interface-craft

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

• Third-party content exposure detected (medium risk: 0.30). In design-critique.md (Design Critique skill), the runtime path for critique ingests outsider-authored free text when the user supplies a URL and the skill uses WebFetch to retrieve and analyze the page content (“Live URL (Tertiary)”).

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's design-critique mode explicitly accepts a user-provided URL and instructs the agent to "Use WebFetch to retrieve the page content" at runtime, meaning external page content would be fetched and injected into the model's context to drive the critique (user-provided URL via WebFetch).