SkillEvo: Self-Renewing Evolution Gradients from Multi-Turn Interaction Feedback
The paper’s claim is that skill evolution stalls unless feedback keeps exposing multi-turn failures.
SkillEvo turns simulated follow-up interactions into the feedback source, using later-turn breakdowns to guide each revision. It adds a separate governance layer meant to repair factual degradation and structural bloat instead of only rejecting bad candidates with a score. The authors report gains across six cloud-service categories, 9 production Skills, and 98 reference files: 23.0 points over self-reflection evolution and 15.4 over single-turn QA-driven evolution. HF Daily Papers' note
SkillEvo turns simulated follow-up interactions into the feedback source, using later-turn breakdowns to guide each revision. It adds a separate governance layer meant to repair factual degradation and structural bloat instead of only rejecting bad candidates with a score. The authors report gains across six cloud-service categories, 9 production Skills, and 98 reference files: 23.0 points over self-reflection evolution and 15.4 over single-turn QA-driven evolution. HF Daily Papers' note
score 4