Curious Companion: WTF are Claude Skills?

LISTEN TO THE EPISODE

In this episode I break down the best way to teach Claude how to complete specific tasks in a repeatable way: Claude Skills. I explain how Skills work, walk through how to build your own, and cover the real differences between Skills and Projects. As per always, the best way to understand this stuff is to try it yourself first hand, but hopefully this episode will give you some ideas and a nice jumpstart.

The Best Way to Learn

Today we’re talking about Claude Skills, and while they’re absolutely not new (they’ve been around since October of 2025), I recently went on a walk with my guy Khe, and it inspired me to finally give them an official try.

As per always, when it comes to AI (and honestly, everything) the best way to truly learn and fully understand this stuff is to try it yourself. I actually watched quite a few long-form YouTube videos about Skills when they first launched last October and I kinda got the concept, but I really couldn’t fully wrap my head around it. Fast forward to last week, and I played with creating a custom Skill for about 20 minutes before making this episode…and just like that it was all clear as day.

What Exactly Is a Skill?

Let’s start with a definition, right off the bat, straight from Anthropic’s support page:

Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. Skills teach Claude how to complete specific tasks in a repeatable way.

The best skills “solve a specific, repeatable task” and are “focused on one workflow rather than trying to do everything.” The core file in a skill is a markdown file, and remember, a markdown file is simply a plain text file that uses symbols for formatting, it’s the preferred file format for LLMs.

Translation: A skill is simply instructions for a set way of doing something.

The cool part here is that once a skill has been created, Claude will load and use it automatically when it’s relevant for the task you’re performing. Worth noting, if you’re performing a task that Claude should be calling a particular skill for and it’s not, you can also just tell Claude to use it, something like, “Use my brand guidelines skill for this.”

Now let’s get techy for a secy, because IMO one of the coolest parts about Skills is how they actually work, which is called progressive disclosure.

It would be hugely inefficient for Claude to read the entire Skill folder and everything in it any time it thought that Skill might be helpful for a particular task. Instead, Claude uses progressive disclosure, which means it scans the skill description first (which is limited to 200 characters) and only loads the full instructions for a skill once it decides that skill is relevant to the task. Save them tokens. Efficiency for the win.

Finding Skills

Claude has a directory of skills that you can access directly inside of Settings, but there aren’t that many in there (at least not that I could find). To me, this wasn’t a huge deal because creating your own custom Skills is definitely the way to go.

Also, worth noting for my techies in the crowd, you can go on GitHub and search through a ton of skills there. Anthropic open sourced a repo, anthropics/skills, but as always, be careful about what you download to your computer.

How Do You Make a Skill?

As per always, just ask the robot.

Claude actually has a skill creator skill, how meta, but honestly all you need to do is tell Claude that you want it to help you create a Skill and you’ll be off to the races.

A few things to note:

The better the input, the better the output. Go back and forth with Claude about what you want this thing to do, and then test it to see if it works the way you want.
Claude will actually write and format the skill for you.
It’s supposed to be able to save it to your skills library on its own, but I had to actually hit the save button on the file it created. No big deal, but worth noting.
The name of the skill and the description are super important. Remember, that’s the only part Claude reads before deciding whether to load the rest of the skill, so a vague description means a skill that technically exists but never fires.

All of this can be done directly inside of a chat with Claude, but there’s another, more formal way to create a skill.

Your Skills live in Settings → Customize section of Claude. At the top of the Skills screen you’ll see a button that says “Add,” and that opens a dropdown with four options:

Create with Claude
Write Skill instructions
Upload a Skill
Record your screen – That one is very cool, and while I haven’t tried it yet, it’s definitely worth noting that it exists.

This Skills page is also where you can access your Skills at any time if you want to edit them. Or you can take the easy road and just have Claude edit them via chat.

My Custom Skill: The Client-Debrief Skill

So again, your best bet for truly understanding what Skills are, and for getting ideas of how they could be helpful for you, is to go through the process of building one yourself. But to give you an example, I’m gonna share the one I built as I was doing “research” for this episode.

The skill I built is called client-debrief. Here’s the description, which Claude wrote:

Helps process a client call after it’s over, when the user is still stuck on something or didn’t fully land where they wanted to during the call itself. Pulls the client’s transcript, surfaces every distinct thread in what’s going on for them, and offers a synthesis read of the pattern underneath, then stays in the conversation as a thinking partner. Use whenever the user names a specific client and says they’re still thinking about the call, didn’t wrap it up the way they wanted, or wants to work through what to do next with someone, e.g. “I’m stuck on what to do with [client]” or “that call with [client] didn’t really land, help me think it through.”

I also made a skill that pulls exact verbiage from the transcripts of these coaching calls that can be used for marketing materials.

Realistically the sky is the limit when it comes to creating a Skill, which I realize probably makes it harder to come up with one that would be helpful for you. AI continues to be “here’s a solution, go find a problem.”, but if you have a task that you want always done in the same way, that’s the foundation for a good use case for a Skill.

Skills vs. Projects

Hopefully you’re sitting there thinking, wait, this sounds like how you’ve taught us to use a Project, and you’re correct. Honestly, to me there is a lot of overlap, and creating my first Skill is what helped me understand the differences.

Skills can be used anywhere, in any conversation, in Claude Code, and in Cowork, whereas the instructions for a Project only apply to conversations in that Project.
Skills are simply specialized procedures, while Projects allow for, and are designed for, providing background knowledge that’s always loaded with every chat.
You can create custom prompts inside of a Project, but the main utility of having the Project is that you’ve created a workspace that always has context about the topic. For example, I have a Project that’s for Prompting Curiosity, this way I don’t have to explain the podcast and the audience every time.

Again, for the eleventy billionth time, it’ll make way more sense when you actually try it out for yourself.

How I Used AI This Week

Each episode I share a quick example of how I used AI that week.

This week I used AI to transcribe my coaching calls, via a tool called Granola. I absolutely love it. Maybe you’ve seen ads for it, if not, you’ll probably see one after reading this episode.

I used Zoom AI for a while, but what Granola produces is definitely better, particularly in the AI-enhanced post-call notes that it generates.

You do have to connect Granola to your calendar, which I didn’t want to do at first, but eventually I sucked it up and did it because I really wanted a solution for the post-call notes pain, and everything has been 100% fine.

Again, the notes that Granola generates literally immediately after the call (no wait time) are so good, and because it’s also transcribing in real time you can scroll back in the transcript if you missed something or if you want to reference something verbatim. There’s no annoying notetaker to let into Zoom, you don’t have to wait for Zoom to do all of its processing and then sign into Zoom to get the transcript or the notes. Granola is convenient, it’s sleek, and most importantly, the output is really good.

I’ve got no affiliate link or anything like that. I’m currently on the free tier and we’ll see if I bump up. Either way, I figured I’d share this tool with you because I’ve been loving it and perhaps it’ll be helpful for you.

Da Wrap-up

Claude Skills are just folders of instructions that let Claude handle a specific, repeatable task the exact same way every time. If you’re like me, it might be a little bit difficult to conceptualize what they are, but as per always, your best bet is simply to dive in and learn as you go.

As always, endlessly grateful for you and your curiosity.

Catch you next Thursday.

Maestro out.