What Are Claude Skills (and Agent Skills)?

A skill is a folder of instructions that teaches an AI agent how to do a specific task the way you want it done. At its simplest it is a single file, SKILL.md, holding a name, a description, and step-by-step guidance, plus any optional scripts or reference files. The agent loads the skill only when your request matches what the skill is for, so you define a method once instead of re-explaining it in every conversation.

"Claude skills" and "agent skills" are the same idea. Claude skills are the version most people met first, through Claude and Claude Code. Agent skills is the vendor-neutral name for the same open format, which now runs across several coding agents. We use the terms interchangeably here.

What a skill actually contains

A skill is a lightweight folder, not a piece of software. Inside it:

• SKILL.md: the one required file. Its frontmatter holds a name and a description, and the body holds the instructions, the actual know-how the agent should follow.
• Optional resources: scripts the agent can run, reference documents, templates, or examples, bundled alongside the SKILL.md when a task needs more than text.

The description is the most important line in the file. It is how the agent decides whether a skill is relevant to your request, so a precise description is the difference between a skill that triggers at the right moment and one that never fires.

How skills work

Skills use progressive disclosure. The agent reads only the short name and description of each available skill up front. When your request matches a skill's description, it pulls in that skill's full instructions and resources, and not before. That keeps the agent's context window focused and avoids loading a wall of instructions it does not need.

In practice this means three things:

• Context efficiency: detailed instructions load on demand, not on every message.
• Repeatability: the same task runs the same way each time, because the method is written down instead of improvised.
• Standardization: tacit knowledge, your coding standards, brand voice, or a release checklist, becomes an executable, shareable procedure.

Skills vs MCP: the difference that confuses people

This is the most common point of confusion, so it is worth being precise. Skills and MCP solve different problems and work best together.

• MCP (Model Context Protocol) gives an agent access. It connects the agent to data and actions: a database, a GitHub repo, a Slack workspace, an internal API. MCP is about reach.
• A skill gives an agent know-how. It is the operational logic: the best practices, the formatting rules, the step-by-step method for doing the job correctly. A skill is about judgment.

A simple way to hold it: MCP lets the agent touch your systems, a skill tells it what to do once it can. A reporting skill might define exactly how your quarterly numbers are formatted, while an MCP server is what actually pulls those numbers from your database. You often use both at once.

Skills vs plugins and slash commands

A plugin is a packaging and distribution wrapper. In Claude Code, a plugin can bundle skills along with commands, hooks, and MCP servers so you install them as one unit. A skill can ship inside a plugin or on its own.

Slash commands are closely related to skills. Some skills are user-invoked, meaning you trigger them deliberately by name, which behaves like a slash command. Others are model-invoked, meaning the agent reaches for them automatically when your request matches. Many skill sets include both.

The types of skills

You will run into three broad kinds:

• Built-in skills: bundled by the vendor, for example the official skills that let Claude produce Word, PowerPoint, or Excel files, or enforce a considered frontend design.
• Community skills: open-source skills shared by other developers, the bulk of what is out there.
• Custom skills: ones you or your team write to encode your own standards and workflows.

Do skills work across different agents?

Mostly, yes. SKILL.md is an open standard, and a skill written for Claude often runs in other agents with little or no change. Claude Code, Codex, and OpenCode all read the format, and the standard is increasingly treated as cross-vendor, with skills also working in tools like Cursor and GitHub Copilot.

The caveat is the part most lists skip. Anything Claude-specific can degrade elsewhere: hooks, certain MCP setups, bundled scripts, and the differences between the underlying models all change how reliably a skill behaves on another agent. That is why a skill that is excellent in Claude Code can be merely good in Codex or OpenCode. It is also why every skill in our collection carries a cross-agent compatibility note rather than a blanket "works everywhere" claim. If you want the honest version of which skills hold up where, that is the whole point of our skills marketplace.

What skills are good for

The best way to understand skills is to see real ones. A few we have used and reviewed:

• Engineering: Matt Pocock's Skills encode design discipline, grilling sessions that pin down requirements before any code, domain modeling, and TDD.
• Marketing: the Agentic SEO Skill runs evidence-backed SEO and GEO audits with prioritized fixes.
• Operations: n8n Skills teach an agent to build production-ready n8n workflows without looping on validation errors.

For more, browse our curated skills marketplace by role, or see worked skill examples.

How to start using skills

The short path: pick a skill, drop its folder where your agent looks for skills, and let the agent load it when it is relevant. The exact location and install command differ by agent, and a few skills need an MCP server configured first. We walk through it for Claude Code, Codex, and OpenCode in how to use Claude skills.

Frequently asked questions

Are skills the same as MCP? No. MCP connects an agent to data and actions, a skill gives it the method for a task. They are complementary and often used together.

Do skills work in Codex and OpenCode? Usually, because SKILL.md is an open format both read. Claude-specific features can degrade, so check the compatibility note on each skill.

Are skills the same as plugins? No. A plugin is a way to package and distribute skills (and sometimes commands, hooks, and MCP servers). A skill can ship with or without a plugin.

Do I need to be a developer to use skills? No. Using a skill is mostly dropping a folder in place and letting the agent trigger it. Writing a good skill takes more care, but using one does not.

---

Want the opinionated shortlist instead of the theory? Start with our curated skills marketplace.