Ant Group and HKUST(GZ) Propose Skill-MAS, Turning Multi-Agent Orchestration into Evolvable Meta-Skills

Ant Group and the Hong Kong University of Science and Technology have introduced the Skill-MAS model for designing multi-agent systems. This model utilizes evolvable meta-skills instead of updating model parameters, enabling the accumulation of experiences in system design. Skill-MAS comprises three main components: task decomposition models, role allocation strategies, and structural patterns for defining communication channels. These components allow the system to accurately identify issues arising from incorrect task decomposition or inappropriate role allocation.