Claude's new "Skills" feature lets you build reusable AI instruction sets that work across multiple platforms
A solo tutorial by Nate B Jones demonstrating Claude's newly launched Skills feature and how to use it beyond Claude itself.
Summary
Nate B Jones covers Claude's newly launched Skills feature — which Anthropic calls "capabilities" — and argues it represents one of the most significant AI productivity developments of the year. The core idea is that users can build reusable, composable instruction sets (structured as markdown files in a zip folder) that Claude automatically invokes in relevant conversations, removing the need to re-enter lengthy context each time. Jones demonstrates how to build a skill for job search strategy, how to use a second LLM like ChatGPT to critique and improve that skill, and — crucially — how the exact same skill files can be uploaded and used in ChatGPT and Gemini, not just Claude. He argues this portability is an underreported aspect of the launch, and that the files effectively function as transferable onboarding packages for any AI system.
Key Takeaways
FULL TRANSCRIPT
Claude's Skills Launch and Why It Matters
Nate B Jones: I think this is one of the biggest pieces of news in the entire year. Claude launched a way for us to get past the tyranny of the prompt, and I worked out a way to use that same technique outside Claude — in ChatGPT, in Gemini, and other places. We're going to dive into the whole thing, but the first piece I want you to realize is that everything has been prompt-dependent, and that has made hard work really difficult.
Basically, if you want to do something complicated — like a PowerPoint, or an Excel file, or a full financial analysis — it's possible, but it's really been prompt-dependent. And anything we can do that makes it slightly easier, slightly less dependent on the prompt, is a really big deal, because it makes it so much easier for so many more people to do that hard work.
That is what we got today from Claude. It's called the Skills launch. The idea is that you have these composable Lego bricks — they're called capabilities in your settings section. I'm going to show you here in a second. All you have to do is enable capabilities that Claude can call in any conversation, in any combination.
I'll give you a specific example. One of the things that has been difficult for people to do with AI has been a complicated task like figuring out a job search strategy. Let's say you need to identify candidate job postings, compose a resume that fits those postings, make sure your search strategy actually works for those jobs, and that your targeting strategy works for outreach. This is complicated work. Normally it would take a lot of prompts — many prompts in a row. Some people sign up for tools and services for this.
What you can do with Claude is make skills that refer to this complicated work and give Claude special instructions to do it well. Your job search strategy skill could instruct Claude specifically what sites you prefer, what job family and roles you prefer, what levels you prefer, what compensation you want. All of that would be stored inside this Lego brick of context that Claude can go and call — not when you tag it, not when you have to remember it. You just say, "Hey, I'm trying to think about what job search strategy I should use for this week," and give it some context about your last few weeks and your learnings. Claude will look at that initial prompt, understand you're talking about job search, go dig up the skill, and use it on the fly. It's super intuitive.
The magical thing is that because of the way they've simplified the file structure — the Lego brick is just a markdown file — you can use this in other places. That's what I figured out. You can use it in ChatGPT, in Gemini. What this does is make the whole process of doing a difficult, complicated, multi-step piece of work so much easier.
This is not just a story about whether we can make PowerPoints easier, or Excel easier, or job search easier. It's a story about whether we can do hard work with much less effort. It's like Claude gave us a lever — a 10x lever on our prompting.
The Claude Capabilities Settings Page
Without further ado, I'm going to jump in. The first thing I'm going to show you is the settings page — how this actually looks when you sign up for it. But we won't spend much time there. We're going to spend most of our time diving into how you create these, because the magic of this is that you are not locked into whatever the Anthropic and Claude team give you. You can make these yourself. And I can make them for you — and I have. I'm making them for you.
It's so exciting because we can basically make these big packages that help us effectively do lots and lots of complex work. Another example: AI vendor risk estimation, AI vendor build-versus-buy solution estimates — super complicated work. You can have a skill for that now. You don't have to write a gigantic prompt in the same way you did. You can just say, "Hey, please assess this vendor," and it will call the skill. It will be so much more intuitive. I'm so excited for this.
Here we are looking at the skills that are in Claude. This is my Claude, and I want to show you that there are both standard skills and custom skills here. Standard skills — you're going to see brand guidelines, canvas design, MCP builder. These are all skills that Claude provides for us as a way of getting this whole ecosystem started.
I created one called Prompting Pattern Library, and it says "added by me, 59 minutes ago." It describes what it is — basically a file of prompting best practices so that if I talk about how I want to create a great prompt, it will just invoke the Prompting Pattern Library and make that so much easier. This is super easy to enable. All you have to do is enable the skill section at the top and then enable these toggles piece by piece, and you're done.
Building a Skill: Job Search Strategy Example
I want to show you an actual skill that I'm building. I said to Claude, "Can you help me build a skill for strategizing job search?" I want you to note that this is not the world's best prompt — and I did that on purpose. I wanted to call out that it's possible to build a good skill and the prompt doesn't have to be perfect. I just laid out a bunch of things that I wanted: guidance on how to interview the user, how to target the job, and so on.
Claude then starts to think about how to create skills. One of the things that Anthropic released — which I think is really smart — is documentation for Claude on creating skills. So Claude read the instruction manual for creating these skills and then came back and said, "This is really great. I have some questions — here are some questions about how I should analyze the company's news, strategies that are most important," and so on. I came back with a response, gave it some examples, had an LLM help me build some of this, and then said, "Great, excellent details, I'm going to start building."
It spent the last five or so minutes building this skill and came back with a job search strategist zip file — a complete skill, done. It's as easy as that. I can now download this zip file, re-upload it to Claude in the capabilities section, and I will have a usable job search strategist skill.
Now, if I want to make it better — which is always one of the things I focus on — I always think about how I can evaluate this skill through the lens of another LLM. I can take this skill over to ChatGPT and say, "Hey, can you crack open this zip file, look inside, and assess whether or not this is high quality and make specific recommendations for how to improve it?" Then I can pull that ChatGPT critique, bring it back into Claude, and have Claude further improve the skill. So I'm not just sitting here taking what Claude gives me. I'm actually using a multi-LLM strategy to make this skill as useful as I possibly can.
Using Claude Skills in ChatGPT and Gemini
Now let's jump over to ChatGPT and I'll show you how I can use a skill that I developed in Claude — for Claude — in ChatGPT.
Here we are in ChatGPT. You can see that I've uploaded a zip file. This is literally the exact same zip file that I would upload if I was in Claude. I am uploading it per chat because that's how ChatGPT works. All I say is, "Using this file, can you help me come up with a really strong prompt for instructing an LLM to build a PowerPoint? You'll need to crack open the file to do so." It opens the file and immediately comes up with a very strong PowerPoint prompt. I don't have to give it all of this context — it reads it in the file.
I love the Claude interface. It's very intuitive for it to call in the skills automatically — I think that's a lovely touch. But I can use the exact same files in a ChatGPT chat and get a great result. I can use them in a Gemini chat and get a great result. And nobody is talking about that. Nobody is saying that what has really been invented here is a way of working with AI that gives you composable Lego bricks that are portable. I think that's fantastic.
You can just pop this in, and all you have to do is build the ones you want. You literally just say what you want and work with Claude to build it — Claude has the skill creation process built in. You could do the same thing with ChatGPT, to be honest. You can say, "I want you to create a markdown file. The markdown file needs to be a complete, clear set of instructions to help an LLM accomplish a very complex task. In addition, please create these other file types that help the LLM do the task." If it's an analysis file, you may want to give the LLM some specific instructions about how to analyze, and so on. Then you put it all together into a zip file and you're done. You have your skill. You can upload that and say, "Hey, refer to this skill — this is how I like things done."
What Should Be a Skill?
You might be wondering now: what things should be skills? The answer is quite simple. If it is something that you would want to onboard someone with — let's say you have an employee and you want to onboard and train them — just give them a skill. That's what this is for. That's the level we're at.
If it's a one-off task, if you're not going to repeat it, if it's not very high value, don't do a skill. It's just too much trouble. But if it's something where you're going to be doing it repeatedly — if you had someone doing this for you, you would give them some training, you would give them some onboarding materials — that's when you need the skill.
The beauty of the whole ecosystem is that we can all just create the skills and trade them around, and they'll grow. That is part of what makes this so powerful — it's not locked into Anthropic. People will come up with ways to use these exact files in other interfaces. I guarantee that people will come up with ways to use these exact same Claude skills folders and files in a dozen other tools in the next month, because they're so powerful.
The Catch: You Still Need to Prompt Clearly
If you're asking me what the catch is, where you need to be careful, I would tell you this: we were using the skills before we knew they existed. When I did my video on Excel, when I did my video on PowerPoint, Claude was using those skills. And what I found in practice — weeks ago and since, as I've been working with those tools daily — is that even though the skills are like a massive 10x step up in helping you get stuff done, you still need to prompt well.
It does not get you away from prompting well when you do serious work. Prompting well is like giving this massive, powerful skill package clear direction. It doesn't have to be as long a prompt, but it has to be very clear and unambiguous about where you want to take all of this power, all of this context you've got. That's the catch. You still have to be clear in what you're asking for — you just don't have to be as lengthy as you used to have to be.
If I had to do a complicated piece of work like the job search thing, I would honestly not even do it in one chat. I would do it in multiple chats — help me with my resume, help me with my job search, and so on — and you'd have to reinvent the prompt and reinvent the context every time. The power of this is you can do that much more complicated work in a much less heavy-lift way.
That is why I call them super prompts. It's like we've taken this load that we've all had to carry for prompting for so long and lifted it. Yes, you still have to be clear. It's good to be clear about what you want. It's good to bring your specific context — bring that specific job description, bring the specific experience you have that nobody else has. But you don't have to carry as heavy a load on just explaining everything in exhaustive detail as you used to have to. And that is a huge breakthrough for all of us.
I'm putting together a whole set of Claude skills that you can use in ChatGPT and Gemini — and in Claude, of course. I can't wait to show them to you. We are just at the beginning of this, and I want to give you a great get-started pack so you're off to the races on using Claude skills. Have fun with your new super prompting powers.