Transcription
This is a graph of the growth of Skills, a new standard for interacting with AI agents, which is currently going crazy with hype. When I started studying this topic, I honestly didn't understand why all this was necessary and how to apply it in real projects. After all, it seemed like everything had already been invented: we have prompts, we have a knowledge base, why do we need any additional skills? But after personally trying dozens of skills, studying all popular marketplaces, watching dozens of videos on this topic, and reading tons of documentation, I realized how cool this thing is and how much it simplifies life. And not only for developers, but for everyone who uses artificial intelligence in their daily tasks. Seriously, I had an epiphany and became an ambassador for skills. Now, in this video, I've tried to give you all the information about skills. What they are, why they are needed, how to install them, how they differ from prompts, where and how they can be applied, how a neural network understands when to launch which skill, and everything, absolutely everything you need to know on this topic. We will also analyze several real-world cases from no-code development to motion design and copywriting, where skills can actually be applied. And, of course, we will create our own skill from scratch, so that you can then learn to create your own skills. And at the end, we will discuss the future a little, where all this is heading, and how the world will change in general. In short, I put a lot of effort into this video and tried to make the most detailed guide on Russian-language YouTube, while removing all the fluff, leaving only the substance, making it concise, short, super clear, but at the same time sufficiently dense. I hope you appreciate it. Please like. Subscribe to the channel. You are on the Yaishny channel. Let's go.
To create some skills, you may need external APIs of various neural networks. And for this, it is very convenient to use the Polza AI service. This is an aggregator where hundreds of the most popular models are collected, everything is connected and configured very easily via API. The service is cool because you don't need to register separately on all AI platforms. You have a single account with any generative models for images, text, and video. You can top up your balance with a Russian card and not worry about VPNs, blocks, verifications, various foreign SIM cards, and other dances with a tambourine. The guys have Russian-language support, which responds very quickly. There are no subscriptions or monthly charges. You just deposit money and use it, track your balance in your personal account and top up when it runs out. If you are building your own projects, creating automations, or generating content, I highly recommend trying it. The service is compatible with OpenAI API and easily connects to any no-code tools or automation software like ANY. And there is documentation, very clear in Russian. This is what the personal account looks like. Everything is very simple here. You can view usage statistics, request history, remaining balance. In general, everything is in Russian, very clear and simple. By the way, for those who don't want to bother with the API, there is a playground section in the personal account. Here, in chat mode, you can interact with hundreds of models for text, images, and video. Look, there are really a lot of models. Everything is as usual here. Write your prompt in the chat, press Enter, and get the finished result. I will leave a link to Polza AI in the description. Try it.
In 2025, Anthropic first introduced Skills in their Claude. They did this to solve the problem of model hallucinations. The problem is that when you give a neural network a large prompt with a description of roles, tasks, and step-by-step instructions, the context window is quickly forgotten, and very soon the neural network starts forgetting what you wanted from it at all. In other words, it starts hallucinating. And using these huge prompts is not very convenient. You have to constantly copy-paste them from one chat to another. Skills work a little differently. Here, the AI itself decides which skill to launch and when, based on your request. Thus, you don't pass kilometer-long prompts to the neural network. The context window remains free, you save money on tokens, and the neural network hallucinates less itself. A little later, I will show you exactly how this works, how the neural network decides which skill to launch and when. So, skills have proven to be so cool that they have transformed from a small feature of Anthropic into an international standard that other corporations and model developers are now implementing. Recently, Google added support for skills in its Antigravity, and, for example, Codex has focused on skills in its interface.
And to give you a complete picture, we need to make one short digression. And this digression is about a completely new way of interacting with AI. Very unusual, but at the same time super effective. And I promise, immediately after this, we will move on to skills themselves, we will get to the substance. But this digression is needed so that you have a complete picture, so that you understand the philosophy of skills and their structure. Look, before the advent of skills, we were, in a way, hostages of platforms. That is, you have, for example, an account in ChatGPT, there you have your projects, different folders, custom GPTs, you upload your knowledge base, your prompts, and instructions there. And all this lives somewhere on the corporations' servers. Thus, you are tied to their interface, their subscription. Moving everything from one service to another is quite difficult. Skills, on the other hand, are just folders and files created according to a special structure. They are stored on your computer just like regular text files. And if you get tired of, for example, paying a subscription for Claude, you simply take the folder with your skills, open it in any other software, be it Antigravity, the same Codex, or Winsf, VS Code, anything, and just use it without any complex, long migrations. Any software that supports skills will immediately understand everything, load it, and work as it should. And all the software I listed above is designed for developers. And in general, skills were initially created for developers to write code better, to draw designs, to test, and so on, and so on. But personally, I use skills more and more in my daily tasks. This includes copywriting, some research, brainstorming, writing reports, creating presentations, publishing content, and much, much more. I simply have a folder with personal skills, where everything is described as I need it, and I don't care what software it is. Because I use the software simply to communicate with the model and launch these skills at the right moment. In my opinion, it's very convenient because I can build entire chains of workflows this way. For example, for YouTube, I have a research skill that gathers some information. Then there's a skill that helps me write a script. Then a skill that turns all of this into a video structure. Then illustration, publication, maybe even a skill that helps with editing, and then a skill that writes posts for all of this to a Telegram channel. In general, entire chains of workflows, skills that launch one after another.
The closest analogy here is The Matrix. Remember, at the beginning there was a moment when Neo was loaded with various skills like kung fu and so on into his head? Well, it's very similar here. You simply create a skill, put it into your AI agent. And this skill contains a description of the skill itself, the process, and some additional files, scripts, so that your agent can directly perform the tasks you need. As a result, you load your agents with new skills, they become smarter and smarter, and they solve more and more of your tasks. It's like training your employee or assistant. In short, what am I leading up to? The very format of how we interact with AI agents is changing now. And for now, it's something complex and incomprehensible in the language of developers, but if you can overcome this entry barrier right now, understand this topic, you will get a huge boost in productivity and a great advantage.
Well, now let's move on to the skills themselves, I'll show you what they are. We'll try to create our own skill, install a few ready-made ones, and look at different use cases. As I said, a skill is just a folder with files created according to a specific structure. If you look at the documentation, you can see what this folder consists of and what should be in it. Let's immediately try to create our own simple skill and analyze everything in parallel. Create a new empty folder on your computer. Inside it, create another new folder. Type the name dot skills. Press Enter and choose to use the dot. If, by the way, you get some error here and this folder doesn't appear, then on Mac you can use the combination Command Shift Dot to hide and show these system folders. Open this folder, and inside it, create another folder called Skills. And I'll tell you right away that you won't have to do this all the time. You only need to do it once for educational purposes. Then we will download skills from other marketplaces. So, inside the Skills folder, create another folder with the name of our skill. Let it be a skill for copywriting on Telegram. Let's call it Telegram Writer. And it's time to open this folder in Antigravity to create a file there. Well, you can open it in your editor that you use. So, we opened all this in Antigravity. We have the Agent Skills folder and our Telegram Writer. And inside it, create a new file named skill.md. The main prompt for our skill will be stored here. Let's put the prompt that I prepared in advance here. And what do we see here? We have this header of this document. Here we have name, description, and allowed tools. Name is Telegram Writer. It is named the same as the folder with our skill. Description creates posts for a Telegram channel with in-depth topic research. And allowed tools are search, various MCPs. And, you can add any tools here that you allow your AI agent to perform. And then below is just a regular prompt. Note that the name should be written like this, with a hyphen. This is indicated in the documentation. I will also leave a link to the doc in the description. There is also a description section, where we have the description of this skill. And it is precisely based on this that the neural network understands that this skill needs to be launched. That is, here you need to specify keywords by which the neural network should understand from the correspondence with you that this skill needs to be launched now. For example, here I can say that always use when you need to write a post, posts for a Telegram channel. And then we just have the prompt, our standard instruction.
Now let's try to ask the model to write some post and see how it behaves. Press Enter and see what happens. You see, it saw our skill.md file that we created, read it itself. That is, it understood from the context that we need a Telegram post, found the right skill itself, read it, and went to study what is in this skill. And now it will use this prompt to write our post. And look, as written in the prompt, the first round, clarifying questions, it starts asking me clarifying, uh, questions. And let's try to answer them. And it will now go to step two. In theory, it can now launch a search on the internet to find some additional information. Here it gave me a hook and theses. It didn't go to search the internet. Well, that's its decision. Thesis, thesis, thesis. Everything is exactly as per the technical specification. Okay, let's agree on what the final post will be. Well, you see, right? I specifically made such an example. It differs from what the neural network would simply give us as a post. Of course, the prompt needs to be refined, add your own examples, and we will do that now, expand it a little, but the model is already doing something different from its standard behavior.
So, we've seen that our skill was successfully launched, but this is just the cherry on top. The skill.md file itself recommends making it no more than 1.24 characters, and it's better to store all other information in other files. So, you can create a references folder inside the skill and a reference.md file. This is a special catalog for your various examples, anti-examples, and various hints that you can give the neural network. Let's insert more info about our Tone of Voice here and see how the result changes. Let's write a new post on the same topic. As you can see, it has now also picked up the reference.md. That is, it read the skill, and the reference, and now it will take into account my tone of voice examples in the text. Look, cool, right? It asked a few questions and went to search for additional information because I asked it to mention a marketplace for a ready-made skill in the post, and it, accordingly, uses the search to find more examples and, accordingly, enrich the post with some facts. And here it gives me theses. And look, look what's interesting now, that it, first of all, took my analogy about The Matrix, which is even a little scary, and wrote very similarly to how I write and speak. Well, let's even look at an example, how it looked here. We are used to chatting with AI, but Technology Skills is changing the game. It's like loading skills in The Matrix. Let's see what we had in the previous hook. Antigravity is now not just a universal assistant constructor with which you can assemble an ideal employee for any task. Well, in short, there is indeed a big difference here. It took into account my examples. And now let's write the post.
Another interesting idea here is that you can create a file, for example, my.md. Here you can create a new file in the references folder, write my.md, and store some context about your life here. For example, I have my personal skill that I use, it says this: I update it about once a week and simply dictate what has happened in my life. For example, I have now flown to Uzbekistan for a week. Here I need to go to the hospital because the treatment I need is very expensive in Serbia. And here I am in field conditions making a video about skills. And also, I will have a child soon. And I also have a dog. And I also started a second channel. And considering all this. And I can record such context once a week, I can set up automation to write it through a Telegram bot. And it turns out that every time the neural network writes a post for me, it can take this context into account, because I can add the file with references here. Take into account the context from the my.md file and write the post according to this context, that is, give examples from there, and then it will give you a very cool result. So, let's see what's here. Remember the scene from The Matrix where Neo loads the IT program in a second? With Technology Skills, it's the same now. You don't need to train it with context every time and write simple prompts. Just throw the file into the folder and it's a pro in your stack. Well, I don't really talk like that. It's too intrusive, or something. But the tone itself has changed, of course. All this can be refined, as with any prompts, and brought to some cool result.
Well, we've created our first skill. Congratulations. Now let's look at the marketplaces for ready-made skills. I'll show you what cool skills are there and which ones you should install first. The most popular marketplace is called Skills. There are now over 50,000 skills for all occasions. For example, a separate section of skills for marketing. Look, there's a pricing specialist, a paid advertising specialist, and an SEO promotion specialist, tons and tons of everything. A skill that can work with presentations, a skill for Excel spreadsheets, a skill for designing Android applications, and thousands, thousands, thousands more skills. By the way, note that some of these skills require a paid API. That is, some guys have already taken the initiative and are making money from this. For example, these YouTube skills, for generating previews, require access to a paid API.
Let's now install a few skills from here and see how to use them. And the first skill that I recommend you install is the Skill Creator, that is, a skill for creating skills from Anthropic themselves. It will teach your model to create correct folders, the entire structure, so that you don't do it manually, but simply write the necessary prompts, and the agent will create everything itself, everything that is needed. We find its page in the marketplace, open it, copy this line, return to Antigravity, open the terminal, and paste this line here. And we see that such an installer has launched. Now it will clone the repository and offer me to install. So, look, here you can use the spacebar to select your software where to install it. That is, you can press the spacebar and see, it turned green, and it installs for Open Claude, for Claude, and for, for example, Antigravity separately. But I use Antigravity. I don't select anything here, I just make sure that there are no green dots here, because some default options are already selected here, and Antigravity already understands where and how to use it. If I select, for example, Claude Code here as well, it will create an extra folder .Claude here and will do it according to Claude's structure. In short, you can clutter it up quickly here. Therefore, I just do it like this. Let me make it a little wider. I just skip everything here so that nothing is selected, and just press the Enter button. And now. Let's close this. Now I choose how to install the skill. It can be installed as a Project, meaning it will only work locally in this folder, or globally, if you want the skill to work in all folders, in all projects that you launch. I will install it as a Project. Here I just press Enter. Enter. And, accordingly, our skill is installed. Everything, it says that the skill is ready. I will click refresh here and see that a new skill, Skill Creator, has been created.
Now let's create a skill for drawing illustrations. Now that we have Skill Creator, we can do it through AI, without creating any folders manually. Let's write a prompt. I need to create a skill for creating illustrations. I want to make illustrations with text that I specify. I use Nanobana. I use Nanobana, which is in Google itself. That is, write in the prompt just generate an image. What is the style of these pictures? It should be 16x9. It should be a dark gradient background, like blue gradient, more black, so something black, grainy, gray, like that. And, uh, what should the skill do? I send it the inscription for which I need to generate a picture, and it generates a picture with this inscription. The inscription should be written in a thin, small font in the bottom right corner of the screen. This is the prompt we set. And now let's see. This thing should launch Skill Creator. And this Skill Creator will create its own skill for us.
Okay, done. I created the Illustration Generator skill. Let's open it and see. Here I open this skill and see that it creates illustrations with text in the style of Nanobana. Only let me say that I want it to create a folder where it will put these pictures. You see how easily we can create skills, without worrying at all, just through no-code. Now it will make changes to my document, and we will try to test it all. I specifically set such a prompt where there is a specific instruction on what and how it should be located and look. Here it corrected my skill and said that it is necessary to determine the current date, create a folder and put a file there. Okay. Now let's click here and try to launch our skill. Oh, I said presentation, but in fact, it should be an illustration. Here, send it. And you see, it launched the skill. We asked to create an illustration. The illustration creation skill was launched. And it's going. Here it is creating a folder where it will put our illustration. Let's collapse all this here. And if we click refresh here, we will see that our images folder has appeared. Here is a folder with the date. And now our picture is being drawn. By the way, for those who didn't know, in Antigravity, you can draw illustrations with Nanobana. Right here, through chat.
Okay, done. Let's open the picture and see what's there. Of course, well, this is basically the technical specification, but the text is too small. I, well, it's clear that it can be improved, but the basic task is done, right? Well. You must agree that it turned out not bad from the first prompt. Well, it's hard to argue with that. Now we can, of course, refine all this, bring it to perfection, but I won't torture you. And note that it put our image in the folder, as we asked it to. And here you can ask for anything. You can ask it to upload these folders to Google Drive, make a presentation out of them, for example, generate five pictures, make a presentation out of them or make a PDF, or send it, I don't know, make a mailing list to the client base. Well, in short, tons and tons of use cases for what you can ask skills to do. At the same time, you can also create skills through prompts using no-code. This, in my opinion, is really very cool. And very quickly, in just 5 minutes, we assembled a full-fledged skill for creating illustrations in our style. What's most interesting is that it will create all illustrations in this style of ours or in the style we specify. Here you can attach, for example, an archive with your photos and make carousels. Yes, you can do anything. You can overlay text, even create videos. The use cases are endless. When I was talking to you in the first part of the video about the new way of interacting with AI, that we now do it through folders on our computer, it's a bit more cumbersome than doing it in a web interface, but notice what freedom you get. You can do anything with these files further, upload them to Notion, to Google Drive, integrate them into a CRM system, integrate them with your workflows. In short, a million things where this can be applied. I think this is wow, in general. Just wow.
And now I will show you the coolest skill that humanity has ever seen. This is the Remotion skill. This is a skill that can create videos, but not those videos like VO3 or Sora, which are generated through prompts, but full-fledged motion animations. That is, this skill programs videos. Remotion is a program for programming videos. For example, some inserts in this video were created using the skill. I didn't ask any freelancers to do this. We go to the skill page, copy this line, return to Antigravity, open the terminal. You can create a new terminal here. Let's create our skill. Everything is basically the same here. Project TeamLink. Yes. Done. And now let's ask the neural network to install everything we need to use this Remotion skill. Here it is. Remotion best practice. Let's ask the neural network to install everything we need to use Remotion. So, when everything is installed, we can simply follow the instructions that the agent gives us. Just go to the test project folder and press npm start. Remotion will start. Remotion, as I said, is a service for creating videos. And it is available to us at the address localhost 3000. Now it will load, and we will see what the neural network has created for us. It has created a video called Hello World. Let's try to play it. You see, we have a full-fledged motion animation, right, which we can simply press the render button and create a video.
But let's try to create something more interesting, more complex, right? Let's imagine that I need to create some beautiful intro for this video, like a big inscription Skills and below it a subtitle: "A new way to interact with AI agents that will change everything." Now create a new video for me. I need an intro for my video about Skills. It should be a YouTube format video, a horizontal video, with a dark background. Possibly a dark background with a gray gradient line. And there should be two inscriptions. The first inscription is Skills. It should be written in a large, bold font and appear very epically, so that it just bursts onto the screen. And then the inscription "A new way to interact with AI agents that changes everything" should be typed. This is the kind of video I need to create. An intro and make an incoming animation and an outgoing animation. Also, at the end, make a short purple glint, so that there is a glint and all inscriptions disappear. Done. And let's see. Here we have a new composition called Skills Intro. And it already has something more serious here. Look. Bam! A new way to interact with AI agents that changes everything. And such an intro. In my opinion, it turned out very cool. And in one prompt, friends, this is very cool. It's really very cool. Imagine what possibilities open up. Six months ago, I said that it would take another year or two for neural networks to conquer editing. It seems I was a little wrong. Half a year is enough for us. Really, you can now create motion design with voice. Look how cool. Create your own design system and overlay these subtitles, make these inscriptions. All this can be done in literally a few clicks. Just by recording, damn it, tasks for artificial intelligence. By the way, write if you are interested in a separate video on Emotion. I will also make a detailed guide.
Well, friends, I don't want to overload you further. The video turned out to be quite dense for YouTube format, and most likely, something like this is going on in your head. >> I didn't understand. Well, it's very interesting. It's always like this when some new technology comes along, IT specialists and geeks use it first. They do it the way it's convenient for them. Like me, for example. But over time, the technology comes to mass adoption, becomes more accessible and understandable. And I think that a whole revolution in skills awaits us, because, well, you can see for yourself how huge the potential is. For example, OpenAI in its Codex software has already focused so much on skills in its interface. Look how cool it looks here. There are already some avatars, names, descriptions. All this is installed in one click. You don't need to open any terminal. Everything is very cool, trendy, youthful, and understandable for ordinary people. By the way, I recently recorded a separate review of Codex on my second channel, Yaishny Live. There I post lighthearted unboxings and service reviews. If you are interested, I will leave a link to the video in the description. Go to my second channel. I'll be happy to see you there too.
So, what are my thoughts on this whole topic? I'm a maximalist in this regard, and I might say complete nonsense from your point of view, but it seems to me that in a year and a half to two years, we won't be buying any courses, but we will be buying skills for our AI agents, we will be loading them with skills, there will be paid skills, free skills, marketplaces for skills and applications for AI agents will appear. Possibly by this time, we will already be loading skills into our home robots. Who knows what awaits us, really. Honestly, I'm not surprised by anything anymore. On my Telegram, by the way, we were recently joking about this. By the way, if you're interested, go and subscribe. I write there quite rarely, I write good, high-quality posts, I don't post any junk there, so there's such activity. Look how many reactions, comments in the chat. I regularly respond there, so go to Telegram, it's very fun there and there's content that is not posted on YouTube.
Please like this video, subscribe if you liked the content. I really tried very hard to make a detailed guide on skills, without overloading it or dragging out the timing. In general, this is my first experience creating such a complex video. And if you are interested in individual no-code training, then write to me privately. I will also leave a link to my Telegram in the description. In March, February-March, I am ready to take a couple of people for personal training. We will create your own track for your tasks. We will create the project you need together, solve what you need to solve, and learn at the pace you need, so that everything is clear to you. And don't forget to try Polza. It's a really cool service, link in the description. That's all, thank you everyone, hugs to all. Bye-bye.