Skip to Content
Introduction

skillet_agent

skillet_agent is an AI agent orchestration engine. You describe a system in a single YAML file — a skillet — and the runtime spins up one orchestrator agent plus one sub-agent per skill. The orchestrator routes each user request to the right skill, which runs shell scripts in its own folder to do real work. Drive it from a CLI (chat interactively, run one-shot, or embed it as a library), queue it as durable background jobs and crews on a scheduled dispatcher, or expose the whole thing to end users through the multi-user web client.

A skillet is a single YAML file plus a few markdown files. Here is the entire definition for a todo-list agent:

# todo_list.skilled_crew.yaml
version: "1.0"
id: "todo_list"
 
agents:
  todo_list_agent:
    instructionsPath: ../dotclaude_todo_list/AGENTS.md
    skills:
      - name: create-task
        folderPath: ../dotclaude_todo_list/skills/create-task
      - name: list-tasks
        folderPath: ../dotclaude_todo_list/skills/list-tasks
      - name: complete-task
        folderPath: ../dotclaude_todo_list/skills/complete-task
      - name: delete-task
        folderPath: ../dotclaude_todo_list/skills/delete-task

Each skill folder holds a SKILL.md (a short markdown file describing what the skill does) and a scripts/ directory of executables the skill can run. The orchestrator reads AGENTS.md for triage instructions. That’s the whole shape.

What this site covers

These docs are aimed at operators — people writing skillets, running the CLI, and shipping agents into production. If you’re embedding skillet_agent as a library, head to the API Reference instead.

  • Getting Started — install, set environment variables, run your first skillet.
  • Concepts — how the orchestrator and skill agents fit together; where session state, logs, cache, and cost data live.
  • Configuration — the .skilled_crew.yaml schema, SKILL.md / AGENTS.md / .command.md contracts, and MCP server integration.
  • CLI — every command and flag, plus the slash commands available inside chat.
  • Jobs & Scheduling — run skillets as durable background jobs and multi-step crews, on a recurring schedule, via the dispatcher.
  • Models — supported providers (OpenAI, LMStudio, Ollama) and how to pick a model per skillet.
  • Evaluations — write evals.json, run a suite, grade with an LLM-as-judge.
  • Web Client — the multi-user web app: authenticated chat, an inbox of results, a job board, and recurring schedules.

Why “skillet”?

Each skill folder is a self-contained unit you can drop into any skillet — heat up the ingredients you need, leave the rest. A skillet YAML is just a recipe that tells the runtime which skills to load and which agent gets to use them.

Last updated on
Install