deepgram-data-handling
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes transcription data from external audio sources, creating a surface for indirect prompt injection where malicious instructions in the audio could influence the agent or downstream systems.\n
- Ingestion points: Transcription results are processed in
SKILL.mdandreferences/implementation.md.\n - Boundary markers: No explicit boundary markers or delimiters are used for the processed transcript text.\n
- Capability inventory: The skill utilizes file system access, S3 storage operations, database deletions, and network requests via the fetch API.\n
- Sanitization: While PII redaction logic is included, there is no logic to sanitize for command or instruction injection patterns.\n- [DATA_EXFILTRATION]: The
AuditLoggerclass inreferences/implementation.mdtransmits audit logs to a remote endpoint.\n - Evidence: The
sendToSIEMmethod uses thefetchAPI to send event data to a URL specified in theSIEM_ENDPOINTenvironment variable.
Audit Metadata