hallmark

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

• Third-party content exposure detected (medium risk: 0.30). In hallmark study, when the user supplies a https://... URL (URL mode) the runtime performs WebFetch of that user-authored page and parses its returned HTML/CSS for design facts before generating the diagnosis.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's "study" verb explicitly WebFetches user-supplied URLs (any input with an http:// or https:// prefix) at runtime and parses the returned HTML/CSS to extract a "DNA" that directly influences the agent's instructions and subsequent builds, so runtime fetching of those external pages can control the agent's outputs (http:// / https://).