Anthropic's Official 'Introduction to Agent Skills' Course, Now a Korean Tutorial

A familiar scene in organizations that have adopted Claude Code: one engineer asks for a PR (pull request) review and gets back something that already matches the team's conventions. The engineer at the next desk uses the same tool and has to spell out the format from scratch every time. The difference between them is not ability. It is the difference between teaching the tool once and explaining yourself again every time.

Skills are the Claude Code feature that removes this repetition.¹ Put a repeated instruction in a SKILL.md file once, and Claude applies it automatically whenever it recognizes a related request. The problem is what to learn from. Most of the skills tutorials that turn up in search are blog summaries or video recaps. They leave you with fragments of knowledge rather than a framework you can apply across a team.

Modern Web Labs has published a Korean tutorial based on Anthropic's official course 'Introduction to agent skills'². The tutorial goes beyond translation. It covers ground the original could not, such as rules files, which Claude Code added after the course was released. This post walks through how the tutorial is put together and what an organization gains by learning from an official curriculum instead of a generic skills tutorial.

Six chapters, an official curriculum in under two hours

The Korean tutorial has six chapters. Each takes about 15 to 20 minutes, and the whole series runs under two hours. Chapters 1 and 2 are open to everyone; chapters 3 through 6 are reserved for Modern Web Labs clients.

Each chapter embeds the original lesson video unchanged, with a Korean summary alongside it so you can follow along without watching. We added everyday examples, such as ordering your usual at a cafe or filling in a meeting-notes template, so non-developers can grasp what a skill is. The series page also collects eight FAQ entries, among them "How is a skill different from CLAUDE.md?" and "What do I do when a skill doesn't trigger?"

What sets the official curriculum apart for an enterprise

There is no shortage of content on skills. Most of it stops at individual productivity: create a SKILL.md, verify the trigger, speed up your own work. It leaves out what an organization weighing team-wide adoption actually needs to know. Who should own which skills, which one wins when names collide, and how do you deploy and govern them across the whole organization?

One employee who is good at Excel does not standardize the company's report templates. One person using a tool well and an organization adopting that tool as a standard are different problems. The value of Anthropic's official curriculum is that it builds that organizational axis into its design from the start. Five points stand out.

• Priority hierarchy (Chapter 2): When the same skill name exists in several locations, priority runs Enterprise → Personal → Project → Plugins. The hierarchy puts organizational standards ahead of individual settings by design.

• Three deployment paths (Chapter 5): The curriculum spans every scale: committing to a repository to share with the team, distributing through plugins and marketplaces, and rolling out organization-wide through enterprise managed settings.

• Tool access restrictions (Chapter 3): allowed-tools limits which tools Claude can use while a skill is active. It is the control you need for read-only tasks and security-sensitive workflows.

• Managing context cost (Chapter 3): Progressive disclosure keeps SKILL.md light and has Claude read the detailed material only when it is needed. Operational knowledge of this kind shows up directly in token cost and response quality.

• Troubleshooting framework (Chapter 6): The curriculum walks through a diagnostic sequence for triggers that never fire, load failures, priority conflicts, and runtime errors. It also covers the skills validator and claude --debug.

If you are evaluating Claude Code adoption and building official skills for a team, these five points are your checklist. Take the material you are working from today and see whether it covers priority, deployment, and governance. If it does not, you are reading a personal productivity guide, not a team adoption guide.

What Modern Web Labs added

The official course has its gaps too. It cannot cover what Claude Code shipped after the recording. Rules files are the clearest example (official docs). They let you split CLAUDE.md across several files, and once you set the paths frontmatter, the instructions in a rules file load only when Claude works with files under those paths. That "load only when needed" behavior is what makes them easy to confuse with skills.

So we added a comparison of the two to Chapter 4. The deciding factor is the trigger. A skill matches on the user's request; a path-scoped rules file matches on the files Claude is working with. To answer "write a PR description," reach for a skill. To attach validation rules every time someone edits a file under src/api/, reach for a rules file.

We also documented when subdirectory CLAUDE.md files load. Claude loads the CLAUDE.md in the working directory and in every directory above it at session start, but a subdirectory CLAUDE.md loads only when Claude reads a file in that subdirectory. So pulling out the instructions that apply to a single module saves context much the way a path-scoped rules file does. A single diagram lays out when the root CLAUDE.md, a rules file, and a subdirectory CLAUDE.md each load in a monorepo. Chapter 5 adds a directory-structure diagram for distributing skills as plugins.

Every addition is labeled "Modern Web Labs extra tip" in the text so it stands apart from the original. Readers can trust both sources only when they can see where Anthropic's official curriculum ends and our additions begin.

Where to start

Chapter 1, "What are skills" and Chapter 2, "Creating your first skill" are live right now. If skills are new to you, start with the examples in Chapter 1; if you already know the concept, jump to Chapter 2 and write a SKILL.md yourself.

Chapters 3 through 6, which cover deploying and governing Claude Code across a team, are available to clients who have worked with Modern Web Labs on consulting or training. For skill design and governance shaped to your own team, see our consulting overview. When the premium chapters open to the public, the newsletter will carry the news first.

1. Claude Code official documentation, Agent Skills. The primary source for skill structure and configuration options.

   ↩

2. Anthropic, Introduction to agent skills (Skilljar). The original course behind the Korean tutorial.

   ↩