If you use an AI agent every day, you have probably felt this. You open a fresh session and paste the same long prompt you pasted yesterday. You re-explain how you like research done, what your brief should look like, which sources count, what tone to use. The agent is capable, but it does not know how you work. So every task starts from zero.
Agent Skills fix that. A Skill is a small, reusable file that teaches an agent how to do one kind of task the way you want it done. You write it once. After that, the agent loads it on its own when the task comes up, and you stop re-pasting instructions. Skills are quickly becoming one of the core building blocks for reusable AI workflows.
This guide walks through how to build your first one. It mirrors a hands-on lab where you build a real deep-research skill, extend it, and run it on actual source material inside a live sandbox in your browser. If you would rather build than read, the lab is the fastest path in.
What is an Agent Skill?
A prompt is something you type each time. A Skill is something the agent reads automatically when the situation calls for it. It lives as a small markdown file, usually named SKILL.md, inside a folder in your project. The file has two parts. A short header that names the skill and describes when to use it, and a body that spells out the actual steps the agent should follow.
The header is what makes the magic feel automatic. When you ask the agent to research a topic, it scans the descriptions of the skills it has, notices that your deep-research skill is a match, loads it, and follows the playbook inside. You did not invoke anything by hand. You told the agent ahead of time when to reach for the skill, so it picked the right one on its own.
That is the whole idea. No framework, no SDK, no build step. A Skill is a markdown file in a folder that turns a general-purpose agent into a specialist for a task you care about.
Why build your own
General agents are good at general things. Your work is not general. You have a specific way you research a market, a specific structure for a brief, a specific set of sources you trust. A Skill captures that and hands it back to the agent every time, without you having to remember the details.
The payoff compounds. Write a skill once and it works across every future session. Share it with a teammate and they get your exact process, not their best guess at it. Improve the skill in one place and everyone who uses it gets the upgrade. Reusable AI workflows are how small teams get leverage out of agents, and Skills are the simplest way to start.
There is also a learning payoff. Building a skill forces you to write down how you actually do a task. That alone tends to make the work sharper, whether or not an agent ever runs it.
What you will build in this lab
The lab is built for beginners. It is four hands-on labs that run in a live agent sandbox in the browser, so there is nothing to install. The course description sums up the goal. Learn how to create reusable Agent Skills that turn an agent into a specialist for any task you care about, whether that is research, writing, analysis, or anything else. By the end you will have built and tested a real, reusable skill.
You start small. The first lab has you build the smallest possible skill so you can see the file format with nothing else in the way. You learn what goes in the header, what goes in the body, and how the agent decides to load it. It is deliberately tiny, because the point is to understand the shape of a SKILL.md before you put real work inside one.
From there the capability grows. In the second lab you encode a research methodology. You build a reusable deep research agent skill that teaches the agent to research any topic thoroughly, not just the one example in front of you. This is where a skill stops being a toy and starts being something you would actually reach for. You are taking a repeatable process out of your head and writing it down in a form the agent can follow every time.
The third lab adds reusable resources. A skill is not limited to a single file. You extend yours with a template for the output and a source-format file that describes how your sources are structured, then reference both from SKILL.md. This is a big step. It teaches the agent not only the steps to follow but the exact shape of the result you expect, which is usually the difference between an answer you can use and one you have to rewrite.
The fourth lab puts the whole thing to work on something real. You run your deep research skill on four Costco strategy sources and produce a structured brief. You watch the skill you wrote pull themes across separate documents, organize them against your template, and hand back a brief in your format. By the end you have a working skill and proof that it does the job on material it has never seen before.
Read as a sequence, the four labs are a single arc. Understand the format, encode a real method, give the agent the resources it needs to match your standards, then run it on real inputs and see the output hold up.
What's under the hood
You build everything in your browser. There is nothing to install and no API key to manage. Each lab spins up a real Linux sandbox running the Claude Agent SDK, and you work alongside the agent in a live workspace.
The skill format itself is plain text. A skill is a single SKILL.md file with a short YAML frontmatter block and a Markdown body, stored at .claude/skills/<name>/SKILL.md. That is the same format agents like Claude Code load on your own machine, so the skill you write here also works outside the lab. As you go, automated checkpoints inspect the files in the sandbox and turn green when your skill is correct, so you always know where you stand.
Who this is for
You do not need to be an engineer. If you have used Claude Code or any agent and felt the friction of re-explaining yourself, you have everything you need to start. The lab assumes no prior experience with Skills and runs entirely in the browser, so you are building within minutes rather than configuring an environment.
It is a good fit for researchers, writers, analysts, founders, and anyone whose work has a repeatable shape worth handing to an agent. The Costco example is about strategy research, but the pattern carries. Once you can write one skill, you can write a skill for whatever task you do most.
Start building
The best way to understand Skills is to build one and watch it run. In under an hour you will go from the smallest possible SKILL.md to a working deep research agent that produces a real brief from real sources. That is a skill you can keep, improve, and reuse on your own work the next day.
Want more like this? Check out our courses, explore the rest of our labs, or join the community to ask follow up questions about this post.
