design-review

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly navigates to and scrapes arbitrary target URLs (e.g., "Phase 1: Navigate to the target URL", $B goto and $B js extraction calls in Phase 2/3) and also uses WebSearch ("Search Before Building"), so untrusted public/web content is read and directly influences design findings and subsequent fix/commit actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's setup instructions run "curl -fsSL https://bun.sh/install | bash" if bun is missing, which fetches and executes remote code at runtime and is required for the browse build, so it directly executes remote code during skill runtime (https://bun.sh/install).