AutoGen Multi-Agent Systems

Build conversational multi-agent workflows with Microsoft AutoGen. Generates agent role definitions, group chat orchestration, tool-using assistants, code-execution sandboxes, and human-in-the-loop checkpoints.

AutoGen turns LLMs into collaborating agents that plan, code, critique, and execute. This skill scaffolds AssistantAgent and UserProxyAgent setups, configures group chat managers, sandboxes code execution in Docker, and adds human approval gates for risky steps.

When to use

Use when one prompt is not enough — research agents that delegate, code-writing pairs with a critic, ops bots that need human approval before taking destructive actions.

Examples

Coder + Critic pair

Generate, review, and iterate on code with two agents

Set up AutoGen with a Coder assistant and a Critic that reviews each output until tests pass, sandboxed in Docker

Group chat with human approval

Multiple agents collaborate with a human gate

Build an AutoGen group chat for a research workflow: planner, retriever, writer — with human approval before any web actions