polish-by-diff

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

• Third-party content exposure detected (high risk: 0.70). The workflow reads outsider-authored free text from the manuscript section files (e.g., .writing/manuscript/*.tex) that the operating user did not author, and then injects that prose into the agent’s LLM context during Step 5 diff review (presenting full -/+ hunk text).