AI & LLMs

Agent Skills

also: skills

Packaged, on-demand instruction bundles an agent loads only when the task calls for them.

An agent skill is a folder of instructions (plus optional scripts and resources) describing how to do one kind of task — deploy steps, a review checklist, a house style. The agent sees only each skill’s name and one-line description; the full content loads just in time when relevant. Skills are context engineering productized: expertise stored outside the window, selected on demand instead of bloating every prompt.

Worked example: packaged, reusable capabilities an agent loads on demand — a bundle of instructions (and sometimes scripts/resources) for a specific task, so the agent gains a procedure without bloating its base prompt. Gotcha: the value is progressive disclosure — only the relevant skill loads into context when the task calls for it, keeping the base prompt small; the design challenge is discoverability (the agent must recognize when a skill applies) and keeping each skill focused rather than a catch-all that always triggers.