Skill Creator

To create token-efficient agent skills, author SKILL.md documents using progressive disclosure and a three-level loading system. This structure minimizes token consumption by keeping metadata compact and placing heavy content in lazy-loaded resources.

Progressive disclosure in prompt design is a three-level loading structure: metadata, the SKILL.md body, and lazy-loaded resources. This mechanism minimizes token usage while enabling scalable agent systems to access information only when needed.

Structure references and assets by organizing heavy content into dedicated directories like references/ and assets/. This organization accelerates scalable skill creation across teams by enabling lazy loading and keeping the core SKILL.md body compact.

The best way to reduce token usage in AI skills is applying imperative, token-efficient writing rules within a three-level loading system. Place heavy content in references and use templated resources to maximize information density while minimizing token consumption.

Yes, your agent skill needs structured frontmatter and size limits. Ensuring frontmatter, core workflow, and references are structured with specific size limits minimizes token use and enables efficient lazy loading for scalable agent systems.

Use lazy loading for agent skill resources when heavy content like scripts and references would otherwise bloat the SKILL.md body. This approach minimizes token consumption and ensures scalable agent systems load data only when required.