Guides
aaaa.fyi · 2026-07-22 · 8 min read
A Claude skill is a folder containing a SKILL.md file — a markdown document that teaches Claude Code how to perform a specific task: reviewing code against your standards, generating PDFs, debugging systematically, designing frontends. When your request matches a skill's description, Claude loads its instructions into context and follows them.
Think of skills as installable expertise. The model already knows how to write code; a skill tells it exactly *how your team wants it done*.
my-skill/ SKILL.md ← required: frontmatter + instructions reference.md ← optional supporting files scripts/ ← optional helper scripts
The SKILL.md frontmatter carries a name and a description. The description matters most: Claude uses it to decide when the skill applies. The body holds the actual instructions.
One command:
npx skills add <owner>/<repo> --skill <name>
Project-level skills land in .claude/skills/ (shared with your team through git); global skills live in ~/.claude/skills/ and apply to every project.
Install counts are the honest signal. Our skills directory tracks live install numbers for over a thousand skills — the same numbers the skills CLI reports — plus stars and last-push activity from GitHub. Start with the most-installed skills for code review, debugging and your framework, then branch out.
Skills change how Claude *thinks* (instructions); MCP servers change what Claude *can touch* (browsers, databases, APIs). They compose: a skill can instruct Claude on how to use an MCP server well.
Do skills work outside Claude Code? Yes — Cursor, Codex and other agents read the same format via the skills CLI.
Do skills cost tokens? Only when triggered. An installed-but-unused skill costs just its one-line description.
Are skills safe? A skill is plain text — read it before installing. Thirty seconds of reading tells you exactly what it will push your agent to do.