literature-review

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

• Third-party content exposure detected (high risk: 0.78). The skill’s runtime workflow includes fetching and ingesting full-text/metadata from outsider-authored public sources (e.g., WebSearch/WebFetch of arbitrary publisher pages and CrossRef/WebSearch results, plus “obtain full text” from external sites), which can introduce free-form text from those pages into the agent’s LLM context.