What Is an Agent Skill? Definition, Examples, and How They Work

An agent Skill is a packaged capability an AI agent loads on demand to do a task. Learn what agent Skills are, how SKILL.md works, and why they matter.

An agent Skill is a packaged capability — a SKILL.md file plus optional scripts and reference files — that an AI agent loads on demand to perform a specific task. In harnesses such as Claude and Codex, Skills can be distributed inside agent plugins: plugins are the apps users install, while Skills are the capabilities those apps execute.

This guide explains what agent Skills are, how they work, and where they fit inside the plugin-first agent stack.

Agent Skill, defined

An agent Skill is a small, declarative bundle that teaches an agent how to do one job well. At minimum it is a single Markdown file, SKILL.md, containing:

A name and a description that tell the agent when the Skill applies.
A body of instructions the agent reads when the Skill is triggered.

Optionally, a Skill also ships scripts (for deterministic steps), reference files (loaded only when needed), and resources like templates. The agent doesn't run a Skill because you pressed a button — it reads the description, decides the Skill is relevant to your request, and loads it.

How agent Skills work

Discovery. The harness keeps every installed Skill's name and description in context. This metadata is tiny, so many Skills can coexist.
Triggering. When your request matches a Skill's description, the agent loads the full SKILL.md body.
Execution. The agent follows the instructions, optionally calling bundled scripts or loading reference files (a pattern called progressive disclosure — load detail only when needed).
Outcome. The Skill produces a result: a document, a reconciliation, a code change, a report.

Agent Skills vs. tools, plugins, and MCP

A tool is a single function call the agent can make. A Skill can orchestrate many tools.
An MCP server exposes tools and data over a protocol. A Skill can use MCP tools but is itself just instructions and assets.
A plugin is the installable app-level bundle that can contain Skills, commands, MCP configuration, and other integrations. A Skill is one executable capability of a plugin.

In short: plugins are the unit of distribution; Skills are the unit of capability; tools and MCP are the unit of access.

Why agent Skills matter inside plugins

Skills get invoked, succeed, fail, get retried, and get revised — exactly like software. As agent harnesses become the operating system for knowledge work, plugins become the apps. The teams who measure the Skills inside their plugins can see which shipped capabilities actually work.

That's the layer Telvine provides: analytics, version comparison, and management for agent plugins through capability-level telemetry.

Frequently asked questions

Is an agent Skill the same as a Claude Skill? A "Claude Skill" is an agent Skill that runs in a Claude harness (Cowork or Claude Code). The format — SKILL.md — is the same idea across harnesses.

Do I need to code to build a Skill? No. A useful Skill can be a single Markdown file. Scripts are optional, for steps you want to be deterministic.

Where do agent Skills run? Inside an agent harness such as Claude Cowork, Claude Code, or Codex CLI.