AutoGen Multi-Agent System

Create a collaborative multi-agent system using AutoGen for [Task: code generation/research/analysis].

Agent Definitions

1. UserProxyAgent

• Represents the human user
• Executes code on behalf of user
• Triggers agent conversations
• Can request human input

2. AssistantAgent

• Main coding/analysis agent
• Custom system message for role
• Generates code/solutions
• Uses GPT-4 for best results

3. CriticAgent

• Code reviewer
• Identifies bugs and issues
• Suggests improvements
• Quality assurance role

4. Executor

• Runs code safely
• Tests generated solutions
• Reports execution results
• Validates functionality

Configuration

Model Selection

• Use GPT-4 for critical agents (Assistant, Critic)
• GPT-3.5-Turbo for less critical tasks
• Optimize for quality vs cost

Safety Limits

• max_consecutive_auto_reply - Prevent infinite loops (e.g., 10)
• Timeout configurations
• Rate limiting

Termination Conditions

• Task completion keywords
• Maximum message count
• Success criteria met
• User intervention

Code Execution

• Docker container for isolation
• Safe execution environment
• Resource limits
• No access to host system

Human-in-the-Loop

• Request approval for:
  • Critical decisions
  • External API calls
  • File system operations
  • Deployment actions

Conversation Flow

Collaboration Pattern

1. User provides task
2. Assistant generates solution
3. Critic reviews and suggests improvements
4. Assistant refines based on feedback
5. Executor tests the solution
6. Iterate until success

Message Routing

• Define which agents can talk to whom
• Structured conversation flow
• Avoid chaotic interactions

Error Handling

• Try-catch blocks around agent calls
• Graceful degradation
• Error recovery strategies
• Logging for debugging

Shutdown

• Proper cleanup
• Save conversation history
• Release resources
• Final summary

Requirements

• Complete implementation
• Show full conversation flow
• Production-ready error handling
• Well-documented agent roles