human-eval-review

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

• Third-party content exposure detected (medium risk: 0.65). The skill embeds the runtime items JSON (including outsider-provided item.source and candidate.excerpt fields) into the HTML’s inline <script> as const DATA = __DATA__, and then renders cand.excerpt via e.innerHTML = cand.excerpt (HTML-capable), so any outsider-authored free text included in those fields becomes LLM-context text via the reviewer’s browser session.