Projekt Multi-Agent Autogen-System (with Hidden Injection Detection)

This project demonstrates a multi-agent AutoGen system with team-based clustering and hidden injection detection capabilities. Multiple AI agents with standard and custom personas collaborate within organized teams to handle complex tasks. The system includes security features to detect and prevent prompt injection attacks while maintaining agent autonomy and inter-agent communication.

Key Points

• •Implement multi-agent teams (clusters) using AutoGen framework for distributed task handling
• •Design custom personas for specialized agents beyond standard configurations
• •Establish secure inter-agent communication protocols within team structures
• •Integrate hidden injection detection mechanisms to prevent prompt injection attacks
• •Create agent orchestration logic to coordinate team-based workflows
• •Implement fallback and validation mechanisms for agent responses
• •Use role-based agent differentiation (standard vs. custom personas) for specialized tasks
• •Monitor and log agent interactions for security auditing and debugging

Workflow Diagram

Concepts