苏格拉底式技能学习：教学与实践交替提升智能体技能迁移
原标题：Paper page - Socratic Skill Learning: Interleaving Teaching and Practice for Transferable Agent Skills
AI 摘要
该论文提出 Socratic Skill Learning（SSL）方法，通过让专家（Educator）指导与学生（Student）实践交替进行，提升 LLM 智能体技能的可迁移性。作者构建了 SkillTransferBench 基准（35 个任务族、408 个可执行用例），在主队列上 Fixed-Window SSL 达到 0.801 效用，优于无技能（0.723）、轨迹蒸馏 Trace2Skill（0.761）和对比蒸馏 SKILL-KD（0.751）。Issue-Guided SSL 进一步以更少 token 成本达到 0.805 效用。
正文节选
Socratic Skill Learning: Interleaving Teaching and Practice for Transferable Agent Skills Abstract Socratic Skill Learning improves reusable agent skills by interleaving expert guidance with student practice and targeted review, outperforming trajectory and contrastive distillation methods. Agent Skills improve LLM agents by making task knowledge reusable across sessions. Learning transferable Skills requires evidence about a procedure and its conditions of use, which a successful task attempt c