Skip to content
atomic-agent docs

Skills

Skills are how you teach atomic-agent reusable workflows. A skill is a Markdown playbook with optional approved scripts attached.

How they load

Full skill bodies load only when needed. Their names and descriptions stay available so the agent knows what it can reach for, but the heavy content is pulled into the prompt budget on demand — keeping the stable prefix lean and the KV-cache reusable.

What’s in a skill

  • A Markdown body describing the procedure
  • Optional scripts that the agent may run (subject to approval gates)
  • A short description used to decide when the skill is relevant

This mirrors the agent’s broader design: keep state outside the model, load context only when it earns its place in the budget.