#15 - SkillOpt: Stable Text-Space Optimization for Self-Evolving Agent Skills

SkillOpt is a novel optimization framework designed to improve the performance of AI agents by treating their skills as a trainable, external text document. Unlike traditional methods that rely on manual prompting or model fine-tuning, this system uses a separate optimizer model to analyze successes and failures from task trajectories and propose structured edits. These edits are governed by deep-learning-inspired controls, including a textual learning rate to limit change size and a validation gate that only accepts improvements. This process ensures the agent’s strategies remain stable and reproducible without requiring expensive updates to the underlying model weights. Across diverse benchmarks, the method significantly boosts accuracy and produces compact, human-readable skill artifacts that are easily understood. Furthermore, these optimized skills demonstrate strong generalization, retaining their value even when transferred across different model scales and execution environments.