paper-read

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly fetches papers from open web sources (e.g., "arXiv URL: Fetch the paper via WebFetch" and "Paper title + 'find it': Use WebSearch to locate the paper, then fetch it" in Phase 1), and the agent is required to read and act on that content (analysis, ingest, and downstream tooling), so untrusted third‑party content can materially influence tool use and decisions.