1 paper
Qianli Wang, Boyang Ma, Minghui Xu +1
LLM agents often rely on Skills to describe available tools and recommended procedures. We study a hidden-comment prompt injection risk in this documentation layer: when a Markdown…