Home › Rules › Cursor rules vs skills

Cursor rules vs skills: which one to use (plus AGENTS.md and SKILL.md)

Based on Cursor's rules and Agent Skills docs and Claude Code's skills docs, checked Oct 20, 2026. Sources below.

• Formats
• Cursor
• Claude Code
• Codex

Short answer: a rule is a short instruction the AI should keep in mind ("use TypeScript", "ask before changing the database"). A skill is a step-by-step procedure it runs when needed ("deploy to staging: test, build, deploy, verify"). Rules load as context in every chat or every matching chat. A skill's full steps load only when the agent decides it's relevant or you type /skill-name. Beginners should start with one AGENTS.md file and add a skill only when they keep pasting the same multi-step instructions.

Rules vs AGENTS.md vs skills at a glance

The first and last rows of the rules and skills columns come straight from Cursor's own "When should I use skills instead of rules?" table.

How Cursor rules load (the 4 types)

Each rule is an .mdc file in .cursor/rules/. A plain .md file there is ignored, because it has no frontmatter. The type you pick sets three frontmatter fields:

---
description: How we write React components in this project
globs: src/components/**/*.tsx
alwaysApply: false
---
- One component per file, named the same as the file.
- Use the existing Button and Card components before making new ones.
- Keep components under 150 lines; split anything bigger.
- No inline styles. Use the Tailwind classes already in the project.

More copy-paste rules: 10 Cursor rules examples.

How skills work

A skill is a folder with a SKILL.md file inside. The frontmatter needs a name (lowercase letters, numbers and hyphens, matching the folder name) and a description that says what it does and when to use it. The agent sees the list of skills and their descriptions, and pulls in the full steps only when one fits. You can also run one yourself by typing / in Agent chat.

Cursor looks for skills in .agents/skills/ and .cursor/skills/ in your project, and in ~/.agents/skills/ and ~/.cursor/skills/ for skills you want in every project. For compatibility it also loads .claude/skills/ and .codex/skills/, so one skill folder can serve Cursor, Claude Code and Codex.

---
name: pre-launch-check
description: Run the pre-launch checklist before publishing or deploying. Use when I say launch, publish, deploy or "is it ready".
---

# Pre-launch check

Do these in order. Stop and report after each step; don't fix anything until I say so.

1. Run the tests and the build. Report anything that fails.
2. Search the code and env files for secret keys in front-end code (NEXT_PUBLIC_, VITE_, EXPO_PUBLIC_ prefixes). List each one.
3. List every database table and whether row-level security is on.
4. Check every page at phone width and list anything that overflows or can't be tapped.
5. Give every public page a title under 60 characters and a meta description under 155. Show me the list.
6. Finish with a short summary: Ready / Not ready, and the 3 most important fixes.

Want a skill that only runs when you ask for it, like an old-style slash command? Add disable-model-invocation: true. Then the agent won't use it on its own, and it runs only when you type /add-feature.

---
name: add-feature
description: Add one new feature safely in small steps. Use when I ask for a new feature or page.
disable-model-invocation: true
---

# Add a feature

1. Restate the feature in one sentence and list the files you expect to touch. Wait for my OK.
2. Build the smallest working version only. No extra options, no refactors.
3. Tell me exactly how to test it by hand.
4. After I confirm it works, update AGENTS.md if anything about the project changed.

Skills can also carry scripts/, references/ and assets/ folders. The agent reads those only when it needs them, which keeps your chats lighter.

Should I turn my rules into skills?

• Keep as rules: anything with alwaysApply: true or globs. These have clear triggers. Cursor's built-in /migrate-to-skills leaves them alone too.
• Good skill candidates: "Apply Intelligently" rules (a description, no globs) that are really procedures, and old slash commands. /migrate-to-skills (Cursor 2.4+) converts these for you; commands become skills with disable-model-invocation: true.
• Rule of thumb from Cursor's docs: use a rule when a short instruction is enough, and a skill when the agent needs a detailed, repeatable process.

To make one from scratch, type /create-skill in Cursor's chat and describe what you want. For rules, /create-rule does the same.

AGENTS.md vs skills (and CLAUDE.md vs skills)

AGENTS.md and CLAUDE.md are always-on project notes: what the app is, how it's built, and the rules to follow. Skills are on-demand procedures. Claude Code's docs put it this way: make a skill when you keep pasting the same checklist or multi-step procedure, or when part of CLAUDE.md has grown into a procedure rather than a fact, because "a skill's body loads only when it's used". Claude Code has also merged custom commands into skills: .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both create /deploy.

# AGENTS.md

## What this is
A habit tracker web app for beginners. Next.js + Supabase.

## Rules
- Work in small steps. Tell me how to test each one.
- Ask before changing the database, auth or payments.
- Never put secret keys in front-end code.
- Explain errors in plain English before fixing them.

More on these files: AGENTS.md vs CLAUDE.md, Cursor rules vs AGENTS.md, CLAUDE.md examples, every rules file format.

A simple setup for beginners

1. Start with one AGENTS.md in the project root. Grab a free one for your project type from the rules library or the generator.
2. Add a Cursor rule only for file-specific guidance, like React components or database migrations.
3. Add your first skill the second time you paste the same multi-step instructions. The pre-launch check above is a good first one.

Related

• Cursor rules examples (10 copy-paste .mdc rules)
• Cursor rules vs AGENTS.md
• AGENTS.md vs CLAUDE.md
• CLAUDE.md examples and best practices
• Vibe coding security checklist (turn it into a skill)

Sources

• Cursor docs: Agent Skills
• Cursor help: Skills (When should I use skills instead of rules?)
• Cursor docs: Rules
• Claude Code docs: Extend Claude with skills
• Agent Skills open standard

Checked Oct 20, 2026. These tools change often; if something here no longer matches, tell us at [email protected].