Hello. And today I'm going to show you how to create your very first AI agent skill directly in VS Code. And then we'll see how and why it's the most efficient way to extend your agent capabilities today.
So what are agent skills? Agent skills are an open standard that Anthropic introduced just a few months ago. The idea behind them is very simple.
You can give your agent focused, reusable capabilities like interpreting PDF documents, managing Excel files, and doing image conversions. Now, sure, you can pull some of this off through system instructions, or using MCP servers, but agent skills are really about keeping things clean, simple, and efficient inside your agents. Instead of the agent having to chew through long system prompts or analyze a bunch of tools in MCP servers, this is really about efficiency.
The agent will only read the skill name and description to determine how to use that skill. That's less context to process. That will result in faster processing and less tokens, hopefully resulting in less overall costs.
skills use what's called progressive disclosure, which basically means the agent only sees what it needs when it needs it. It will start with the name and description and go further if it needs to call that skill. And the part I like best about agent skills is that they're crossplatform.
You build an agent once in VS Code and you can use the exact same code in Claude, Cursor, and even anti-gravity. In this video, I'll walk you through building your first AI agent skill step by step in VS Code and then I'll show you how to run it using your agents. Let's dive in and get started.
Okay, I've now jumped into VS Code and let's get started creating our first agent skill. Now, the very first thing we need to do is go in the bottom left to go to the gear icon and go to settings. And we need to make sure that agent skills are enabled.
So, do a search for skills. And what you'll see here, give you a little more room, is the agent skills checkbox. Now, by default, this is not enabled.
You go ahead and check that checkbox. Now, depending on your version of VS Code, this might look a little bit different than this. Here's what it looks like in the latest experimental release of VS Code.
It's very easy to see what folders are being leveraged. You can turn them on and off. You can add custom ones here as well.
And you have the ability to see both workspace level folders and then the overall profile level folders. And we'll take a look at the profile level folders uh in just a moment. So for now, we're back in normal VS Code.
Make sure that checkbox is checked. And we'll go ahead and get started. I'm now over here in a workspace.
And [snorts] the first thing I want to do in my workspace is create a new folder. And as we saw right here, it needs to be called GitHub. Then inside there, I'm going to create another new folder called skills.
And then I'm going to go ahead and create a folder that's going to house our agent skills components. Now, this is going to be called game of chess. And then inside here, I create a folder simply called skill.
md. Now, what I've noticed is the casing is very particular and a lot of things about the file layout is very particular. So if you're creating a skill and you don't see it picked up, go ahead and check your spelling of the folder name has to match the name of the skill which we'll create in just a moment.
And all the casing has to be right and so it's very particular, very finicky. So keep that in mind. Now in addition to the skill.
mmd file, there's other types of files and folders that can be created under here. Underneath the skill folder, you can have a scripts folder that contains scripts that can be referenced. These would have Python, Bash, or Node scripts that you could have in there that could do additional work.
You could have another folder for references. So, if you had things like templates that you wanted it to use, they could be under there, as well as assets. Assets could be under there.
So, you could have logos and images as well. So, for now, we're just creating a basic skill here. So, it's not going to have anything under there.
So, now I'm going to go ahead and create this file. First thing I need to do is have a name. As I mentioned, the name has to match exactly what the folder is.
There's also other rules, limits, 64 characters and no double dashes and stuff like that. Then you need a description. The description is what the agent is going to use to determine whether or not it's going to look at this skill further at the time of execution.
So in this case, we just want this to be a skill to return a game choice based on the user wanting to play a game. Okay? And you're going to want to make this as descriptive as possible.
Basically, this is your selection criteria on when you want this skill to be executed. Now underneath here is where we'd have the logic in markdown of what we want the skill to do. So again, if you are referencing scripts, you would have the relative path to those scripts to have them executed.
In this case, we're going to use a skill to respond, would you prefer a good game of chess for the user asking to play any type of game with the system. Now, this is just a trivial skill here just to kind of get used to the skills and how you would create them. Obviously, this isn't real business value that you would see here.
So, with that, I'm going to go ahead and save this file. And that should be all we need to do in order for this skill to be picked up and leveraged. Now, real quick, let me show you the two skills that we have at my profile level.
So, I'm going to jump over to the explorer window. And under my user profile here, I have a copilot folder that I created. Has a skills folder in there as well.
And in here, you see I have two additional skills that I've created. A dad's joke and a positive affirmations. Just simply will return a joke.
And then if you're feeling down or getting frustrated with your code outcome, you can ask for a positive affirmation to kind of get a pickme up. So both of these are here. Thought these would be useful in any project so that they are profilewide.
So let's jump back to VS Code. Let's come over here to the agent and let's ask the agent what skills are available. So now let's ask this agent all available skills that it has access to.
I'm going to say list all available skills. Better be specific. agent skills you have access to.
And now this will give it just a moment. It's going to interrogate those folders and determine what skills it has access to. And as we expected, we get all three of the skills back.
The game of chess that we just created here, the dad joke and the positive affirmations. Now, we could run these at any time. I said, let's play a game of poker.
And we'll go ahead and see what this does. You can see here it's interpreting what it needs to do. It's reading that skills MD file.
And now it did echo back. Wouldn't you prefer a good game of chess? Which is exactly what we wanted.
And now let's try this with telling me a joke. Tell me a dad joke. And let's go ahead and see if that can find that as well.
And again, it's reading our skills. mmd file. And in this case, it does uh call out to an outside API.
It is asking me for permission to do that. and it's going to go ahead and call that out. And here's a dad's joke.
Um, so what this was intended to do was just kind of show you high level how to create your first agent skill. Um, some of the places I've used this for is validation of like workflow files, things that are in XML or JSON markup. Wanted to apply uh consistent validation of files that maybe other developers are working on.
Things that you wouldn't really want to put in system instructions because not everything we work on is going to be a workflow file. So I could put these in an agent skill and ask the agent skill to then apply our corporate best practices to these files that are created and it can go out and know to do that at just a skills skill by skill basis actually. And if you want more information about this, the full spec for the agent skills is defined here agentskills.
io. You can go to specification and it does outline everything that you can use and build inside these skills. Anthropic has some predefined skills for you as well.
As I mentioned, you create skills here in VS Code and use those same skills in other IDEs and agents as well. And here's a whole bunch of skills repository that you can get access to as well as the GitHub awesome-copilot account here has a tremendous amount of agent skills as well where you can go grab these and do things like uh work with PDF files and um all kinds of good um things that you'd want to leverage. everything's already built for you.
This site uh also has a lot of good uh system instructions if you're looking for system instructions and uh stuff like that. So, I'll have links to both of these, all three of these actually, in the comments below, as well as the simple markdown file that we created to create our game of chess. So, with that, hopefully you can jump out and make your own agent skills and save speed and token count and hopefully increase your productivity.
And with that, thank you for watching this video. Don't forget to like and subscribe. Thank you.
Hello. And today I'm going to show you how to create your very first AI agent skill directly in VS Code. And then we'll see how and why it's the most efficient way to extend your agent capabilities today.
So what are agent skills? Agent skills are an open standard that Anthropic introduced just a few months ago. The idea behind them is very simple.
You can give your agent focused, reusable capabilities like interpreting PDF documents, managing Excel files, and doing image conversions. Now, sure, you can pull some of this off through system instructions, or using MCP servers, but agent skills are really about keeping things clean, simple, and efficient inside your agents. Instead of the agent having to chew through long system prompts or analyze a bunch of tools in MCP servers, this is really about efficiency.
The agent will only read the skill name and description to determine how to use that skill. That's less context to process. That will result in faster processing and less tokens, hopefully resulting in less overall costs.
skills use what's called progressive disclosure, which basically means the agent only sees what it needs when it needs it. It will start with the name and description and go further if it needs to call that skill. And the part I like best about agent skills is that they're crossplatform.
You build an agent once in VS Code and you can use the exact same code in Claude, Cursor, and even anti-gravity. In this video, I'll walk you through building your first AI agent skill step by step in VS Code and then I'll show you how to run it using your agents. Let's dive in and get started.
Okay, I've now jumped into VS Code and let's get started creating our first agent skill. Now, the very first thing we need to do is go in the bottom left to go to the gear icon and go to settings. And we need to make sure that agent skills are enabled.
So, do a search for skills. And what you'll see here, give you a little more room, is the agent skills checkbox. Now, by default, this is not enabled.
You go ahead and check that checkbox. Now, depending on your version of VS Code, this might look a little bit different than this. Here's what it looks like in the latest experimental release of VS Code.
It's very easy to see what folders are being leveraged. You can turn them on and off. You can add custom ones here as well.
And you have the ability to see both workspace level folders and then the overall profile level folders. And we'll take a look at the profile level folders uh in just a moment. So for now, we're back in normal VS Code.
Make sure that checkbox is checked. And we'll go ahead and get started. I'm now over here in a workspace.
And [snorts] the first thing I want to do in my workspace is create a new folder. And as we saw right here, it needs to be called GitHub. Then inside there, I'm going to create another new folder called skills.
And then I'm going to go ahead and create a folder that's going to house our agent skills components. Now, this is going to be called game of chess. And then inside here, I create a folder simply called skill.
md. Now, what I've noticed is the casing is very particular and a lot of things about the file layout is very particular. So if you're creating a skill and you don't see it picked up, go ahead and check your spelling of the folder name has to match the name of the skill which we'll create in just a moment.
And all the casing has to be right and so it's very particular, very finicky. So keep that in mind. Now in addition to the skill.
mmd file, there's other types of files and folders that can be created under here. Underneath the skill folder, you can have a scripts folder that contains scripts that can be referenced. These would have Python, Bash, or Node scripts that you could have in there that could do additional work.
You could have another folder for references. So, if you had things like templates that you wanted it to use, they could be under there, as well as assets. Assets could be under there.
So, you could have logos and images as well. So, for now, we're just creating a basic skill here. So, it's not going to have anything under there.
So, now I'm going to go ahead and create this file. First thing I need to do is have a name. As I mentioned, the name has to match exactly what the folder is.
There's also other rules, limits, 64 characters and no double dashes and stuff like that. Then you need a description. The description is what the agent is going to use to determine whether or not it's going to look at this skill further at the time of execution.
So in this case, we just want this to be a skill to return a game choice based on the user wanting to play a game. Okay? And you're going to want to make this as descriptive as possible.
Basically, this is your selection criteria on when you want this skill to be executed. Now underneath here is where we'd have the logic in markdown of what we want the skill to do. So again, if you are referencing scripts, you would have the relative path to those scripts to have them executed.
In this case, we're going to use a skill to respond, would you prefer a good game of chess for the user asking to play any type of game with the system. Now, this is just a trivial skill here just to kind of get used to the skills and how you would create them. Obviously, this isn't real business value that you would see here.
So, with that, I'm going to go ahead and save this file. And that should be all we need to do in order for this skill to be picked up and leveraged. Now, real quick, let me show you the two skills that we have at my profile level.
So, I'm going to jump over to the explorer window. And under my user profile here, I have a copilot folder that I created. Has a skills folder in there as well.
And in here, you see I have two additional skills that I've created. A dad's joke and a positive affirmations. Just simply will return a joke.
And then if you're feeling down or getting frustrated with your code outcome, you can ask for a positive affirmation to kind of get a pickme up. So both of these are here. Thought these would be useful in any project so that they are profilewide.
So let's jump back to VS Code. Let's come over here to the agent and let's ask the agent what skills are available. So now let's ask this agent all available skills that it has access to.
I'm going to say list all available skills. Better be specific. agent skills you have access to.
And now this will give it just a moment. It's going to interrogate those folders and determine what skills it has access to. And as we expected, we get all three of the skills back.
The game of chess that we just created here, the dad joke and the positive affirmations. Now, we could run these at any time. I said, let's play a game of poker.
And we'll go ahead and see what this does. You can see here it's interpreting what it needs to do. It's reading that skills MD file.
And now it did echo back. Wouldn't you prefer a good game of chess? Which is exactly what we wanted.
And now let's try this with telling me a joke. Tell me a dad joke. And let's go ahead and see if that can find that as well.
And again, it's reading our skills. mmd file. And in this case, it does uh call out to an outside API.
It is asking me for permission to do that. and it's going to go ahead and call that out. And here's a dad's joke.
Um, so what this was intended to do was just kind of show you high level how to create your first agent skill. Um, some of the places I've used this for is validation of like workflow files, things that are in XML or JSON markup. Wanted to apply uh consistent validation of files that maybe other developers are working on.
Things that you wouldn't really want to put in system instructions because not everything we work on is going to be a workflow file. So I could put these in an agent skill and ask the agent skill to then apply our corporate best practices to these files that are created and it can go out and know to do that at just a skills skill by skill basis actually. And if you want more information about this, the full spec for the agent skills is defined here agentskills.
io. You can go to specification and it does outline everything that you can use and build inside these skills. Anthropic has some predefined skills for you as well.
As I mentioned, you create skills here in VS Code and use those same skills in other IDEs and agents as well. And here's a whole bunch of skills repository that you can get access to as well as the GitHub awesome-copilot account here has a tremendous amount of agent skills as well where you can go grab these and do things like uh work with PDF files and um all kinds of good um things that you'd want to leverage. everything's already built for you.
This site uh also has a lot of good uh system instructions if you're looking for system instructions and uh stuff like that. So, I'll have links to both of these, all three of these actually, in the comments below, as well as the simple markdown file that we created to create our game of chess. So, with that, hopefully you can jump out and make your own agent skills and save speed and token count and hopefully increase your productivity.
And with that, thank you for watching this video. Don't forget to like and subscribe. Thank you.
