Cursor guide
Best agent Skills for Cursor
The best Skills for Cursor are frontend-design for UI direction, superpowers for multi-step engineering work, and the vendor-published Skills for whatever you integrate against — Stripe, Supabase, Resend, and Prisma all publish their own. 78 of the 191 Skills we catalogue are verified working in Cursor, which makes it the broadest non-Anthropic target; the gap is mostly Skills that depend on Claude-specific tooling.
Updated August 26, 2026 · first published August 19, 2026
Install first
frontend-designThe highest-scoring Skill we catalogue, and verified on Cursor. It is also the best answer to Cursor's characteristic failure mode: fast, fluent completions that produce competent but generic interfaces.
Best when: Building UI where the visual result matters and there is no existing design system to inherit from.
Install model
How Skills install on Cursor
npx skills add github.com/anthropics/skills --skill frontend-design --target cursor- 1Copy the Cursor install command from the Skill page on this site — most use the shared skills CLI with --target cursor.
- 2Run it in the project root so the Skill lands in the repo rather than a global directory.
- 3Reload Cursor so it picks up the new Skill folder.
- 4Trigger it with a task matching its description, and check the output changed — a Skill that does not change behaviour did not load.
Cursor reads SKILL.md through its plugin layer rather than as a first-party primitive, which is why compatibility here is verified per Skill rather than assumed. Skills and Cursor Rules stack rather than compete: Rules are always-on and repo-specific, Skills load on trigger and carry broader method. Where both speak to the same concern, keep the Rule — it is the more specific instruction.
Ranked
The shortlist
Ordered by what to install first for the common case, not by raw score. Every entry here carries verified Cursor compatibility — open the skill page for the composite score, licence, and full install commands.
| Skill | Source | Score | Best when |
|---|---|---|---|
| frontend-design | Official | 92 | Building UI where the visual result matters and there is no existing design system to inherit from. |
| superpowers | Community | 63 | A feature that will span days and files, not a single edit. |
| simplify | Official | 50 | Before every PR, on code you already think is correct. |
| trail-of-bits | Verified | 49 | Code that handles auth, untrusted input, secrets, or money. |
| supabase-skill | Verified | 55 | Building on Supabase — especially row-level security and migrations, where generic advice is actively misleading. |
| webapp-testing | Official | 50 | UI changes where the question is whether the flow still works, not whether the code compiles. |
See every skill verified on this agent, ranked by composite score, on the Cursor page.
Why these
Pick by pick
- 1frontend-design
The highest-scoring Skill we catalogue, and verified on Cursor. It is also the best answer to Cursor's characteristic failure mode: fast, fluent completions that produce competent but generic interfaces.
- 2superpowers
The strongest counterweight to ad-hoc prompting, which Cursor makes very easy. It imposes a plan-then-build pipeline with TDD at each stage, so multi-file work keeps its structure across sessions.
- 3simplify
A quality-only review pass — reuse, simplification, efficiency — that runs on a staged diff. Verified on Cursor, and the cheapest way to stop accreted AI-generated code from becoming the codebase.
- 4trail-of-bits
Security review Skills from a firm that does this for a living, Apache-2.0 licensed. A quality review and a security review are different lenses that miss different things; this is the second lens.
- 5supabase-skill
Published by Supabase, so it encodes their own migration, RLS, and client-usage conventions rather than generic Postgres advice. The pattern generalises: in Cursor, prefer the vendor's Skill for whatever you integrate against.
- 6webapp-testing
Drives a real browser against a running app, so the agent verifies behaviour instead of asserting it. Verified on Cursor, and the fastest way to close the loop between a generated change and evidence it works.
Limits
What this agent cannot do
Cursor is not a native SKILL.md host — it reads Skills through its plugin layer, so each entry on this site lists Cursor only after it was actually tested there. 57 of our 191 Skills are absent from the Cursor list, and the pattern is consistent: Skills that assume Claude-specific tooling, Claude.ai upload flows, or Claude API skill_id references. A Skill that names Claude Code tools in its instructions tends to degrade quietly rather than fail loudly, which is exactly why we do not mark compatibility we have not verified.
FAQ
Common questions
Do Claude Code Skills work in Cursor without changes?
Often, but not always, and the failure is quiet rather than loud. The two required frontmatter fields are the same, so the file parses; what breaks is instruction content that assumes Claude-specific tools or a shell the agent does not have. Our compatibility list reflects Skills actually tested in Cursor.
Skills or Cursor Rules — which should I use?
Both, for different jobs. Rules are always-on, repo-specific, and cheap: use them for conventions that must apply to every edit. Skills load on trigger and can bundle scripts and reference files: use them for methods you invoke deliberately, like a review pass or a document workflow.
Where do Skills live in a Cursor project?
In the project directory the install command targets, which keeps them committed with the repo so the whole team inherits them. Reload Cursor after adding one — like Claude Code, it reads Skill metadata at startup rather than continuously.
Can Skills in Cursor run shell commands?
Where Cursor itself can, yes — which means action-taking Skills (deploys, migrations, paid API calls) carry the same risk here as anywhere. Check the capability level on the Skill page and keep credentials out of anything the agent can invoke unattended.
Other agents
Same question, different agent
- Best agent Skills for Claude Code →
The best verified Skills to install in Claude Code, in install order — frontend-design for UI, superpowers for multi-step work, simplify for review, skill-creator for writing your own, and the Anthropic document Skills for real files.
- Best agent Skills for Codex CLI →
The best SKILL.md Skills verified in OpenAI's Codex CLI — 51 of 191 catalogued Skills, led by frontend-design, superpowers, Trail of Bits security review, and the vendor Skills from Stripe, Paddle, Resend, and Supabase.
- Best agent Skills for Claude.ai →
The best Skills to upload to Claude.ai — the Anthropic document Skills (pdf, docx, xlsx, pptx) for real files, frontend-design and web-artifacts-builder for artifacts, and the marketing set for non-engineering teams.
- Best agent Skills for the Claude API →
The best Skills to use through the Claude API and Skills SDK — the document Skills for file generation in production, firecrawl and Valyu for retrieval, and Composio for tool routing.
- Best agent Skills for Gemini CLI →
The Skills verified working in Google's Gemini CLI — 11 of 191 catalogued, led by frontend-design, Firebase Hosting, the Stripe set, and Composio.
- Best agent Skills for Antigravity IDE →
The best Skills for Antigravity IDE, which ships a large pre-installed bundle — what is already there, what is worth adding on top, and the 36 Skills verified on this surface.
New to Skills? What is a Claude skill · How to install · How to audit a skill first