caveman-compress

To reduce token consumption of markdown project memory files, compress natural language documentation into a concise format that preserves code blocks and technical specifications while lowering LLM input costs.

Yes, compressing a CLAUDE.md file preserves all code blocks, file paths, URLs, and technical instructions exactly as written, ensuring functional accuracy is maintained alongside token optimization.

The best way to optimize markdown documentation for LLM input costs is to use Python-based validation and Claude-driven summarization to compress natural language text while strictly maintaining structural integrity.

Compressing markdown documentation does not affect technical accuracy because the process strictly preserves all code blocks, file paths, URLs, and technical specifications exactly as originally written.

To start compressing project notes to save tokens every session, run the compress command on your markdown memory file in the terminal to trigger the Python-based validation and summarization process.

Markdown compression targets natural language documentation and project notes, so it should not be used when uncompressed conversational context or verbose explanations are explicitly required for LLM reasoning.