ERSkill: Evolving for Skill-Guided Adaptive Memory Retrieval
arXiv:2608.12720v1 Announce Type: new Abstract: While Large Language Model (LLM) agents increasingly rely on long-term memory for persistent interactions, the retrieval mechanisms governing this memory are rarely treated as evolvable components. This static approach limits performance on heterogeneous memory queries, which often demand diverse evidence construction strategies. To address this, we introduce \textbf{ERSkill}, a retrieval-centric framework for self-evolving, skill-guided memory acc
arXiv에서 원문 보기