What Are Claude Skills?

▶

The Problem Skills Are Designed to Solve

Every frequent Claude user has experienced it: you find a prompt that works well for a specific type of task, and you want Claude to behave that way every time you do that task. Maybe you want all code explanations to use simple analogies because you are teaching a beginner. Maybe you want Claude to always respond in a structured table format for comparison questions. Maybe you need Claude to know that whenever you say “client doc,” you mean a particular style of executive summary.

Without Skills, your options are limited. You can retype the instruction every time — tedious and easy to forget. You can paste it from a note — slightly less tedious but still manual. You can try to rely on Claude remembering your preferences from earlier in a conversation — but new conversations start fresh. None of these solutions scale if you have multiple recurring preferences across many types of tasks.

Skills solve this by allowing you to define a behavior once, give it a name, and have Claude apply it automatically whenever the context makes it relevant. You teach Claude something once; it knows it going forward.

What a Skill Actually Is

A Skill is a named, persistent instruction set that lives in your Claude.ai account and automatically activates when Claude determines it is relevant to your current conversation. It is not a saved conversation. It is not a prompt template you trigger manually. It is a behavioral definition — a description of how Claude should act in a particular context — that Claude monitors for relevance in the background.

The word “automatic” here is doing important work. With a regular one-time prompt, you tell Claude what to do right now. With a Skill, you tell Claude what to do whenever a certain kind of situation arises. Claude takes on the responsibility of recognizing when the Skill applies and incorporating it into its response, without you having to remember to invoke it.

A Skill definition typically includes: a name that helps Claude recognize the context; a description of when to apply it; and the actual instructions for how to behave. For example, a Skill named “Executive Summary Style” might say: “When I ask you to write a document for a client or executive audience, always structure the response with a three-sentence summary at the top, followed by bullet-pointed key points, followed by any detailed content. Never exceed 500 words unless I explicitly ask for more.”

Creating Your First Skill

Creating a Skill is a one-time definition process. In Claude.ai, you navigate to the Skills section of your profile and describe the behavior you want to define. You give it a name, write the instructions, and optionally specify trigger conditions — contexts in which Claude should apply it.

The best Skills come from noticing patterns in your own usage. Ask yourself: what instructions do I find myself giving Claude repeatedly? What preferences do I have that apply to a whole category of tasks rather than just one conversation? What domain-specific knowledge would be useful for Claude to remember — a company name, a style guide, a set of terminology?

Good Skill candidates include formatting preferences (“always use markdown headers for long responses”), domain context (“I work in pharmaceutical regulatory affairs, so assume familiarity with FDA submission requirements”), workflow steps (“when I ask you to review code, always check for security issues first before commenting on style”), and output templates (“when I ask for a weekly update, use this exact structure: [template]”).

Skills that are too narrow (“respond in a British accent when I use the word ‘cheerio’”) tend not to justify the overhead of creating them. Skills that are too broad (“always be helpful”) are already the default behavior and add nothing. The sweet spot is specific enough to be meaningful but general enough to apply across many conversations.

Examples of Useful Skills in Practice

Consider a few concrete examples of Skills that deliver recurring value:

A freelance writer might create a Skill called “Style Matching” that includes a few paragraphs of their own writing as examples, with the instruction: “When I ask you to draft content in my voice, match this style: short sentences, active verbs, minimal adjectives, conversational but precise.” Without this Skill, they would need to paste style examples into every session.

A software team lead might define a “PR Review Checklist” Skill: “When I ask you to review a pull request description, check it against these criteria: clear problem statement, explanation of the approach taken, testing evidence, link to related issues, no jargon that a junior developer would not understand.”

A customer support manager might create a “Response Policy” Skill with their company’s specific tone guidelines, escalation criteria, and prohibited phrases — so any time they draft support responses with Claude’s help, those guidelines are automatically applied.

Skills vs. CLAUDE.md: Knowing the Difference

If you use Claude Code — Anthropic’s command-line developer tool — you may already be familiar with CLAUDE.md, a file you place in your project repository to give Claude project-specific instructions. It is important to understand how this differs from Skills.

Skills live in your Claude.ai user profile and are applied in the Claude.ai web and mobile interface. They are personal to you and travel with your account. They are designed for the conversational, chat-based interaction model.

CLAUDE.md is a file in a code repository. It is read by Claude Code when it operates on that project. It is project-scoped (shared with anyone who checks out the repo) rather than user-scoped. It is specifically designed for the agentic development workflow — telling Claude about the project architecture, coding conventions, testing commands, and deployment processes.

The two systems are complementary, not competing. A developer might have Skills in Claude.ai for their general working preferences (code explanation style, response format), while also maintaining CLAUDE.md files in each project for project-specific technical context. When using Claude Code specifically, CLAUDE.md is where project-level customization belongs.

Browsing and Discovering Existing Skills

One of the underappreciated features of the Skills system is discoverability. Claude.ai includes a Skills directory where users can browse and install Skills created by other users and by Anthropic. This means you do not have to start from scratch when you want to adopt a new workflow — you can find a Skill someone else already built, install it, and optionally customize it for your specific needs.

This community library model means the Skills ecosystem compounds over time. As more users contribute their most useful behavioral definitions, the library becomes a resource for best practices across many domains — writing styles, research workflows, technical review processes, educational formats, and more.

When evaluating Skills from the directory, read the description carefully to understand what trigger context the Skill is designed for. Test it on a few representative conversations before relying on it heavily. And remember that you can always modify an installed Skill to better fit your specific context — treating it as a starting point rather than a fixed definition.

🧠 Check your understanding

4 questions · your answers are saved in this browser only

1. 1. What is the key difference between a Skill and a regular one-time prompt?

   Skills are longer and more detailed than regular prompts Skills are applied automatically whenever relevant; regular prompts require manual entry each time Skills only work in Claude Code, while prompts work in Claude.ai Skills can only define formatting preferences, not behavioral instructions

   The defining characteristic of a Skill is automation: you define the behavior once, and Claude monitors for relevant contexts and applies it without you having to remember to include the instruction. Regular prompts require you to type or paste them into each conversation.

2. 2. Where do Claude Skills live and apply?

   In a CLAUDE.md file in your code repository In your Claude.ai user profile, applied across conversational sessions In a special file on your computer that Claude Code reads In a shared team workspace that all employees access

   Skills are tied to your Claude.ai user account and apply across your conversations in the Claude.ai interface. They are user-scoped and personal, unlike CLAUDE.md which is project-scoped and shared with anyone working on the same repository.

3. 3. Which of these is the best candidate for a Skill?

   A one-time instruction to summarize a specific document A formatting preference you apply to every technical explanation you request An extremely narrow instruction that applies to one specific phrase A reminder to be helpful, which is already Claude's default

   The sweet spot for Skills is specific enough to be meaningful but general enough to apply across many conversations. A formatting preference for a recurring task type fits perfectly. One-time instructions, overly narrow triggers, and redundant defaults are poor Skill candidates.

4. 4. How does CLAUDE.md differ from Claude Skills?

   CLAUDE.md is for Claude 3; Skills are for Claude 4 CLAUDE.md is project-scoped and used in Claude Code; Skills are user-scoped and used in Claude.ai CLAUDE.md supports images; Skills only support text There is no difference — they are two names for the same feature

   CLAUDE.md is a file in a code repository, read by Claude Code to understand project-specific context. Skills live in your Claude.ai profile and apply to your conversational sessions. They serve complementary purposes for different contexts.

Reset quiz

🛠️ Build it yourself

Follow these exact steps to reproduce it yourself

Build It: Create Your First Three Skills

Step 1: Audit your recent Claude usage for patterns

Open your last 15-20 Claude conversations (or think back through them if you cannot access history). For each conversation, note any instructions you gave that were not specific to that conversation’s content — instructions about format, style, tone, level of detail, or domain context. List every repeated instruction you find.

Rank your list by frequency. The top three are your first Skill candidates.

Step 2: Write and publish your first Skill

Navigate to Skills in your Claude.ai profile settings. Create a new Skill for your highest-frequency instruction. Write the Skill definition using this structure:

• Name: Short, descriptive (e.g., “Concise Technical Explanations”)
• When to apply: Describe the trigger context specifically (e.g., “When I ask you to explain a technical concept or how something works”)
• Instructions: Write the behavioral instruction clearly (e.g., “Use a concrete analogy first, limit the explanation to three key points, and end with one practical implication”)

Save the Skill and test it immediately by starting a new conversation and asking a question that should trigger it. Verify the behavior matches your intent.

Step 3: Build a domain context Skill for your primary work area

Your second Skill should be a domain context Skill that gives Claude background knowledge relevant to your work. This type of Skill reduces the need to repeatedly explain your professional context. Include:

• Your role and industry
• Key terminology or jargon Claude should understand
• Any assumptions Claude should make about your audience or goals
• Any information that is confidential and should not be referenced in shared outputs

This Skill does not need a specific trigger condition — it is background context that applies broadly. Once you have created it, start a fresh conversation and ask a question about your work area. Notice how Claude’s baseline understanding improves without any additional context from you.