extract-data

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

• Third-party content exposure detected (high risk: 0.70). SKILL.md specifies a subagent that receives paper_contents (full text of papers from the paper-research workflow), so if those paper full texts are fetched/ingested from outsider-authored sources at runtime, their free-form prose would be placed into the subagent’s LLM context for extraction.