Home › Rules › Claude Code skills examples

Claude Code skills examples: 7 copy-paste SKILL.md files for beginners

Based on Claude Code's "Extend Claude with skills" docs, checked Oct 20, 2026. Sources below.

• Formats
• Claude Code
• Cursor
• Codex

Short answer: a Claude Code skill is a folder with a SKILL.md file in it. Save it at .claude/skills/<name>/SKILL.md in your project (or ~/.claude/skills/ for every project). The top of the file has a short description that tells Claude when to use it; the rest is the steps. Claude loads a skill when your request matches the description, or you run it yourself by typing /name. Below are 7 skills you can copy, each showing one feature.

Where skill files go

If a personal and a project skill have the same name, the personal one wins. Keep each SKILL.md under about 500 lines; put long reference material in extra files in the same folder and mention them from SKILL.md.

The parts of a SKILL.md

All fields are optional, but Claude Code's docs recommend always writing a description. The frontmatter must start on the very first line of the file, between two --- lines.

7 Claude Code skills examples

1. Explain an error in plain English (A skill Claude can pick up on its own)

---
name: explain-error
description: Explain an error message in plain English before fixing it. Use when I paste an error, a stack trace, or say something is broken or red.
---

When I share an error:

1. Say in one or two plain sentences what the error means. No jargon.
2. Point to the file and line most likely causing it, and say why.
3. Suggest the smallest fix. Don't change any code until I say go.
4. If you're not sure, say so and tell me what to check first.

No disable-model-invocation, so Claude can use it whenever an error shows up. The description says when to use it, in words you'd actually type.

2. Add one page, step by step (Passing what you type with $ARGUMENTS)

---
name: add-page
description: Add one new page to the app in small, testable steps.
disable-model-invocation: true
argument-hint: [page name and what it does]
---

Add this page: $ARGUMENTS

1. Restate the page in one sentence and list the files you'll create or change. Wait for my OK.
2. Build the simplest working version. Reuse existing components and styles.
3. Add it to the navigation.
4. Tell me exactly how to check it in the browser, including on a phone-sized screen.

Run it as /add-page pricing page with 3 plans. Everything after the name replaces $ARGUMENTS. disable-model-invocation: true means it only runs when you type it.

3. Commit my changes (Pre-approving tools with allowed-tools)

---
name: commit
description: Stage and commit the current changes with a clear message.
disable-model-invocation: true
allowed-tools: Bash(git status *) Bash(git diff *) Bash(git add *) Bash(git commit *)
---

1. Run git status and git diff to see what changed.
2. If you see secret keys, .env files or large build folders, stop and tell me. Don't commit them.
3. Write a one-line commit message in plain English that says what changed and why.
4. Stage the files and commit. Don't push.

allowed-tools lets Claude run those git commands without asking each time, but only during the turn you invoke the skill. It never pushes, because pushing isn't on the list and the steps say not to.

4. What did I change? (Pulling live data in with !`command`)

---
name: what-changed
description: Summarize my uncommitted changes and flag anything risky. Use when I ask what changed, want a commit message, or ask for a review.
---

## Current changes

!`git diff HEAD`

## Instructions

Summarize the changes above in 2 or 3 plain-English bullets.
Then list anything risky: deleted checks, hard-coded keys or URLs, missing error handling, or tests that need updating.
If there are no changes, say so.

The !`git diff HEAD` line runs before Claude reads the skill, so Claude sees your real changes instead of guessing. This is a Claude Code feature; other tools that read SKILL.md may not run it.

5. Our design rules (Reference knowledge that loads only for matching files)

---
name: ui-style
description: Design rules for this app's pages and components. Use when creating or editing UI.
paths: src/components/**, src/app/**
---

- Reuse the existing Button, Card and Input components before making new ones.
- Use the Tailwind classes already in the project. No inline styles.
- Mobile first: every layout must work at 375px wide.
- Buttons and links need a visible focus style.
- Keep the color palette to the brand colors in tailwind.config.

paths means Claude only loads it automatically when it's working on matching files, so it doesn't crowd every chat. Good for rules that grew too long for CLAUDE.md.

6. Ask before touching the database (A safety rule tied to risky files)

---
name: database-changes
description: Rules for any change to the database schema, migrations, row-level security or auth. Use before editing anything in supabase/ or prisma/.
paths: supabase/**, prisma/**
---

Before any database change:

1. Explain the change in plain English and what could break.
2. Show me the migration before running it. Wait for my OK.
3. Never turn off row-level security, and never drop a table or column without asking twice.
4. After the change, tell me how to check that existing data is still there.

Claude Code's docs say a rule that must hold every single time belongs in a hook, not a skill. For beginners this is still a good guard, but keep the most important line near the top.

7. Pre-publish SEO check (A manual checklist you run before launch)

---
name: seo-check
description: Check every public page's title, description and headings before publishing.
disable-model-invocation: true
---

For every public page:

1. List the title tag and its length. Flag any over 60 characters or duplicated.
2. List the meta description and its length. Flag any missing or over 155 characters.
3. Check there's exactly one h1 and it says what the page is about.
4. Check every image has alt text.
5. Show the results as a table, then the 3 most important fixes. Don't edit anything until I say so.

A skill like this costs almost nothing until you run it, because only its one-line description sits in Claude's context the rest of the time.

Who can run a skill

Skills vs CLAUDE.md

CLAUDE.md is always-on: facts about the project and rules Claude should keep in mind every time. A skill is on-demand. Claude Code's docs suggest making a skill when you keep pasting the same checklist or steps into chat, or when part of CLAUDE.md has turned into a procedure rather than a fact, because "a skill's body loads only when it's used". Start with a short CLAUDE.md (examples here) and add skills as you notice yourself repeating things.

Do these skills work in Cursor and Codex?

Mostly, yes. Claude Code skills follow the Agent Skills open standard, and Cursor also loads skills from .claude/skills/. The parts every tool shares are name, description and the steps. Extras like paths, argument-hint, $ARGUMENTS and the !`command` line are Claude Code features, so other tools may ignore them. More on this: Cursor rules vs skills.

Skill not working?

• Claude never uses it: add the words you'd actually say to the description, ask Claude "What skills are available?", or run it directly with /name.
• It runs when you don't want it: make the description more specific, or add disable-model-invocation: true.
• /name works but Claude never picks it: the YAML at the top probably has a typo, so Claude loads the steps but can't read the description. Start Claude Code with claude --debug to see the error.
• Claude follows it at first, then drifts: word steps as standing rules ("run the tests after every edit"), put the most important lines at the top, and run the skill again after a long chat gets summarized.

Related

• Cursor rules vs skills (rules, AGENTS.md and SKILL.md compared)
• CLAUDE.md examples and best practices
• AGENTS.md vs CLAUDE.md
• Every rules file format
• Vibe coding security checklist (turn it into a skill)

Sources

• Claude Code docs: Extend Claude with skills
• Claude Code docs: How Claude remembers your project (CLAUDE.md)
• Agent Skills open standard
• Cursor docs: Agent Skills

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