arxiv-search

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

• Third-party content exposure detected (high risk: 0.70). The skill performs live HTTP fetches from the public arXiv API (e.g., building URLs like "http://export.arxiv.org/api/query" in _search_arxiv_once and _best_effort_enrich using urllib.request.urlopen) and parses titles/abstracts/authors into records, so it ingests untrusted, user-authored third-party content the agent reads.