SkillMAS lets LLM agents evolve skills and restructure teams autonomously
New framework solves mis-specialization by coupling skill evolution with team restructuring.
SkillMAS, a non-parametric framework for LLM-based multi-agent systems, couples skill evolution with multi-agent system restructuring. It uses Utility Learning to assign credit from verified execution traces, bounded skill evolution to refine reusable procedures without unfiltered library growth, and evidence-gated MAS restructuring when retained failures and Executor Utility indicate a structural mismatch. Across embodied manipulation, command-line execution, and retail workflows, SkillMAS is competitive under the reported harnesses while clarifying how post-deployment specialization is attributed, updated, and applied.
- SkillMAS uses Utility Learning to credit individual agent contributions from verified execution traces
- Bounded skill evolution refines procedures without library bloat, enabling efficient post-deployment learning
- Evidence-gated restructuring triggers team reorganization when failures indicate structural mismatch
Why It Matters
Enables AI agent teams to self-improve after deployment, reducing manual tuning for dynamic enterprise workflows.