deepgram-webhooks-events

Pass

Audited by Gen Agent Trust Hub on Mar 14, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill's architecture for handling asynchronous transcription results introduces an indirect prompt injection surface. Malicious content within the audio being transcribed could potentially influence the agent if the resulting transcript is treated as instructions.
  • Ingestion points: The webhook endpoint POST /webhooks/deepgram in SKILL.md ingests transcript data from Deepgram.
  • Boundary markers: The example implementation lacks explicit markers or instructions to treat the transcript content as untrusted data.
  • Capability inventory: The skill is configured with Read, Write, Edit, and Bash(curl:*) tools, which increases the potential impact of a successful injection.
  • Sanitization: The provided TypeScript examples show data extraction and storage but do not include sanitization or escaping of the transcript strings.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 14, 2026, 04:32 AM