If you're building skills for Claude, this is your starting point. It walks through the entire structure: what goes in SKILL.md versus bundled scripts versus reference docs, how to write frontmatter that actually triggers when it should, and when to lock down behavior with scripts versus leaving room for the agent to adapt. The progressive disclosure section is especially useful, it explains the three-level loading system and why you'd split a 600-line skill file into smaller references. Treats the context window like a scarce resource, which it is. The examples show real patterns like splitting PDF docs by feature and organizing multi-domain skills so you don't load irrelevant context. Worth reading before you write your first skill or if yours keeps firing at the wrong times.
npx -y skills add tldraw/tldraw --skill skill-creator --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
anthropics/skills
obra/superpowers
github/awesome-copilot