Introduction to Agent Skills

Agent skills are reusable instruction sets packaged as SKILL.md files that shape how Claude Code behaves. Think of them as specialized knowledge modules that teach Claude how to handle specific tasks, follow particular conventions, or work with certain technologies.

Skills are fundamentally different from other AI customization approaches:

Skills vs. System Prompts

System prompts are monolithic instructions baked into an AI application. They set the baseline behavior for every interaction but can’t be easily swapped out or combined. Skills, by contrast, are modular and composable — you can mix and match multiple skills in a single session, and different skills can activate for different tasks.

Skills vs. MCP Servers

MCP (Model Context Protocol) servers provide tools — external capabilities like reading databases, calling APIs, or accessing file systems. Skills provide behavioral instructions — they tell Claude how to use those tools, what patterns to follow, and what conventions to respect. Think of MCP as expanding what Claude can do; skills define how Claude should do it.

Skills vs. Projects (Claude.ai)

Claude.ai’s "Projects" feature lets you create custom instruction sets with uploaded knowledge. These are powerful but platform-specific — they only work in the Claude.ai web interface. Skills work in Claude Code CLI, can be version-controlled alongside your codebase, and can be distributed through marketplaces for others to use.

Brief History

The agent skills concept evolved from Cursor’s .cursorrules convention — a single markdown file at the project root that influences Cursor AI’s behavior. Claude Code’s SKILL.md convention builds on this foundation but adds directory-based structure (skills can include templates, scripts, and reference files), multi-scope resolution (global, project, or marketplace), and a plugin ecosystem for sharing skills across the community.

Claude Code resolves skills from three distinct scopes, each serving different use cases:

Precedence Rules

When multiple skills exist with the same name across different scopes, project-level skills override global skills. Additionally, explicitly triggered skills (using /skill-name) take precedence over implicit activation based on context.

This hierarchy gives you fine-grained control: use global skills for your personal coding conventions, project skills for team-wide patterns, and marketplace skills for community best practices.

Let’s walk through installing and using a skill from Anthropic’s official marketplace. We’ll use the brand-guidelines skill as an example — it helps ensure visual design elements align with brand standards.

Step 1: List Available Marketplaces

First, see which plugin marketplaces you have access to:

You should see anthropic-agent-skills among the results — this is Anthropic’s curated skills collection (sourced from the anthropics/skills GitHub repo).

Step 2: Install the Plugin

The brand-guidelines skill is part of the example-skills plugin. Install the entire plugin:

This downloads all the example skills — including brand-guidelines, skill-creator, frontend-design, and others. Claude Code will make them available for all your sessions.

Step 3: Start Claude Code

Start an interactive Claude Code session so you can try out the skill:

Step 4: Explicit Trigger

The most reliable way to use a skill is the explicit trigger — a slash command. You don’t need to memorize the full namespaced name. Just type /bran and press Tab — Claude Code will autocomplete it to /example-skills:brand-guidelines for you.

Try it:

This expands to:

The full name example-skills:brand-guidelines is the namespaced form: plugin-name:skill-name. Tab completion means you rarely need to type it out — just enough characters to uniquely identify the skill.

When the skill activates, you’ll see it listed in the skill usage area of Claude’s response. The output will reference Anthropic’s specific brand colors, typography, and design standards — not generic color theory advice.

Step 5: Implicit Trigger

Skills can also activate automatically when Claude’s assessment of your prompt matches the skill’s description. This is called implicit triggering — and it depends on how well your phrasing matches the skill’s description.

First, try a generic prompt that doesn’t mention Anthropic:

Claude will likely respond with generic color theory advice — asking "which brand?" or offering general contrast analysis. The skill didn’t activate because nothing in your prompt matched its description closely enough.

Now try again with a prompt that includes the keyword the skill is looking for:

This time the skill should activate — Claude will reference Anthropic’s specific brand palette, typography, and design standards instead of asking for context.

The difference? The skill’s description says "Applies Anthropic’s official brand colors…​" — the word "Anthropic" in your prompt is what triggers the match. This is an important lesson: implicit triggering depends entirely on how well the prompt matches the skill’s description. You’ll practice writing effective descriptions when building your own skills in Module 2.

Step 6: Compare With and Without

To see the difference clearly, try the same question without invoking the skill:

Then with the skill:

Without the skill, Claude gives general design advice. With it, Claude applies Anthropic’s specific palette and guidelines.

Step 7: Verify Installation

You can confirm the plugin is installed by checking the installed plugins config:

You should see example-skills listed among your installed plugins. You can also browse the cached plugin contents to see the skills it includes:

In this module, you learned three foundational concepts:

You also installed your first skill from the official marketplace and saw how it shapes Claude’s responses differently than default behavior.

Next, in Module 2, you’ll build your own skill from scratch — learning the anatomy of a SKILL.md file and how to package instructions Claude can follow.