mlx-whisper
Pass
Audited by Gen Agent Trust Hub on Feb 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Unverifiable Dependencies & Remote Code Execution (LOW): The skill installs the mlx-whisper package from PyPI and downloads model weights from Hugging Face. The project originates from the official Apple ml-explore repository, which is a trusted source.
- Command Execution (SAFE): The skill's primary function is executing the mlx_whisper command line tool to process audio files. No unsafe command injection patterns or privilege escalations were found.
- Indirect Prompt Injection (LOW): The skill ingests external audio files for transcription. 1. Ingestion points: audio.mp3 and other audio formats processed by mlx_whisper; 2. Boundary markers: Absent; 3. Capability inventory: Shell execution of transcription commands; 4. Sanitization: None detected. This is a common surface for speech-to-text skills.
Audit Metadata