agents

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly defines webhook and client tools (see SKILL.md and references/client-tools.md) that call arbitrary external APIs (e.g., https://api.example.com/weather, https://api.mystore.com/orders/lookup) and ingest those responses into the agent's conversation flow, allowing untrusted third-party content to influence tool use and agent decisions.

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

• Potentially malicious external URL detected (high risk: 0.70). The embed example loads and executes remote JavaScript at runtime from https://unpkg.com/@elevenlabs/convai-widget-embed, which is a required dependency for the widget and thus a runtime-executed external script (remote code) affecting agent behavior.