skill-creator

To structure a Claude Code skill, use a SKILL.md file with required YAML frontmatter fields (name and description) and a progressive disclosure layout so the model receives lightweight metadata first, then detailed instructions only when triggered.

To prevent Claude from auto-invoking a skill with risky side effects, include the disable-model-invocation and user-invocable controls in the SKILL.md frontmatter to restrict automatic execution and require explicit user triggering.

Use a Claude Code skill instead of a prompt or CLAUDE.md when you need reusable, project-wide documentation and automated workflows that the model can safely discover and invoke at the appropriate times.

Write effective name and description fields for skill architecture by applying quality constraints, focusing the scope to a specific task, and including at least one example of invocation and expected output to improve model effectiveness.

Yes, you can link external reference files when building Claude Code skills by referencing additional resources such as REFERENCE.md and templates directly within the SKILL.md file for optional resource linking.

Progressive disclosure in Claude Code skill design is an architecture concept where models get lightweight YAML metadata first and detailed step-by-step instructions only when the skill is triggered, preventing context overload.