Junwei Quan
Titles as the sources gave them. In 2 entries, with 0 checked quotations.
In the Base
Every entry they appear in.
SkillScreen: High-Recall Screening of Malicious Agent Skills with Decision Models
According to the description, a preprint presents SkillScreen, a compact model that scores from a skill instruction file whether an agent skill is malicious, in one pass. The authors report that it flags 20-25% of benign skills at 99% recall on unseen attack families, and that false alarms rise sharply on unfamiliar benign sources.
RH-Detect: A Unified Benchmark for Reward Hacking Detection
According to the abstract, RH-Detect combines reward-hacking subsets from eleven public datasets, which it reports as 92,761 rows in six behavior categories, into a common schema so detectors can be compared.