agentphone

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

• Third-party content exposure detected (high risk: 0.90). The skill's SKILL.md explicitly describes ingesting untrusted, user-generated call/SMS content — e.g., voice webhook payloads with "agent.message" events and the "recentHistory" transcripts, plus hosted voice mode and endpoints returning full call transcripts — which the agent/LLM is expected to read and which can drive actions (spoken responses, transfers, DTMF presses, tool calls), so third-party input can materially influence behavior.

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

• Potentially malicious external URL detected (high risk: 1.00). AgentPhone sends real-time POSTs to user-configured webhook URLs (e.g., https://your-server.com/webhook and https://your-server.com/agent-webhook) and uses the webhook's JSON/NDJSON responses to directly determine what the agent speaks, so these external endpoints are runtime dependencies that control agent prompts.