create-agent-skills

To structure Claude Agent Skills with XML-first prompts, author a SKILL.md file containing YAML frontmatter for discovery and a pure XML body with sections like objective, quick_start, and success_criteria for maintainable execution.

Progressive disclosure in agent skill authoring is the practice of keeping the main SKILL.md file concise by offloading extended context, workflows, and advanced features into separate reference files and external scripts.

Add YAML frontmatter to your SKILL.md file by including mandatory fields like name and description at the top, ensuring the agent skill is reliably discovered and indexed across tasks.

Yes, you can organize complex agent skills by separating advanced features, workflows, and context into external reference files, keeping the core XML body concise while supporting safety, testing, and deployment workflows.

When creating agent skills, follow validation practices that check XML structure, directory naming conventions, frontmatter fields, and safety guidelines to ensure skills remain discoverable and maintainable.

Yes, this XML-first skill authoring approach supports both single concise skills and complex multi-file skills by scaling progressive disclosure through reference files, external scripts, and structured templates.