Personalized Attacks of Social Engineering in Multi-turn Conversations -- LLM Agents for Simulation and Detection