ecommerce-policy-readiness

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow (per references/checks.md) fetches policy text from public storefront URLs (curl on $SHIPPING, $RETURNS, $TERMS, $SUPPORT) at execution time, so an outsider who authors those pages can inject free text that the LLM reads.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs running references/checks.md which invokes curl on user-supplied storefront URLs (placeholders $PDP, $CART, $SHIPPING, $RETURNS, $TERMS, $SUPPORT) at runtime to fetch policy text that is then used to ground and control the agent's outputs, so those external URLs are runtime dependencies.