Claude Code Skills: Folder, Install, and Examples

What Claude Code skills are

A Claude Code skill is a reusable procedure stored as a folder. Its required SKILL.md file tells Claude what the skill does, when it applies, and how to carry out the work. Optional scripts, references, and assets can make the procedure deterministic or give it domain context.

This is different from keeping every rule in CLAUDE.md. Project instructions stay active as general context. A skill is discovered and loaded for a matching task, so a detailed release checklist or debugging method does not consume attention during unrelated work.

If you are new to the concept, read What Is a Claude Skill? first. This page focuses on Claude Code setup and usage.

Where the Claude Skills folder lives

For a project-specific skill, use this structure:

your-project/
└── .claude/
    └── skills/
        └── release-check/
            ├── SKILL.md
            ├── scripts/       optional
            ├── references/    optional
            └── assets/        optional

The important path is .claude/skills/<skill-name>/SKILL.md. Keeping it in the repository makes the workflow available to collaborators and lets normal version control review every change.

User-level skills are appropriate when a workflow should follow you across projects. Plugins are another distribution route: they can package one or more skills together with other Claude Code extensions. The exact choice depends on whether the procedure belongs to one project, one user, or an installable bundle.

The minimum SKILL.md file

A useful starting point looks like this:

---
name: release-check
description: Verify tests, versioning, changelog, and release artifacts before publishing a package.
---

# Release Check

1. Read the repository release instructions.
2. Run the required test and build commands.
3. Verify the version and changelog.
4. Inspect the package contents.
5. Report blockers before publishing.

The description is not decorative metadata. Claude uses it to decide whether the skill applies. Name the trigger, expected work, and output instead of writing a broad phrase such as “helps with releases.”

For a fuller authoring workflow, use How to Create a Claude Skill and the SKILL.md format guide.

How to install official Anthropic skills

Anthropic publishes a public skills repository. Its current Claude Code instructions use the plugin marketplace flow:

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

The repository also documents an example-skills plugin. Check the README before installing because package names and supported surfaces can change.

The first command adds the repository as a marketplace source. The second installs a plugin containing a group of skills. This illustrates the practical relationship: the skill is the reusable workflow, while the plugin is the installation and distribution package. The focused Skills vs Plugins comparison explains the boundary.

How Claude Code activates a skill

Claude Code can inspect lightweight metadata before loading a skill’s full content. When your request matches the description, Claude can open the instructions and then read supporting files as needed. Anthropic describes this pattern as progressive disclosure.

Activation quality therefore depends on scope:

• “Review code” is broad and may fire at the wrong time.
• “Review a pull request for correctness, regressions, security risks, and missing tests; report findings by severity” gives a much clearer trigger.
• Two skills with overlapping descriptions may compete, so give each one a distinct job.

To confirm activation, ask Claude which skill it is using and why, then inspect whether the expected procedure appears in the work. Do not rely on the folder existing as proof that every request uses it.

Claude Code skill examples worth studying

Good examples solve repeated procedures rather than storing a one-off prompt:

Example	What makes it a skill-shaped problem
Systematic Debugging	Reproduce, isolate, test a hypothesis, and verify a fix in a stable order
Test-Driven Development	Preserve a repeatable red-green-refactor loop
Frontend Design	Apply a reusable visual and implementation standard
Using Git Worktrees	Set up isolated work safely with predictable checks
Skill Creator	Build and validate other skills using a consistent structure

These directory entries are source-based profiles, not hands-on reviews. They expose the repository, exact path, license, maintenance signal, and platform evidence so you can inspect the candidate yourself.

Skill, command, or project instruction?

Use a project instruction when the rule should apply to nearly every task in the repository. Use a skill when the detailed procedure belongs to a recognizable class of task. Use an explicit command when the user should decide exactly when the workflow starts.

Modern Claude Code skills can support both automatic discovery and explicit invocation. Legacy command files may still work, but the skill folder format supports richer resources and clearer activation metadata. See Skills vs Commands for edge cases.

Security checklist before installing

Treat a third-party skill as code and instructions from another author:

1. Verify the repository owner and exact folder, not only an aggregator page.
2. Read SKILL.md and every script the workflow can execute.
3. Check package installation, shell commands, network calls, and environment-variable access.
4. Confirm the license and recent maintenance state.
5. Start with a non-sensitive repository or sample input.
6. Remove or disable the skill if its behavior is broader than its description.

Repository stars are not a per-skill quality score when many skills share one repository. Agent Skill Picks never assigns the repository’s total stars to an individual folder. The methodology explains the source and indexing gates.

Troubleshooting

Claude does not discover the skill

Confirm the file is named SKILL.md, the folder is under the expected skills path, and the YAML frontmatter parses correctly. Make the description specific enough to match the request. Restart or open a new session after changing installation state when the client requires it.

The wrong skill activates

Reduce overlap between descriptions. Add the task type, expected input, output, and exclusions. A narrow skill that activates reliably is more useful than a broad skill that competes with every request.

A plugin skill is unavailable

Confirm both steps: the marketplace source was added and the intended plugin was installed. Inspect the installed plugin’s components and namespacing rather than assuming every repository folder becomes a command.

Primary sources

• Claude Code: Extend Claude with skills
• Claude Code: Plugins
• Anthropic public skills repository
• Anthropic: What are skills?