Agent Skills package domain knowledge into folders the agent can discover and use. Cursor implements the open Agent Skills standard (agentskills.io) — the same format used by other AI coding tools, so skills are portable across ecosystems.
.cursor/skills/deploy-checklist/ ├── SKILL.md # required: instructions + YAML frontmatter ├── scripts/ # optional: executable helpers ├── references/ # optional: docs loaded on demand └── assets/ # optional: templates, images
--- name: deploy-checklist description: Pre-deployment verification steps for this repo --- # Deploy Checklist 1. Run the full test suite 2. Verify env config against .env.example 3. Build locally before pushing to the platform
The frontmatter also supports paths globs (auto-activate the skill for matching files) and disable-model-invocation (make it slash-command-only). Skills live at project level (.cursor/skills/, .agents/skills/) or user level (~/.cursor/skills/), with nested directories for monorepo scoping — and Cursor even reads skills from compatible tools' directories.
Type / in the agent input to invoke built-in skills:
| Command | What It Does |
|---|---|
/create-skill, /create-rule, /create-hook | Scaffold a new skill, rule, or hook |
/review, /review-security, /review-bugbot | In-editor code review before you push |
/loop | Repeat a prompt on a schedule until a condition is met |
/split-to-prs | Break a big diff into reviewable stacked PRs |
/automate | Create an always-on Automation (see Cloud Agents) |
/migrate-to-skills | Convert legacy rules/commands into the skills format |
Since Cursor 3.8, the Customize page is one place to manage plugins, skills, MCP servers, subagents, rules, commands, and hooks — at user, team, or workspace scope — plus a marketplace with a leaderboard of popular extensions.