← Back to Guides · CURSOR
Best Cursor Skills in 2026: What SKILL.md Is and 12 Worth Installing
Short answer: a Cursor skill is a folder with a SKILL.md file (usually .cursor/skills/<name>/SKILL.md) that the agent loads when your task matches its description, or that you invoke by typing /. Below: how skills differ from rules and commands, how to write your first one in two minutes, and 12 worth installing as a solo founder, each with a copy-paste snippet.
Skills follow the open agentskills.io format, so the SKILL.md idea is not locked to one editor. Cursor still labels the feature beta in its docs, so check the current docs if a detail below has moved.
Rules vs skills vs commands
Three folders, three jobs. Mixing them up is the most common reason an agent ignores your instructions.
| Rules | Skills | Commands | |
|---|---|---|---|
| Lives in | .cursor/rules/ | .cursor/skills/<name>/SKILL.md | .cursor/commands/*.md |
| Loads | Always, or when a glob matches | When the task matches the description, or when invoked | Only when you type it |
| Triggered by | Cursor | The agent (or you) | You |
| Best for | Short bans and conventions | Longer how-to playbooks | Repeatable multi-step jobs |
Rule of thumb: if it must hold on every turn, make it a rule. If it is a manual the agent only needs sometimes, make it a skill. If you want a button you press, make it a command. The full argument is in Cursor Rules vs Skills vs Commands.
Your first skill in two minutes
Create .cursor/skills/code-review/SKILL.md:
---
name: code-review
description: Review a diff for bugs, missing tests and security issues. Use when the user asks for a review, or before opening a PR.
---
# Code review
1. Read the diff, not the whole repo.
2. List correctness bugs first, then missing tests, then style.
3. For each finding give file, line, why it breaks, and a fix.
4. Say "no blocking issues" if there are none. Do not invent findings.
Two parts matter: the frontmatter description (it decides whether the skill ever loads) and a short numbered procedure. That is the whole format.
12 Cursor skills worth installing
Each snippet is a trimmed sketch, not the full AgenticKit version. Put each in its own SKILL.md, starting with frontmatter that has a name and a trigger-rich description.
1. Code review
Use it before every PR. See the example above, or the code-reviewer agent.
2. Test writer
description: Write tests for new or changed code. Use when asked for tests or after adding a feature.
---
1. Find the changed functions.
2. Test the behaviour, not the implementation.
3. Cover the happy path, one edge case, one failure.
4. Run the suite and fix what you broke.
Pairs with the testing-vitest skill.
3. Security scan
description: Scan for leaked secrets, missing auth checks and unsafe input. Use before deploy or when touching auth, payments or env files.
---
1. Grep the client bundle and repo for sk-, sk_live_, AKIA, ghp_.
2. Check every API route verifies the session and tenant.
3. Check user input is validated before it reaches the database.
4. Report findings by severity with a fix for each.
Background in the 60-second API key check and the security-basics skill.
4. Database migration
description: Plan and write a safe database migration. Use when changing schema.
---
1. Write the migration as additive first (add column, backfill, then drop).
2. Never rename and deploy in one step.
3. State how to roll back.
See postgres-drizzle.
5. Design polish
description: Tighten UI spacing, hierarchy, states and accessibility. Use when a screen works but looks rough.
---
1. Check empty, loading and error states exist.
2. Check heading hierarchy, contrast and focus rings.
3. Use design tokens, not one-off values.
See design-systems.
6. SEO audit
description: Audit titles, H1s, meta descriptions, schema and internal links. Use when asked about SEO or before publishing a page.
---
1. One H1 per page that matches search intent.
2. Keep the title and meta description short enough not to truncate.
3. Add schema that matches the content.
4. Link to related pages with descriptive anchors.
See on-page-seo and the /seo-audit command. For what this skill actually catches in Search Console data, see the 4 real page fixes from our own audit.
7. Launch checklist
description: Run a pre-launch checklist. Use when the user says launch, ship or go live.
---
1. Errors monitored? Backups on? Env vars set in production?
2. Pricing, checkout and emails tested end to end.
3. Landing page, OG image and sitemap in place.
4. Draft the announcement posts.
See /launch.
8. Stripe payments
Webhook signature checks, idempotent handlers and subscription states are common places for generated code to go wrong. See stripe-payments. For a narrower checklist covering Checkout, raw-body verification and subscription sync, see the Stripe billing skill, or pair it with glob-scoped Cursor rules for Stripe codebases to catch the same mistakes without loading a full skill every turn.
9. Auth patterns
Session checks, tenant isolation and OAuth callbacks. See auth-patterns.
10. Error handling
One error envelope, no swallowed exceptions. See error-handling.
11. Landing page CRO
Headline, proof, objection handling and one call to action. See landing-page-cro.
12. Project conventions
Your stack, folder layout and naming. Keep the always-on core as a rule and the long version here. See project-conventions.
For a full worked example, see the Cursor Playwright skill for sign-up, checkout and dashboard tests.
Where to find more, and how to vet a skill
Public repos and community lists carry many skills, but a skill is instructions your agent follows with your permissions. Before installing:
- Read the whole
SKILL.md. It is short. - Look for anything that runs shell commands, fetches URLs or touches env files.
- Check that the
descriptionsays when it should trigger, and that it does not try to override your rules. - Prefer small skills with one job.
You can generate scoped rules for your stack with the Cursor Rules Generator, grade the rest of your setup with the Rules File Grader and move files between editors with the Config Converter.
Common mistakes
- Vague descriptions. "Helps with code" never triggers. Say what it does and when to use it.
- Bloated skills. A 600-line skill burns context. Split by job; run it through the Skill Token-Cost Estimator to see where it's heaviest before you trim it.
- Always-on conventions in skills. If it must apply every turn, it belongs in a rule.
- Duplicating a rule and a skill. Pick one home.
FAQ
Are skills the same as rules? No. Rules apply always or by file glob. Skills load when relevant or when invoked.
Do skills work in Claude Code? The SKILL.md format follows the agentskills.io standard, and AgenticKit ships the same skills to .claude/skills/ for Claude Code. For the exact directory gotchas when moving a skill between the two tools, see Cursor Skills vs Claude Code Skills: What Actually Transfers Between Them.
Can the agent run commands on its own? Commands are manual: you type them. The agent can pick a skill by itself when its description matches.
Is there a Cursor skills marketplace? Skills are shared mostly through repos and community lists. Vet each one as described above.
Skills vs. MCP servers — are they the same thing? No. A skill is an instruction file the agent follows with tools it already has; an MCP server gives it new tools and live external access, like querying a real database or calling the Stripe API. See Best MCP Servers for Claude Code SaaS Founders in 2026 for which servers pair with the skills above.
Starting from rules instead? Best Cursor Rules for Next.js and SaaS in 2026 has ten .mdc files and says which should become skills.
Want a ranked shortlist instead of the full picture? 25 Best Cursor Skills for Solo SaaS Founders in 2026 (Ranked) ranks 25 real skills by setup time and blast radius if missing.
Install the whole set
AgenticKit installs 34 agents, 61 skills (including the ones above) and 47 commands into your repo with one command:
npx agentickit-cursor init --target cursor
Prefer manual prompts? See 12 copy-paste Cursor commands. Browse all Cursor skills, try the free tools, or see pricing.
Related Tools & Agents
Install the whole skill set in one command
34 agents, 61 skills and 47 commands land in .cursor/ and .claude/.
Get the kit →