agentphone

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

• Third-party content exposure detected (medium risk: 0.30). In “webhook” voice mode, AgentPhone POSTs caller speech transcripts (untrusted outsider-authored text) into the skill’s configured webhook endpoint as agent.message/recentHistory, which the example runtime code then builds into the LLM prompt context.

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

• Potentially malicious external URL detected (high risk: 0.90). AgentPhone will POST live call/SMS events to user-provided webhook URLs (e.g., https://your-server.com/webhook and https://your-server.com/agent-webhook), and those webhook JSON responses directly determine what the agent speaks or does during calls, so these are runtime external dependencies that control agent prompts.