What is a Skill?
A Skill is a reusable package of instructions and resources that helps an AI agent perform a task consistently.
Skill definition
A Skill is a portable folder that teaches an agent how to handle a specific kind of work. It starts with a SKILL.md file containing metadata and instructions, then optionally adds references, scripts, templates, and assets that the agent can load only when they are needed.
The important difference from a normal prompt is packaging. A Skill is discoverable, reusable, versionable, and small enough for many agents or projects to share without copying instructions into every conversation.
Common structure
Good Skills are easy to identify, easy to activate, and easy to evaluate. The root file stays focused on the core procedure, while large supporting material lives in separate folders.
my-skill/ βββ SKILL.md # instructions and metadata βββ references/ # supporting documentation βββ scripts/ # optional executable helpers βββ assets/ # templates and resources
Progressive disclosure
Skills work because agents do not need to load every instruction at startup. They first see a compact catalog, then load the full skill only when the user task matches the description.
The agent sees the skill name and description so it can decide whether the skill is relevant.
When the task matches, the agent reads the full SKILL.md instructions into context.
Scripts, references, and assets are loaded on demand instead of being injected upfront.
When Skills are useful
Where can you use Skills?
Agent Skills are not tied to one product. Many coding agents, editors, CLI tools, agent frameworks, and team workspaces now document Skill support.
Open the Client Showcase