validates output, and retries up to 2 times. The original file is automatically backed up as filename.original.md before compression.

Markdown compression works with local memory files, todos, and preferences across projects. It is applied when opening Claude sessions to lower token costs by converting natural language into a minimized caveman format.

Using an automated workflow that saves the original as filename.original.md is the best way to back up markdown files before token compression. This ensures the readable backup is preserved alongside the compressed version.

Validation errors during markdown file structure compression for token savings may occur, but the workflow retries up to 2 times. The tool enforces a deterministic process to ensure code blocks, headings, and tables are preserved.