elevenlabs-agents

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

• Third-party content exposure detected (high risk: 0.95). The required runtime workflow for this skill is a live voice conversation where the agent ingests the user’s spoken transcript (outsider-authored free text) into the LLM context via the SDK’s onEvent/conversation pipeline (e.g., event.type === 'transcript' / agent_response), enabling indirect prompt injection through user speech.

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

• Potentially malicious external URL detected (high risk: 0.90). The agent config explicitly includes a runtime custom LLM endpoint ("custom_llm": { "endpoint": "https://api.openai.com/v1/chat/completions" }) which the platform would call at runtime to generate agent responses (i.e., execute remote LLM code that directly controls prompts/outputs), so this external URL is a required runtime dependency.