deepgram-incident-runbook

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md includes runtime scripts that fetch and ingest external content (e.g., the triage curl to https://status.deepgram.com/api/v2/status.json and the investigateDegradation testUrls which include arbitrary URLs like https://your-test-audio.com/sample1.wav) and uses those transcription results to drive decisions (success/fail, fallback activation), so untrusted third‑party content could influence agent behavior.