compress

To reduce markdown memory file tokens without losing substance, this Skill compresses natural-language prose into a shorter format. It strictly preserves headings, URLs, file paths, and code blocks while converting text into a condensed structure to maximize model input limits.

You need to compress CLAUDE.md notes when maintaining long-running preferences or todos that exceed model input limits. This process shrinks the token count of natural-language memory files, allowing you to fit more context into the available model input window.

Token optimization during markdown compression handles file paths and code blocks by keeping them completely intact. The process validates the compressed output by checking that headings, URLs, code blocks, and path patterns remain structurally preserved.

You cannot safely compress markdown files with sensitive credentials using this approach, as built-in guardrails refuse to process files that appear to contain secrets. This safety mechanism prevents accidental third-party data exfiltration during token optimization.

When markdown compression validation fails, the Skill attempts targeted fixes without full recompression. If repeated validation fails, it restores the original file from a generated FILE.original.md backup to ensure no data is lost.

The best way to automate markdown compression for CLI workflows is using a script-based approach that backs up the original file as FILE.original.md before processing. This ensures token optimization runs safely while preserving technical structure and URLs.