creating-agent-skills

Claude Code Skills require frontmatter metadata, XML-structured content, progressive disclosure patterns, and validation workflows. The official specification defines skill authoring through SKILL.md files paired with reference documents, ensuring consistency and reducing token waste across router patterns and skill designs.

Progressive disclosure keeps SKILL.md concise by placing essential guidance in the main file while detailed patterns, workflows, and scripts live in separate references. This structure enables depth without overwhelming authors or auditors, supporting both quick starts and comprehensive skill refinement.

Router patterns suit complex skills with multiple workflows or conditional branches; simple skills work for focused, single-task actions. The skill authoring guidance clarifies when each pattern reduces complexity and improves validation, helping you match structure to task scope.

Validation patterns catch structural errors, inconsistent XML, missing frontmatter, and token waste early. Best-practice guidelines ensure skills meet Anthropic's specification, serve their intended task clearly, and maintain quality across new authored skills and audited existing ones.

Yes. The guidance covers both authoring and auditing, providing checklists and reference templates to assess skill consistency, structure compliance, and adherence to best practices. Auditors use the same specification framework as skill authors.

Include reference files for detailed patterns, workflows outlining task sequences, and scripts demonstrating implementation. This component structure supports progressive disclosure, enables reuse, and clarifies the skill's reasoning logic for both end users and future maintainers.