SkillMAS: Skill Co-Evolution with LLM-based Multi-Agent System

TL;DR

SkillMAS is a framework that couples skill evolution with system restructuring in LLM-based multi-agent systems, enabling adaptive specialization and improved post-deployment performance.

Contribution

It introduces a non-parametric approach that integrates skill refinement and system restructuring, addressing organization bottlenecks and mis-specialization issues.

Findings

SkillMAS performs competitively across various workflows.

It clarifies how post-deployment specialization is managed.

The framework effectively couples skill evolution with system restructuring.

Abstract

Large language model (LLM) agent systems are increasingly expected to improve after deployment, but existing work often decouples two adaptation targets: skill evolution and multi-agent system (MAS) restructuring. This separation can create organization bottlenecks, context pressure, and mis-specialization. We present SkillMAS, a non-parametric framework for adaptive specialization in multi-agent systems that couples skill evolution with MAS restructuring. SkillMAS 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.