Overview

Configures AutoGen agents (AssistantAgent, UserProxyAgent) and orchestrates multi-agent conversations for scalable conversational AI. It enables group chats, nested conversations, and teachable patterns, with options for code execution and human-in-the-loop oversight to support autonomous-task planning.

How This Skill Works

This skill exposes configuration options for LLM models, temperatures, code execution settings, human input mode, max consecutive auto-replies, and function calling. It defines agent types (AssistantAgent, UserProxyAgent, GroupChatManager, ConversableAgent) and patterns (two-agent conversations, group chats with selection, nested conversations, teachable agents), powered by pyautogen.

When to Use It

• Building a multi-agent AI system for autonomous task planning (multi-agent-system).
• Coordinating conversations between an assistant and a human proxy via UserProxyAgent.
• Setting up group chats or nested conversations among multiple agents.
• Enabling safe code execution within agents for dynamic tasks.
• Implementing human-in-the-loop patterns for oversight and approvals.

Quick Start

1. Step 1: Install pyautogen and load the autogen-setup skill into your bot framework.
2. Step 2: Configure agents and options — choose LLM models, temperatures, code execution, human input mode, max auto-replies, and function calling.
3. Step 3: Deploy a test scenario (e.g., two-agent or group-chat task) and monitor logs, then refine patterns and settings.

Best Practices

• Define clear termination conditions for conversations and tasks.
• Sandbox code execution safely and restrict external effects.
• Write explicit, consistent agent system messages for each role.
• Monitor conversation flow and intervene when needed.
• Test two-agent and group-chat patterns before deployment.

Example Use Cases

• Two-agent planning: AssistantAgent proposes a plan and the UserProxyAgent validates and executes code as needed.
• Group chat orchestration: GroupChatManager coordinates several agents to complete a complex task.
• Nested conversations: Teachable agents learn from user feedback through layered prompt interactions.
• Teachable agents: Agents adapt behavior based on demonstrations and corrections.
• Human-in-the-loop: A supervisor agent reviews critical decisions in real-time before execution.

Frequently Asked Questions