Claude Skills Examples: Real Skills and What They Do

The clearest way to understand Claude skills is to read a few. This guide walks through real examples in four groups: the smallest possible skill, skills that add knowledge, skills that run a procedure, and skills that ship code. Each example comes from Anthropic's documentation or from a published repository, and each shows a pattern you can reuse.

If the term is new to you, start with what are Claude skills.

The smallest useful skill

A skill does not need to be long. This is the complete frontmatter of Grill Me, from Matt Pocock's skills repository:

---
name: grill-me
description: A relentless interview to sharpen a plan or design.
disable-model-invocation: true
---

The body is a single line that hands off to a companion skill in the same repository, grilling, which holds the interview instructions. Two things are worth noticing.

First, the description is one sentence. It says what the skill does in words a person might use.

Second, disable-model-invocation: true is a Claude Code field that stops Claude from starting the skill by itself. You run it with /grill-me when you want it. That suits a skill that changes how the whole conversation goes.

See the Grill Me page for the listing.

Skills that add knowledge

Some skills hold reference material that Claude applies while doing other work. Anthropic's documentation gives this example:

---
name: api-conventions
description: API design patterns for this codebase
---

When writing API endpoints:
- Use RESTful naming conventions
- Return consistent error formats
- Include request validation

There are no steps here. The skill loads when Claude is writing an endpoint and shapes the result.

Published skills of this kind are usually about taste and standards:

• Frontend Design, from Anthropic, describes itself as guidance for distinctive, intentional visual design when building new UI or reshaping an existing one. See Frontend Design.
• Humanizer rewrites AI-sounding text so it reads like the writer, and is based on Wikipedia's "Signs of AI writing" page. See Humanizer.
• Caveman switches the agent to a compressed way of writing that cuts output tokens while keeping technical accuracy. See Caveman.

The pattern to copy: write down the rules you would give a capable colleague, and let the description say when they apply.

Skills that run a procedure

Other skills are step-by-step tasks. This deploy example is from the Claude Code documentation:

---
name: deploy
description: Deploy the application to production
disable-model-invocation: true
---

Deploy $ARGUMENTS to production:

1. Run the test suite
2. Build the application
3. Push to the deployment target
4. Verify the deployment succeeded

$ARGUMENTS is replaced by whatever you type after the command, so /deploy staging passes "staging" into the instructions. The skill is restricted to manual use because nobody wants an agent deciding to deploy.

A commit skill from the same documentation adds one more field:

---
name: commit
description: Stage and commit the current changes
disable-model-invocation: true
allowed-tools: Bash(git add *) Bash(git commit *) Bash(git status *)
---

allowed-tools lets Claude run the listed commands without asking for approval each time while the skill is in use. It does not block other tools. It only removes the prompt for these.

Claude Code ships procedures of its own as bundled skills, including /code-review, /debug and /verify. They are written as prompts, the same way yours would be.

Skills that bring code and data

A skill folder can contain more than Markdown. Scripts run without their source entering the conversation, and reference files load only when needed.

Anthropic's document skills are the best-known case. Four pre-built skills handle PowerPoint, Excel, Word and PDF files. Claude uses them on claude.ai when you ask it to create a document, and they are available through the API by ID: pptx, xlsx, docx and pdf.

Playwright Skill is a community example. Its frontmatter uses optional fields from the open standard:

---
name: playwright-skill
description: Complete browser automation with Playwright. Auto-detects dev servers, writes reusable test scripts, and supports screenshots, responsive checks, UX validation, login flows, link checks, and arbitrary browser automation. Use when the user wants to test a website, automate browser interactions, validate web functionality, or perform browser-based testing.
license: MIT
compatibility: Requires Node.js 20+, npm, and network access on first setup to install Playwright and Chromium.
metadata:
  author: lackeyjb
  version: "5.0.0"
allowed-tools: Bash(node:*) Bash(npm:*) Read Write
---

The description lists what the skill does and then, starting at "Use when", the situations that should trigger it. compatibility states what must be installed. See Playwright Skill.

UI UX Pro Max bundles data instead. Its description says the skill includes searchable local data such as styles, palettes, font pairings and UX guidelines, which the agent looks up while designing an interface. See UI UX Pro Max.

The pattern to copy: put anything deterministic in a script, put anything long in a reference file, and keep SKILL.md as the short guide to both.

Skills that work on other skills

Two examples exist to manage skills themselves.

• Skill Creator, from Anthropic, creates new skills, improves existing ones and measures their performance with test cases. See Skill Creator.
• Find Skills, from the Vercel Labs skills repository, helps you discover and install skills when you ask whether a skill exists for a task. See Find Skills.

Collections

Some repositories publish a set of skills meant to be used together. Superpowers is a well-known example. Its repository holds skills with names such as brainstorming, writing-plans, test-driven-development and systematic-debugging, and it installs in Claude Code as a plugin. See Superpowers.

A collection shows how skills combine: each one stays small and has a narrow description, and the agent loads the one that fits the step it is on.

What the good examples share

Reading across these, four habits stand out.

1. The description names the trigger. The strongest ones say what the skill does and when to use it, in the words a user would type.
2. Side effects are manual. Skills that deploy, commit or take over the conversation set disable-model-invocation: true.
3. The main file is short. Detail lives in scripts and reference files. The specification recommends keeping SKILL.md under 500 lines.
4. Requirements are stated. If a skill needs Node.js, network access or a particular tool, the frontmatter says so.

Try one

Pick an example close to your own work and install it. Most install with one command:

npx skills add owner/repo

Each page in the directory has the exact command. To write your own from one of these patterns, read how to create a Claude skill, and for a short list to start from, see the best Claude skills.