generating-patterns

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime LLM context is populated from per-section spec files that contain “Real content (verbatim from the site)” (headings/body/reviews) extracted from the source storefront/page-builder HTML, which is outsider-authored text; the generator then fills templates using those spec fields.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill explicitly instructs fetching third-party review widget data endpoints at runtime (e.g., Okendo/Junip/Yotpo/Stamped/Loox review-widget JSON APIs) to populate spec.reviews verbatim, meaning external HTTP GETs would directly control generated prompt/content.