literature-mentor

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

• Third-party content exposure detected (high risk: 0.75). High likelihood: the required workflow includes runtime web fetching (web_fetch) of outsider-authored public pages and then using that fetched readable text (plus Zotero fulltext when available) as LLM context for narrative-type judgment, pre-checks, and per-figure reading—an indirect prompt-injection path from arbitrary web content.