deepgram-reference-architecture

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

• Third-party content exposure detected (high risk: 0.80). The skill's implementation (e.g., architecture/sync/server.ts and architecture/async/worker.ts) accepts arbitrary audioUrl values from requests and calls deepgram.listen.prerecorded.transcribeUrl to fetch and transcribe that untrusted, user-provided content—then stores and forwards transcripts—so third-party content is ingested and can materially influence downstream behavior.