genshijin-compress

To reduce input tokens for Markdown memory files, you can compress natural-language text into a condensed format. This process rewrites prose-heavy documentation into a smaller representation while preserving technical accuracy and structural elements.

During token compression, structural Markdown elements are preserved exactly to maintain technical correctness. The process safeguards code blocks, inline code, URLs, headings, file paths, document structure, and YAML frontmatter from any modification.

You can compress memory files in place by first creating a human-readable backup. The system writes the original content to a FILE.original.md backup file before overwriting the target file with the compressed output.

Yes, validation is performed after Markdown compression to ensure code blocks, URLs, and file paths remain unchanged. If validation failures occur, the system performs pin-point repairs on the specific errors without fully re-compressing the file.

You cannot use this to compress files containing credentials due to safety boundaries. The system identifies and refuses likely secret or credential files to prevent accidental exposure during memory file compression.

You should compress Markdown files before memory-heavy conversations or PR review contexts to shrink conversational overhead. Compressing files like CLAUDE.md and todos beforehand reduces the input token usage while maintaining technical correctness.