nature-response

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md + static/core/workflow.md specify that at runtime the agent parses and extracts freeform text from a user-pasted “journal email”/editor decision/reviewer comments (including manuscript and reviewer instruction text), meaning outsider-authored content is directly ingested before routing and drafting.