brand

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

• Third-party content exposure detected (high risk: 1.00). The skill's Phase 1 extraction explicitly directs a real browser (Playwright) to load arbitrary brand-guidelines/marketing URLs and moodboard links and to scrape rendered copy, computed styles, images, meta tags, and logos (SKILL.md Phase 1 and Inputs), which means untrusted public web content is ingested and directly used to populate the brand profile and drive subsequent decisions — creating a clear avenue for indirect prompt injection.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly loads a user-supplied brand guidelines URL at runtime via Playwright (i.e., any https://... brand site the designer provides), extracts rendered page content and assets which are then injected into the brand-profile/board generation and thus directly control the assistant's outputs.