Agent skills
Overview
One markdown file per task, written for your AI to follow.
A skill is a markdown file in agent-skills/ that encodes how to do one
task in this kit correctly: which files to touch, which conventions to keep,
and how to verify the result. Skills are written for AI coding tools, so the
workflow is always the same:
- You say what you want in plain words.
- Your agent matches it to a skill via the table in
CLAUDE.md. - The agent follows the skill and shows you the result.
This works in Claude Code, Cursor, Codex, and any tool that reads project instructions.
What triggers what
| You say something like... | Skill |
|---|---|
| "get this running" | get-it-running.md |
| "add auth" / "Supabase" / "Clerk" | wire-auth.md |
| "add billing" / "Stripe" | wire-billing.md |
| "wire the AI chat" | wire-ai-chat.md |
| "file uploads" / "S3" | wire-file-uploads.md |
| "notifications" / "realtime" | wire-notifications.md |
| "add SEO" / "sitemap" | add-seo.md |
| "add analytics" / "GA4" | add-analytics.md |
| "translate" / "add a language" | add-i18n.md |
| "deploy" / "go live" | deploy.md |
| "make it mine" / "rebrand" | make-it-yours.md |
| "check accessibility" / "perf" / "security" | the audit skills |
The full table lives in CLAUDE.md at the project root, which your agent
reads automatically.
Skills keep sessions consistent
Two different AI tools, or the same tool on different days, produce the same shape of work because the conventions live in the skill, not in the chat.