Skill Best Practices

If you're building skills for AI agents, this walks you through the full authoring workflow: validation scripts that catch metadata errors before you commit, a clear directory structure (scripts, references, assets), and progressive disclosure rules to keep your main logic under 500 lines. It enforces third-person imperative style and includes a Python validator you run inline. The bundled path rule is smart: it makes you resolve helper script paths relative to the skill directory so agents don't silently depend on working directory state. It's opinionated about what belongs in a skill versus general docs, which keeps things focused. Worth using if you're shipping reusable agent tooling and want consistency across a catalog.

Install to Claude Code

npx -y skills add pedronauck/skills --skill skill-best-practices --agent claude-code