Home · Cursor Skills
Cursor skills: what they are, where they live, and which ones are worth installing
Skills turn Cursor's agent from a generalist into a specialist — one folder, one SKILL.md, loaded exactly when a task needs it. This guide covers the mechanics, the install locations everyone gets wrong, and the part nobody else checks: whether a skill actually works and whether it's safe.
What a skill is made of
A skill is a folder. At minimum it contains SKILL.md: YAML frontmatter with a name and a description, followed by Markdown instructions. Well-built skills add reference files the agent reads only when needed, and sometimes scripts. The description is the trigger — when your request matches it, the agent loads the full instructions; until then the skill costs almost no context.
The same format works across Claude Code, Cursor and Codex — a skill written for one usually runs in the others, which is why our reviews test compatibility instead of treating each platform as a separate world.
Where Cursor skills live
| Location | Scope |
|---|---|
.cursor/skills/ in your repo | Project — shared with everyone on the project |
~/.cursor/skills/ | Personal — every project on your machine |
.agents/skills/, .claude/skills/, .codex/skills/ | Also discovered by Cursor — handy for shared team skills |
Step-by-step with troubleshooting: How to add skills to Cursor →
Full path reference: Where are Cursor skills stored? →
Rules vs skills
Rules (.cursor/rules) are always on: project conventions the agent should never forget. Skills are on demand: deep instructions for one kind of task, kept out of context until the moment they're relevant. If you put task expertise in a rule, you pay for it on every request; if you put project conventions in a skill, they may never load. Use both, for different jobs.
The full comparison, with the two mistakes that waste the most context: Cursor rules vs skills →
Cursor skills we've actually tested
Every score comes from a live standard-task run, with and without the skill, under our published methodology. Install counts and stars never add points.
Before you install anything: the safety problem
A skill is instructions your agent will follow — and bundled scripts it will run. Independent scanning (Snyk, ToxicSkills) found security flaws in 36.82% of 3,984 public skills, including confirmed malicious ones. Our reviews screen every skill against six red lines — credential theft, remote code execution, data exfiltration, hidden instructions, obfuscated payloads, privilege abuse — before it can be recommended. How the screen works →
Frequently asked questions
What are Cursor skills?
Cursor skills are reusable instruction packages built around a SKILL.md file. Each one teaches the Cursor agent how to do a specific kind of task — design a page, process a PDF, run a test suite — and loads only when a request matches its description, or when you invoke it by name.
Where do Cursor skills go?
Project skills live in .cursor/skills/ inside your repository and are shared with the project. Personal skills live in ~/.cursor/skills/ and apply everywhere. Cursor also discovers skills in .agents/skills/, .claude/skills/ and .codex/skills/, so most Claude Code and Codex skills work unchanged.
How do I use a skill in Cursor?
Two ways: describe the task and let the agent pick the skill whose description matches, or invoke it directly in chat with /skill-name. Automatic triggering depends entirely on how well the skill description is written — which is one of the things our reviews score.
Are Cursor skills safe to install?
Not automatically. A skill is instructions — sometimes with scripts — that your agent will follow and run. Snyk’s ToxicSkills scan of 3,984 public skills found security flaws in 36.82% of them. Check what a skill asks for before installing it; every review on this site includes a security screen against six red lines.
What is the difference between Cursor rules and skills?
Rules (.cursor/rules) are always-on conventions for a project — style guides, architecture decisions, do-and-don’t lists. Skills are on-demand expertise for a task, loaded only when relevant so they don’t consume context on every request. Most projects want a small set of rules plus a toolbox of skills.