Researchers propose EvoSOP, a framework that allows Large Language Model agents to synthesize atomic actions into reusable Standard Operating Procedures (SOPs) for self-evolution. The system extracts these SOPs from execution trajectories and iteratively optimizes the toolset through construction, merging, evaluation, and pruning.
- Agents create callable higher-order tools that encapsulate multi-step logic, avoiding the need to reinvent low-level workflows.
- The framework implements a systematic lifecycle for extracting and optimizing SOPs from agent interactions.
- Experiments show significant improvements in task success rates and reduced interaction rounds compared to baselines.
This approach fosters reliable and efficient tool-use patterns, providing a scalable pathway for developing self-evolving agents.