Tencent researchers present SkillEvo, addressing the problem that AI agent skills typically fail to improve from interaction failures beyond a single exchange. The method reframes multi-turn user simulation as a feedback generator where follow-up questions expose defects layer by layer, paired with an independent governance layer that actively repairs degraded skills rather than simply rejecting them. In tests across production cloud service skills, SkillEvo outperformed self-reflection-based approaches by 23.0 points and single-turn methods by 15.4 points.
