marketing-sostac

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

• Third-party content exposure detected (medium risk: 0.30). In Phase 0 Auto-Discovery, the required workflow uses agent-browser ... get text body on multiple outsider-authored sources (e.g., Google SERP results pages, G2/Trustpilot pages, Reddit search results, and Quora pages) before any item selection, ingesting the returned free text at runtime.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes an explicit install command that pulls and installs remote code from GitHub (npx skills add https://github.com/vercel-labs/agent-browser --skill agent-browser), which would fetch and execute third‑party code at runtime to provide the agent-browser capability used in auto-discovery.