concepts

What agent skills are

A skill is a reusable instruction package. Instead of pasting the same long prompt every time, you store a structured guide that an agent can load when a task matches.

Typical shape

• A dedicated folder for the skill.
• A SKILL.md (or similarly named) markdown file as the entry point.
• YAML-style frontmatter with at least a name and description.
• Body text: when to use the skill, ordered steps, constraints, and pointers to helpers.
• Optional scripts, templates, or data files the instructions refer to.

How agents use them

Many agent systems keep a catalog of skill names and short descriptions. When your request matches a description, the agent loads the full instructions and follows them for that turn or task. Skills that do not match stay unread, which keeps context focused.

Why they are useful

• Consistency — the same procedure runs the same way across sessions.
• Reuse — encode a multi-step workflow once instead of re-explaining it.
• Scope control — load detail only when needed, instead of stuffing everything into a system prompt.
• Shareability — a folder can be copied, versioned, or reviewed like other project docs.
• Safer defaults — skills can encode checks, refusals, and “do this first” gates.

What they are not

A skill is not a model, not a plugin marketplace by itself, and not a substitute for permissions or security policy. It is documentation plus optional helpers that an agent is instructed to follow.