Create Your First SKILL.md File (Make AI Agents Do Exactly What You Want) | #claude #antigravity

Hey everyone. Today I'm going to show you something that's changing how AI agents work. It's called agent skills. And trust me, this is way simpler and cooler than it sounds. So imagine this. You're working with an AI coding assistant, right? And you want it to do something specific like process PDF files or build a specific type of app, but the AI doesn't really know your exact workflow. That's where agent skills come in. Let me show you what this is all about. All right, so what exactly are agent skills? Think of them like instruction manuals for AI agents, but instead of being complicated, they're just folders with some files inside. Here's the basic idea. You create a folder, let's say called PDF processing. Inside that folder, you put a file called skill.md. That's it. That's the basic structure. Now, this skill.md file has two parts. First, there's some basic info at the top, like the name of the skill and what it does. Second, there are the actual instructions that tell the AI how to do the task. So, when your AI agent needs to work with PDFs, it can read this skill file and know exactly what to do. No guessing, no figuring things out, just clear instructions. And here's the cool part. These skills work across different AI tools. So you can write one skill and use it in Claude, use it in cursor, use it in VS Code, whatever. You're not locked into one platform. Now, let me explain how this actually works when you're using it. Step one, when your AI agent starts up, it looks at all the available skills, but it doesn't read everything. It just reads the name and description of each skill. This keeps things fast. Step two, you ask the AI to do something like, "Hey, extract text from this PDF. The AI sees that there's a skill called PDF processing and the description matches what you need. Step three, now the AI reads the full instructions from that skill. It loads the skill.md file into its context and follows the steps. And if the skill has extra stuff like scripts to run or reference documents, the AI can load those too, but only when it needs them. This is called progressive disclosure. Fancy term, but it just means the AI only loads what it needs when it needs it. Smart, right? So, you might be wondering, is this actually being used or is this just some random idea? Well, a bunch of major AI tools already support agent skills. First up, Claude and Claude Code from Anthropic. Claude Code is their command line tool for coding tasks, and it supports skills out of the box. Then you've got Cursor. If you're into coding, you probably know Cursor. It's that AI powered code editor that everyone's talking about. They support agent skills. GitHub is on board, too. Makes sense, right? GitHub is all about developers and AI tools now. VS Code, Microsoft's code editor, also supports this format. There's also tools like goose, letta, amp, open code, gemini cli, and factory. All of these are using the same agent skills format. The reason this matters is that you can create a skill once and it works across all these platforms. You're not writing different instructions for each tool. One format, many tools. That's the power of an open standard. All right. So, how do you actually create a skill? Let me explain it in a very simple way. Here I created an API design skill focused on building well ststructured REST APIs in Nex.js. In this skill, I added rules like always returning responses in a consistent format, defining reusable TypeScript types for API responses, validating input using ZOD, adding an API logger, and following other best practices for API development. Because of this, our AI agent automatically applies all these rules whenever it creates a new API. For example, when I asked it to implement a CRUD API for users, it immediately read the API design skill file first. Based on that design, it implemented the entire API following all the best practices and it even tested the endpoints using curl commands. So, why should you care about agent skills? If you're someone who uses AI coding tools or AI agents, this means you can teach your AI exactly how you work. You can create skills for your company's workflows, your personal coding style, whatever you need. If you're building AI tools, supporting agent skills means your users can extend your product without you having to build every feature. They can add their own skills and share them with others. Plus, since this is an open format, anyone can contribute to it. It's on GitHub. It's documented and it's designed to evolve with the community. This is still pretty new, but it's already being adopted by some of the biggest names in AI development tools. And I think we're going to see a lot more skills being created as people realize how useful this is. If you want to learn more, check out agentskills.io. That's where all the documentation lifts. You can also find example skills on GitHub to see how people are using this. If you found this helpful, hit that like button and subscribe for more AI and coding content. Drop a comment if you have questions or if you end up creating your own skills. I'd love to hear about it. Thanks for watching and I'll see you in the next one.