deep-research

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

• Third-party content exposure detected (high risk: 0.78). The workflow’s Phase 2 “Investigation” uses a bibliography_agent to perform a systematic literature search and build a source corpus from runtime-fetched public web/academic pages, which are then ingested as readable annotated bibliography text into the LLM context (indirect prompt-injection risk from outsider-authored content).