This is the meta-skill for building skills themselves. It walks you through the structure of SKILL.md files, the progressive disclosure pattern (metadata loads first, body loads when triggered, bundled resources load as needed), and how to balance specificity with context efficiency. The core insight is useful: treat Claude like it's already smart and only add what it genuinely can't infer. Includes practical guidance on when to use scripts versus references versus assets, and emphasizes keeping the main file under 500 lines by splitting content into reference docs. If you're publishing skills or building internal tooling that extends Claude, this codifies the design patterns worth following.
npx -y skills add thinkinaixyz/deepchat --skill skill-creator --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
supercent-io/skills-template
supercent-io/skills-template
huangjia2019/claude-code-engineering
remotion-dev/remotion
reactjs/react.dev