arxiv-search

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). This skill queries the public arXiv repository via arxiv_search.py (using arxiv.Client() / arxiv.Search) and returns paper titles and abstracts from an open, user-submitted archive, which are untrusted third‑party contents the agent will read and could contain adversarial instructions.