speech-to-text
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill strictly adheres to privacy-by-design principles, instructing the agent to process all audio locally and delete files immediately after transcription to prevent data persistence.
- [SAFE]: Comprehensive input validation patterns are provided, including checks for file size, MIME type (via libmagic), and audio duration to mitigate denial-of-service and file-processing attacks.
- [SAFE]: Implementation examples include automated PII (Personally Identifiable Information) filtering for transcriptions, using regular expressions to redact sensitive data such as phone numbers, emails, and credit card numbers.
- [SAFE]: The skill demonstrates secure system-level practices, such as creating temporary directories with restricted permissions (0o700) and using encryption for audio data at rest during processing.
- [SAFE]: Analysis of the input surface (audio ingestion) shows that while the skill processes untrusted data, it includes sufficient sanitization (PII filtering) and lacks dangerous capabilities like shell execution or network exfiltration of transcriptions.
Audit Metadata