skills-optimization

MASkills presents a continual learning framework that optimizes multi-agent LLM systems through agent skills, using skill-conditioned credit assignment and hierarchical aggregation to improve performance on tasks like HotpotQA and GAIA.