arXiv cs.AI· Alan Tang, Lifeng Wang, Gang Wu, Tao Zhang, Yantao Jia, Xiaofeng Zhu·· 3 天前AI 评分23
SkillMarker：基于 KGW 的智能体技能所有权验证水印方法
SkillMarker: A KGW-Based Watermarking Method for Agent Skill Ownership Verification
AI 导读
SkillMarker 是一种基于 KGW 的水印方法，用于智能体技能的所有权验证与泄露溯源，通过质量约束改写与信号导向候选搜索兼顾功能保持与水印强度。在四个数据集上，该方法对下游任务性能影响可忽略；复制粘贴测试中，基于窗口的检测可稳定识别最短 1,536 个连续 token 的抄袭片段，与一至三个背景技能合并后水印检出率仍保持 72.2–77.8%。
正文 · 原文
Abstract:Agent skills encode operational procedures and domain expertise as reusable knowledge assets, yet their distribution in plaintext exposes them to unauthorized copying and leakage. However, practical mechanisms for skill ownership verification and leakage tracing remain underdeveloped. Watermark embedding before distribution offers a potential solution, but structured, compact skill instructions impose stringent requirements for factual and procedural accuracy. To address these challenges, we introduce SkillMarker, a KGW-based watermarking method that jointly targets functionality preservation and strong watermark embedding. SkillMarker unifies quality-constrained rewriting with signal-oriented candidate search. The former separates essential knowledge from flexible expression to preserve instruction fidelity; the latter explores alternative expressions to obtain a stronger watermark signal under preservation constraints. We release an open-source toolkit that integrates watermark embedding, quality auditing, and detection to support the protection of agent skills. SkillMarker targets ownership verification and leakage tracing in distributed skill ecosystems, where content may be copied in whole or in part, or incorporated into other skills. Across four datasets, SkillMarker negligibly affects downstream task performance on average. In our copy-paste tests, window-based detection consistently identifies copied passages containing as few as 1,536 consecutive tokens. Even after merging with one or three background skills, SkillMarker retains watermark detection rates of 72.2--77.8\%.
| Comments: | Project homepage: this https URL |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15725 [cs.CR] |
| (or arXiv:2610.15725v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15725 arXiv-issued DOI via DataCite |
Submission history
From: Alan Tang [view email]
[v1]
Sun, 11 Oct 2026 16:04:26 UTC (2,176 KB)
[v2]
Wed, 14 Oct 2026 12:38:56 UTC (2,194 KB)
来源：arXiv cs.AI · arxiv.org