skill-optimizer

To reduce token usage in Claude Code skills, you can modularize them using Progressive Disclosure Architecture (PDA), which splits monolithic files into an orchestrator and reference components for on-demand loading. This approach cuts token consumption by 80-95%.

Progressive Disclosure Architecture (PDA) is a modularization pattern that breaks monolithic skills into an orchestrator plus reference and scripts directories, loading content on-demand to minimize token costs while preserving original functionality.

To refactor a monolithic skill, split it into an orchestrator and reference files, extract scripts into dedicated directories, and optimize metadata for better discovery. This on-demand loading architecture reduces tokens and improves maintainability.

Yes, skill metadata optimization improves frontmatter for better discovery and trigger coverage. The modularization process enhances metadata alongside splitting monolithic files into orchestrator and reference components to preserve functionality.

Yes, modularizing skills preserves original functionality. The refactoring process splits monolithic files into an orchestrator and reference components using PDA, ensuring on-demand loading reduces token costs without losing features.

You should modularize a Claude Code skill when it becomes large or bloated. Applying PDA by splitting into an orchestrator and reference files reduces token usage by 80-95% and improves maintainability for complex skills.