extract

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly fetches and processes arbitrary public HTTPS websites as part of its workflow (see SKILL.md "Gather evidence from public page data") and the implementation (scripts/extract-blueprint.mjs — functions like fetchWithRedirects, fetchCss, captureWithPlaywright, and sampleComputedStyles) ingests HTML/CSS and screenshots which the agent reads and uses to synthesize the DESIGN.md, so untrusted third-party content can directly influence agent outputs and actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches user-supplied HTTPS targets (e.g., https://example.com) and any linked stylesheets (e.g., https://example.com/style.css) during runtime and uses the retrieved HTML/CSS as extraction evidence that directly shapes the generated DESIGN.md (i.e., it controls the agent's output), so those runtime-fetched URLs present a clear prompt-injection/controlling-content risk.