creating-agent-skills

To create Claude Code Skills, use a structured blueprint that embeds core rules in SKILL.md and structures content with progressive disclosure to ensure consistent discovery, activation, and execution across models.

The router pattern in skill authoring directs tasks to specific references or scripts only when needed. You should use it to audit and upgrade existing skills for better structure and maintainability across models.

Progressive disclosure structures Claude skills by separating content into SKILL.md, references, and scripts, loading only what is needed for a specific task to maintain efficiency and reduce context overhead.

The best way to validate and audit existing Claude skills is using provided templates and validation scripts to standardize outputs, checking for proper XML structure and adherence to progressive disclosure principles.

You should inline essential principles directly in SKILL.md for immediate enforcement, while using references and scripts for deeper content to maintain progressive disclosure and optimal loading behavior.

Claude skills fail to activate consistently when they lack proper progressive disclosure and router patterns, making them poorly structured for discoverability and execution across varying models and tasks.