"What Are Claude Skills? A Clear, Practical Guide"

Claude Skills, officially called Agent Skills by Anthropic, are reusable folders of instructions that teach Claude how to do a specific task. Each skill is a folder with a SKILL.md file inside. That file holds a short description and a step-by-step playbook Claude follows when the task comes up. Claude reads the description, decides if the skill fits what you asked, and loads the full instructions only when it needs them. That keeps the conversation fast and the context clean, while giving you repeatable, expert-level output without re-explaining your process every time.

Think of a skill as a recipe card you hand to a very capable assistant. The assistant already knows how to cook. The card tells it exactly how you like this one dish done.

How Claude Skills Work

A skill is just a folder. The only required file is SKILL.md, a plain markdown file with two parts.

The first part is YAML frontmatter at the top, wrapped in three dashes. It carries two key fields:

---
name: linkedin-hook-writer
description: Writes scroll-stopping first lines for LinkedIn posts. Use when drafting or rewriting a post hook.
---

The name identifies the skill. The description is the important one. It tells Claude what the skill does and when to use it. Claude scans descriptions across all your installed skills and triggers the right one automatically. A vague description means the skill never fires, so this line does real work.

The second part is the body of the file: the actual instructions, written in plain markdown. Steps, rules, examples, formatting guidelines, anything you would tell a teammate.

Skills can also include extra files. A folder might hold scripts the skill runs, reference documents it reads, or templates it fills in. Claude pulls these in only when the task calls for them. This is called progressive disclosure, and it is why skills stay lightweight even when they carry a lot of knowledge.

Skills work in Claude Code, Claude Desktop, Cowork, and through the Claude API, so the same skill follows you across tools.

How to Use Claude Skills

Here is the short version of getting a skill running.

1. Find or create a skill. Browse Anthropic's open skills repository on GitHub, grab a ready-made set, or write your own SKILL.md.
2. Install it. In Claude Code, you can add a skill bundle with the plugin system, for example /plugin install document-skills@anthropic-agent-skills. For a custom skill, drop the skill folder into your skills directory.
3. Add it in Claude Desktop or Cowork. Open settings, find the Skills section, and enable the skills you want. Uploaded skill folders show up there as toggles.
4. Just ask. You do not call a skill by name. Describe your task normally. Claude matches your request to a skill description and loads it on its own.
5. Refine the description. If a skill is not triggering when it should, sharpen the description line so it names the exact situations where the skill applies.

That last point matters more than people expect. The quality of your skill library depends less on clever instructions and more on clear descriptions that tell Claude precisely when to reach for each one.

Claude Skills vs Agents vs MCP

These three terms get mixed up constantly. They solve different problems and work best together.

A skill is knowledge: the how-to. An agent is the worker that runs the loop and decides what to do next. An MCP server is the hands: it gives Claude live access to outside data and actions.

They stack. A skill can tell Claude to call an MCP tool at a specific step. So a "write a LinkedIn post" skill can instruct Claude to first pull your recent top posts through an MCP connection, then draft in your proven style. The skill supplies the method, the MCP supplies the live data, and the agent runs the whole thing.

Claude Skills Examples

Skills shine on any task you repeat with a consistent process. A few common ones:

• A document skill that builds clean, formatted reports or slide decks.
• A code-review skill that checks a diff against your team's standards.
• A brand-voice skill that rewrites copy in your company tone.
• A research skill that gathers sources and structures findings the same way every time.

For a concrete, ready-to-copy set, look at Taplio's free, open library of 17 Claude Skills for LinkedIn, also available on GitHub. It covers the real jobs creators repeat: writing hooks, structuring posts, planning a content calendar, repurposing old posts, and auditing a profile. It is a useful model for what a well-scoped skill set looks like, with one clear job per skill and a description that tells Claude exactly when to use it.

Several of those skills include a "Use the Taplio MCP" section. That means they run live through the Taplio LinkedIn MCP, so instead of guessing, Claude can pull your actual post performance and write against your real data. It is a clean illustration of skills and MCP working together.

If you want curated picks, see our roundup of the best Claude skills, or the marketing-focused set in Claude marketing skills.

How to Add Skills to Claude (Quick Recap)

And if you want Claude to act on a real tool, add its MCP as a custom connector: in Claude or ChatGPT, open Settings, go to Connectors, add a custom connector, and paste the server URL (for LinkedIn, https://mcp.taplio.com). No terminal required.

To add a skill, you need a skill folder containing a SKILL.md file. In Claude Code, install a bundle with the plugin command or place a custom skill in your skills directory. In Claude Desktop and Cowork, upload or enable the skill from the Skills section in settings. Once it is installed, you do not trigger it manually. You describe your task and Claude loads the matching skill automatically.

FAQ

What are Claude Skills in simple terms?

They are reusable instruction folders that teach Claude how to do a specific task. Each skill is a folder with a SKILL.md file containing a short description and a step-by-step playbook. Claude loads a skill only when your request matches it, so you get consistent, expert output without re-explaining your process.

How do I install a Claude Skill?

Get a skill folder with a SKILL.md file. In Claude Code, install a bundle with the plugin command or drop a custom skill into your skills directory. In Claude Desktop or Cowork, upload or enable it from the Skills section in settings. After that, just describe your task and Claude triggers the right skill on its own.

What is the difference between a Claude Skill and an agent?

A skill is knowledge: the how-to instructions for a task. An agent is the autonomous loop that plans, decides, and uses tools to finish a goal. The agent does the work, and a skill can guide how it does that work.

How are Claude Skills different from MCP?

A skill is static instructions stored in a file. An MCP server is a live connection to external tools and data. Skills tell Claude how to do something, MCP lets Claude act on real systems in real time. They pair well: a skill can instruct Claude to call an MCP tool at a given step.

Where can I find good Claude Skills examples?

Anthropic publishes open skills on GitHub, and Taplio offers a free library of 17 Claude Skills for LinkedIn that doubles as a model for well-scoped skills. Several of them run live through the Taplio LinkedIn MCP.

Conclusion

Claude Skills turn your best processes into something Claude can repeat on command. A skill is a folder with a SKILL.md file: a description that tells Claude when to use it, and a playbook that tells it how. They differ from agents, which run the loop, and from MCP servers, which connect to live data. Used together, they let Claude work the way you would, with your method and your real data.

The fastest way to learn is to copy a strong set. Start with the free Claude Skills for LinkedIn and connect the Taplio LinkedIn MCP so your skills run on your actual numbers.