arXiv：cs.AI· Qiming Shi, Yibo Dou, Jiawen Zhu, Yulong Tao, Linbo Jin, Zhaolu Kang, Yunfan Zhou, Di Weng·· 19 小时前
SKILL-KD：面向 LLM Agent 的对比式技能蒸馏框架
SKILL-KD: Contrastive Skill Distillation for LLM Agents
AI 导读
SKILL-KD 通过对比学生 Agent 的失败轨迹与教师轨迹，将二者的可操作差异蒸馏为文本形式的技能补丁，经学生重跑评估并在仍失败时迭代优化。该方法维护与轨迹关联的编辑历史，并通过 Drift-Aware Skill Consolidation 决定每个补丁的添加、合并或跳过，以避免技能漂移。
正文
Abstract:Skill-based prompting has become a practical mechanism for improving LLM agents, yet existing methods often treat skills as summaries of the agent's own experience or of successful demonstrations. This creates a mismatch for weaker student agents. A failed trajectory may not reveal the missing knowledge or strategy, while a teacher trajectory may be too implicit to internalize. We propose SKILL-KD, a contrastive skill distillation framework that treats skills as an explicit distillation medium between agents of different capabilities. Given a student failure and the teacher trajectory on the same task, SKILL-KD distills their actionable discrepancy into a textual skill patch, evaluates it by re-running the student, and iteratively refines it when the student still fails. To prevent skill drift from repeated local updates, SKILL-KD maintains trace-linked edit histories and performs Drift-Aware Skill Consolidation to decide whether each patch is added, merged, or skipped. Across five agent benchmarks and two student settings, SKILL-KD consistently improves frozen student agents over fixed-model adaptation baselines.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.28048 [cs.AI] |
| (or arXiv:2607.28048v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.28048 arXiv-issued DOI via DataCite |
Submission history
From: Qiming Shi [view email]
[v1]
Thu, 30 Jul 2026 11:27:37 UTC (694 KB)
[v2]
Tue, 4 Aug 2026 02:41:24 UTC (694 KB)
[v3]
Mon, 5 Oct 2026 11:46:27 UTC (696 KB)
[v4]
Tue, 13 Oct 2026 09:38:38 UTC (698 KB)
来源：arXiv：cs.AI · arxiv.org