How To Upgrade Claude AI with Custom Skills
Let's learn how to use Claude Code📅 2025-10-17
Transcript
[00:00] Claude just released a massive upgrade called Claude skills. In this video, we're not only going to learn how to enable them. I'm also going to show you how to build out your own custom skill. I'm going to also let you download one of the skills I personally created. And this custom skills use case was provide any CSV and depending on the industry, it'll provide you with the most optimal outputs. And finally, we're going to see how we can use skills in Enthropic API. If that all sounds good, go ahead and strap in, leave a like. Let's do it. Welcome back, y'all. In today's video, we're going to be checking out how to create, use, and leverage Cloud Skills. If you don't even know what a claude
[00:30] skill is, let me explain real quick. TLDDR, this gives the Claude AI model really good performance on specific task. It lets you really laser in and fine-tune for it. To even enable these, simply go to your profile, go to settings. Once you're in settings, come down here to capabilities. Go ahead and scroll down. And in order for these skills to work, you have to enable code execution and file creation. If you have analysis tool enabled, we can't do anything. So do this one. We're going to be good to Now, Anthropics team provided us with a bunch of starter skills such
[01:01] as algorithmic art, internal comms, canvas design. One that seemed kind of interesting was MCP Builder. But in this video, I want to show you actually how to create one of these because in reality to use one, it's really as simple as on or simply turn it on, three ellipses, try and chat, and then hit enter. Anyone can do this, but let me show you how to create one. So, in order to create our skill, we come up here to upload skill and we need a specific type of file here. So, first major resource I'm going to leave in the description down below is just the documentation around how to create a custom skill. But let's go ahead and do it together. I'm
[01:32] going to leave a custom skill that I created with Claude that allows us to put in a CSV file and then get data extraction right away. Essentially the use case here is that we've built out a skill for Claude where it's going to be able to one identify the industry eg if this is a financial report. It knows specifically what types of charts we want to see with that and what kind of numbers we even care about when a CSV data file comes in. This is in the description down below. If you just want a simple Corbin, I want this zip so I can just upload it. Simply come to that get link that I provide. Click this. Go
[02:02] up to these three ellipses. Hit download. This is going to give you a ready to go skill file where you simply just drag and drop. EG got the skill file here. Drag in. It's uploading. And you can see that we have it enabled here. Pretty cool. Also, make sure to give it a star and follow me on GitHub. Now, let's go ahead and dive over to the actual code for this custom skill I created and explain it. I want to be very clear here, though. I added a bunch of extra files here just to help y'all out. So, when you click that GitHub link, there's a nice little readme tutorial. I even give example data you could use as well, like a little CSV sheet of dummy data, which is nice. I
[02:34] like the dummy data. But all you really need to care about when you create your custom skill is one a very important file called skill MD. This is identified within the docs. And in addition to that, if we want to give the claude AI model ability to do very specific actions, this is where the code comes involved with the py python. Now I want you to think of the skill MD as essentially the brain in our cloud AI model brain. What specifically do we want to give as its really good skill it can handle? and for example CSV data summarizer that is a specific skill that
[03:04] is relevant to this skill. Two very important things for you to understand about the skill MD file so you don't get very very annoying errors. First thing this top part line 1 to 7 extremely fundamental extremely important you keep here. This is YAML YL. If you don't format it in this way when you try to upload it to claude here it's going to give you an error and it's going to be frustrating. So ensure that you format it this way. What is the name of your skill? Mine is going to be CSV data summarizer. Is yours? Where is the best coffee near me? Description. What the
[03:35] heck does your thing even do? What's your skill? Give a description. Metadata. What version is this? And the relevant dependencies needed to even run your code. Now, what does this mean? This means essentially in order to run this specific action, the clawed model needs Python 3.8 and all these other dependencies. What is a dependency? Corbin dependency is essentially giving your code different actions. So another way of saying this is that if we were building an application that needed the ability for monetization, eg give me a dollar. I love your dollar. Stripe would
[04:06] be a dependency that we would insult in order to handle this kind of workflow. Long story short, this non-negotiable. Keep it formatted like that. This open-ended. This essentially is going to be your massive, massive, massive prompt on the way you want it to handle the specific information it's receiving on that specific skill. Lucky for you, you can download my custom scale. Use this as a guideline. Use this as a framework. Notice this can be very, very, very in-depth. That's the point. Forbidden behaviors, asking the user what they want, listing options for the user to
[04:36] choose from. One thing I don't like when I want to leverage a skill within an AI model is please, if I provide the CSV and your specific skills to provide me with the best data visualization and data metrics, don't ask me what I want. Just know based off the industry. And I tailored this custom skill to do that. So that's nice. Skillmd the brain analyze.py Pi, the ability to actually move your arm, the action. This is going to be the relevant code file. Now, this is just straight up code, y'all. If you don't want to code it out raw, obviously use a chatbot as you see here. Chad, Claw, Gemini, whatever you want to do
[05:07] with the specific actions that you're looking for. My specific actions was the ability to internalize CSV data, understand the industry that's coming in, and give a relevant output. Analyze.py. Where is this reference in the scale MD? Scroll down files. Analyze.py. use it as your action. So that covers at a base level fundamentally how we create custom skills. I would encourage you to download my CSV data summarizer Claude skill here to give you the best template to start from as right now the template
[05:37] provided by Claude um this is the template. I don't know about you but that's not very helpful. So go ahead and just use mine and then just plug and play. The next topic of discussion here is using this within API. Right now, we've gotten up to the point where we're able to upload a skill and use it within Claude AI. Also, I want to point out if you follow me here on X, you can get early access to when I throw out free stuff like this. So, for example, this little CSV custom thing we created together, I put a little demo video. Click the link. Really cool
[06:08] stuff. So, follow me on X description down below. So, when using agent skills with API, same idea, but obviously we're communicating software to software. Essentially, we're not giving the user input. I don't have a nice little chatbot UI. rather I'm just sending data between different softwares. Eg the whole point of an API right here. So the idea is how do we use this and what's an easy way for me to test if this even works. One, I'll make sure I leave this docs in the description down below as well. Two, we're going to use a software called Postman here. What Postman allows us to do is send dummy calls to see if the data even works without us having to
[06:39] build out the infrastructure of an entire app to handle this kind of logic. I mean, you don't really need to build out the entire app infrastructure. In theory, you could just create a separate file in your repository, but for now, if you don't even know what Postman is, then merry Christmas or happy Hanukkah or whatever you celebrate. This is going to be an awesome tool you can use. Idea here is this though. So, essentially, in order to access a skill set, we need to want to give authorization, right? So, we're going to give our X- API- key. Provide your API key. You don't know where this is, go to cloud console, simply go to API keys, and create one.
[07:09] API key is going to be a long ass letter list, long ass string. Don't even worry about it. Bring it back over here. put it in the value the post where we're going to actually be sending this data to. Eg post means I'm giving anthropic some data to play around with. Where do you want me to send this data is going to be in the messages. In addition to this, we're going to do the anthropic version 2023601. You can increase that. Anthropic beta different things we're going to use in order for this to functionally work. Code execution skills. These are time dates. What time dates means is that this is when it was released to the
[07:39] public. For example, there may be a better version of anthropic version. Therefore, go to the more recent date. Content type application JSON, which is very, very familiar with if you ever done anything with API. All right, this is all cool, Corbin. But how do we actually leverage this new skill thing in the API? Well, first off, they have a new little tab here called skills. Right now, we got stuff out of the box. We got Excel sheets, PowerPoint, PDF, Word document. One that was most interesting to me was the PDF. So, I'm going to show that in today's video. In theory, you can add skill here and do the same process to add a specific custom skill,
[08:10] but within API. To actually call this, what we're going to do is go to the body here and we're going to make a payload. First off, we're go ahead and put in the model that you care about. What's very important here is in the documentation, this is how we identify the relevant skill. Eg anthropic skill ID, PDF. Are we not using PDF? Are we using PowerPoint? PPTX. Put that in there. Put it in. Version latest message. What do we want it to do? We know the output is going to be a PDF, but what do we want it to do? Make a PDF that says coffee is the best with a bullet list of three
[08:41] reasons why. And I don't know about you, but you already knew how to do the best three reasons why coffee is the best. Don't even don't even start in the comments like, "Oh, Corbin, I drink green tea. Green tea is amazing." Okay, coffee is better. With all that done, obviously, if you want a different functionality, we would change the message here. AI input. What we're going to do here is hit send and then I'm going to show you how to access this relevant file that was created with the skill set of PDF. This could be any skill set though. This could be my custom CSV generator. I mean this is why this is powerful in the context API because you can specialize endpoints within the software that you're
[09:11] creating. Now, obviously that's if you're creating it with anthropic API, but you get a status of 200. Okay, that means it went through console files. I did some testing, but you can see here, hit download, and here we go. This is why coffee is the best. It provides an energizing boost to start your day and keep you focused, right? The rich aroma and delicious taste makes every sip a delightful experience. Facts. It brings people together for meaningful conversations, connections. Beat that green tea. There we go. Now, let me just go over one thing that you may get that's very frustrating and annoying. If you're using Postman ever and you get an error
[09:42] of like, hey, it timed out. We can only do 30 second requests. That's built into the architecture here. In theory, you could do a way longer request when it comes to this. You just got to build it out, you know, locally on a code repo on your own. Here though, you're going to get a 30-second time limit. There's other ways within here that we can work around that 30-cond time limit, but that's not the point of this video. If you're running into errors, have questions, or just got cool things to show, check out the school community in the description down below. It's completely free to join. That's probably going to be your best way to troubleshoot different situations that occur. Long story short, think of this
[10:15] like Reddit but more specialized. So, that just about does Claude skills. Are you going to be using them or are you just like, I don't care about this, Corbin, but I want to watch just to see the use case. As you already know, just let me know in the comments down below. But, as we do with these style videos, I'll see you in the next. Why do I feel like in one or two weeks, OpenAI is going to release a similar type of feature and call it OpenAI talents?