← back to paper
arxiv: 2602.20156 · 2 revisions
Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks