hallmark

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s hallmark study <screenshot | URL> verb accepts user-supplied URLs and, in URL mode, performs WebFetch and parses the returned HTML/CSS as runtime-readable text to extract “design facts,” creating an indirect prompt-injection surface from outsider-authored page content.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's study verb explicitly WebFetches user-supplied URLs at runtime and parses the returned HTML/CSS to extract "DNA" that the agent then uses to generate prompts/output (example source-url: https://example.com/), so external URLs supplied at runtime can directly control the agent's outputs.