TY  - RPRT
TI  - OpenSkillRisk: Benchmarking Agent Safety When Using Real-World Risky Third-Party Skills
AU  - Qiyuan Liu
AU  - Tingfeng Hui
AU  - Kun Zhan
AU  - Kaike Zhang
AU  - Ning Miao
PY  - 2026
UR  - https://arxiv.org/abs/2607.20121
ID  - 2607.20121
ER  -