deepgram-performance-tuning

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

• Third-party content exposure detected (high risk: 0.70). The skill's SKILL.md and implementation explicitly call client.listen.prerecorded.transcribeUrl / transcribeUrl and accept arbitrary audio URLs (e.g., Step 5 cachedTranscribe, transcribeMultiple, and the benchmark functions), meaning the agent fetches and transcribes untrusted/public audio from third-party URLs and uses those transcripts in its workflow (caching, benchmarking, downstream processing), which could allow indirect prompt-injection via spoken content.