thematic-coding

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

• Third-party content exposure detected (high risk: 0.75). The required workflow (“Thematic coding requires iterative comparison across all papers … reading, coding, comparing”) indicates the subagent consumes outsider-authored paper text/notes via the runtime paper_notes input, then uses those passages as “Key quotes/evidence” in its LLM context; since paper_notes is produced by separate paper-search/research operations and is not explicitly guaranteed to be user-authored, this is an indirect prompt-injection risk from public/third-party content.