A new framework using GraphRAG aims to enhance LLM-based agents' alignment with human social values by converting principles into actionable instructions. This approach shows improved performance in decision-making and emotional understanding compared to existing methods, potentially paving the way for more emotionally aware AI systems.