caveman-compress

To reduce Claude token usage, you can compress markdown memory files like CLAUDE.md. This process preserves all technical substance, code, URLs, and structure while shrinking the overall token count for faster loads.

Markdown compression does not remove code blocks or URLs from your memory files. It preserves all technical content and structural elements while compressing the natural language text to lower token counts.

To compress markdown files for Claude, you need Python 3.10 or higher and access to the Claude API. The skill uses the API for primary compression and Python's subprocess module as a fallback.

You compress a CLAUDE.md file by running a compression skill that validates output integrity to ensure no technical context is lost. It also creates a human-readable backup of the original file automatically.

Large memory files cause slow session loads in Claude due to high token counts. Compressing the natural language markdown content reduces token usage, which directly lowers costs and improves loading performance.

The Python subprocess fallback for token compression may lack the contextual nuance of the primary API method. However, the skill validates the final output integrity to ensure the compressed file remains accurate.