design-pipeline

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

• Third-party content exposure detected (high risk: 1.00). The skill's Mode 1 (references/mode-reference-gen.md) explicitly instructs the agent to open user-supplied storefront/product URLs with the OpenClaw browser (take_screenshot), analyze those screenshots to produce a design brief, and inject the extracted style elements into generation prompts—so arbitrary public webpages can be ingested and materially steer subsequent actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's Mode 1 explicitly fetches a user-provided store/product URL (the reference URL / [URL] used with OpenClaw's take_screenshot) at runtime and the captured page content is analyzed and injected into the generated design brief and nano-banana2 prompts, so external page content directly controls prompts.