arxiv

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly fetches public content from arXiv (via export.arxiv.org API and arxiv.org PDFs in Steps 2–4) and then reads and summarizes abstracts/PDFs (Step 5) and ingests results into a wiki (Step 6), so untrusted third-party text is interpreted and can materially influence subsequent actions.