sota-survey

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

• Third-party content exposure detected (high risk: 0.90). The skill's workflow (SKILL.md and SKILL.toon) explicitly instructs the agent to fetch and read papers from public sources—e.g., arXiv, Semantic Scholar, Google Scholar, and HF hub (see "Search Order" and the "Semantic Scholar API" curl example)—and to triage and synthesize those third-party documents (3-pass triage), so untrusted web content would directly influence decisions and next actions.