From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents

The paper introduces EvoSOP, a framework for turning granular atomic actions into reusable Standard Operating Procedures (SOPs) that act as callable higher-order tools. Agents extract SOPs from execution trajectories and iteratively optimize their toolset through construction, merging, evaluation, and pruning. The abstract reports higher task success rates and fewer interaction rounds than baselines, while providing no specific datasets, model configurations, numerical gains, or ablation results in the supplied summary.