How to Use Claude Code Skills

A buddy called me last week. He'd seen a post on X about a set of AI marketing skills — CRO, copywriting, SEO, analytics. "Looked amazing," he said. "So I clicked the link. And it took me to GitHub. And I have no idea what to do with any of it."

I knew exactly which post he was talking about. Corey Haines had open-sourced 34 marketing skills as a GitHub repo. Thousands of likes. Over 17,000 stars. The kind of thing that makes you think "I need this."

But my buddy isn't a GitHub expert. He's a marketer. He uses Claude like a power user, but he's never cloned a repo in his life.

So I walked him through it. The whole process — from that GitHub link to a set of personalized skills he could actually use in Claude. Took about twenty minutes. And what came out the other side surprised both of us.

Here's what I showed him.

What Is a Claude Code Skill?

Before I walk you through how to use Claude Code skills, let me explain what you're actually looking at when you open one of these repos.

A Claude skill is a markdown file. That's it. Not code. Not a plugin. Not software you install. It's a text document — written in plain English with some structure — that tells Claude how to approach a specific kind of work. The skill format is just markdown: headings, bullet points, and plain language that any human can read and any AI can follow.

Think of it as encoded expertise. Someone who's really good at CRO sits down and writes out how they think about conversion optimization. The frameworks they use. The questions they ask. The mistakes they watch for. The sequence they follow. All of it, captured in a document that Claude can read and follow.

When Claude has access to that skill file, it stops being a general-purpose assistant and starts behaving like someone who actually knows the domain. Not because the model got smarter. Because someone gave it the criteria for what "good" looks like.

I've written before about how the real bottleneck in AI isn't the technology — it's the gap between what experts know and what they've written down. Skill files are one answer to that problem. They're externalized expertise in a format AI can act on.

Two Ways to Use Claude Code Skills

There are two ways to use someone else's Claude skills, and they depend on which Claude you're working with.

Path 1 is for people using Claude.ai or the Claude desktop app. You don't write code. You don't use the terminal. You have conversations.

Path 2 is for people using Claude Code — Anthropic's command-line tool for developers. You're already in the terminal. You're comfortable with git.

Both paths end up in the same place. The only difference is how you get the skill files in front of Claude.

Path 1: Using Claude Skills in Claude.ai and Claude Desktop

This is the path most people need, and it's simpler than you'd think.

Step 1: Give Claude the repo link

Open Claude.ai or the Claude desktop app. Paste the GitHub URL into the conversation:

"Here's a GitHub repo full of marketing skills: https://github.com/coreyhaines31/marketingskills — can you find and read the skills in there?"

Claude will fetch the repo, find the skills directory, and read through the files. You don't need to download anything. You don't need to understand the repo structure. You just give it the link and tell it what you're looking for.

Step 2: Have Claude explain what it found

This is where it gets interesting. Don't just start using the skills blindly. Ask Claude to walk you through them.

"What skills did you find? Give me a quick summary of each one — what it does, what it assumes, and what strategy it's built around."

This step matters more than people think. A skill file isn't just instructions. It's someone's mental model. Corey's copywriting skill reflects how Corey thinks about copywriting. His CRO skill reflects his conversion optimization philosophy. Those assumptions are baked into every line.

You need to understand those assumptions before you can decide whether they match yours.

Step 3: Let Claude interview you

Here's where this stops being "using someone else's tool" and starts becoming yours.

Pick a skill that's relevant to your work. Then say:

"I want to use the page-cro skill, but I want to adapt it for how I work. Interview me one question at a time — ask me where my approach might differ from what's in this skill."

Claude will read through the skill's assumptions, frameworks, and decision points, and then ask you targeted questions. Not open-ended "tell me about your approach to CRO." Specific things like: "This skill prioritizes above-the-fold clarity over social proof. Do you agree, or do you weight those differently?"

This is the part that connects to what I call the extraction pattern. Most people struggle to articulate their expertise from scratch. It's brutal to stare at a blank page and write down everything you know about your domain. But reacting to someone else's thinking? That's easy. We do that all day. You look at Corey's approach and immediately know where you agree, where you disagree, and where your situation is different.

The skill file becomes the interview guide. Someone else already did the hard work of structuring the domain. You just have to say where you diverge.

I wrote about this same dynamic in the context of building before writing requirements — you learn what you want by seeing what you don't want. Starting from someone else's skill and reacting to it is the same principle. You're not starting from zero. You're starting from someone else's version of good, and chiseling it into your version.

Step 4: Have Claude create your personalized version

After the interview, tell Claude:

"Based on my answers, create a revised version of this skill that reflects my approach. Save it as a markdown file."

Now you've got a skill that started as Corey's expertise but became yours. Same structure, different judgment. Same framework, different priorities. And it's saved as a file you can use again — drop it into any future Claude conversation, attach it to a project, or just paste it in when you need it.

Step 5 (optional): Write your own Claude skill from scratch

Once you've been through this process with one skill, you'll notice something. The structure of a good Claude skill isn't domain-specific. It's a pattern: context, triggers, frameworks, decision points, anti-patterns, output expectations. That's the skill format — and once you see it, you can use it to write a skill for any domain you know well.

So you can say:

"I want to create a brand new skill for [your domain]. Use the same structure as Corey's marketing skills, but interview me to build it from scratch."

This is how one person's open-sourced expertise becomes a template for yours. You don't need to know how to write a Claude skill from a blank page. You just need one good example and a conversation. The value isn't just the 34 marketing skills. It's the skill structure those files teach you about how to encode expertise into something AI can use.

That's what I was getting at in AI Doesn't Replace Expertise — It Makes Expertise Portable. I'm not the world's best video editor. But I knew enough to encode what matters into a set of instructions. The skill file is the portable artifact. And once you see how one is built, you can build your own for anything you know well.

Path 2: Using Claude Code Skills in the Terminal

If you're using Claude Code, the workflow is almost identical. The only difference is how you get the skill files onto your machine. Claude Code reads skills automatically from your project directory — so once the files are there, it knows how to use them.

Step 1: Clone the repo

git clone https://github.com/coreyhaines31/marketingskills.git

Or, if you want to install specific skills directly:

npx skills add coreyhaines31/marketingskills --skill page-cro copywriting

Now the skill files are local. Claude Code will discover them automatically when they're in your project's skills directory.

Step 2: Have Claude Code explain the skills

Same as the Claude.ai path. Inside your Claude Code session:

"Read through the marketing skills I just installed. Walk me through what each one does, what assumptions it makes, and how they relate to each other."

Claude Code reads the markdown files, understands the dependency structure (Corey's skills reference each other — the copywriting skill connects to the CRO skill, which connects to the A/B testing skill), and gives you the full picture.

Step 3: The interview

Same process. Pick a skill. Ask Claude Code to interview you one question at a time about where your approach differs.

The only nuance with Claude Code is that you can be more technical in your divergence. Where a Claude.ai user might say "I prioritize social proof differently," a Claude Code user might say "I want the CRO skill to generate specific test hypotheses as code comments when it writes landing page components."

Step 4: Save the personalized version

"Save the revised skill to my project's skills directory."

Claude Code writes the file directly. It's now part of your working environment. Every future session in that project will read and apply your customized skill automatically.

Steps 5-6: Write and build your own Claude Code skills

Same as above. Use the interview-based approach to create new skills for your domain. Claude Code writes the skill file directly to your project — no copy-pasting, no file management. You describe your expertise through conversation, and Claude Code turns it into a skill that's ready to use in every future session.

The Skill Isn't the Point. The Structure Is.

I want to be direct about something. The 34 marketing skills in Corey's repo are good. Really good. But the most valuable thing in that repo isn't any individual skill. It's the structure.

Because the structure teaches you something: expertise can be encoded. Judgment can be written down. The things you know — the frameworks you apply, the mistakes you watch for, the questions you ask before making a decision — all of that can become a document that makes AI behave like someone who knows your domain.

Most people are still using AI with generic prompts. They type "write me landing page copy" and get generic landing page copy. Then they conclude that AI isn't that useful.

But the person who loads a CRO skill, a copywriting skill, and a product marketing context file before asking for landing page copy? They get something different. Not because the model is different. Because the criteria are different.

I wrote about this in why I never let AI grade its own work. The quality of AI output is only as good as the criteria you give it. A skill file is a set of criteria. And the conversation you have while personalizing that skill — the back and forth where you articulate what you care about and how you work — that's the real prompt engineering. Not a single-shot instruction. A dialogue that extracts your standards.

Start Here

Find a Claude skill repo on GitHub that's close to your domain. It doesn't have to be perfect. It doesn't have to match how you work. It just has to be structured well enough to give Claude something to interview you about.

Paste the link. Ask Claude to read it. Then let it ask you questions.

By the end of that conversation, you'll have something more valuable than someone else's skill file. You'll have your own expertise, externalized, in a format any AI can use. And if you want to go further, you'll know how to write a Claude skill from scratch — because you'll have already done the hard part.

The skills are the starting point. The interview is the work. And what comes out the other side is yours.