audiocraft-audio-generation
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes standard commands for installing the library via
pipand cloning the official source code from the Facebook Research GitHub repository. - [EXTERNAL_DOWNLOADS]: Fetches pre-trained models and library components from official repositories hosted on GitHub and HuggingFace.
- [PRIVILEGE_ESCALATION]: Documentation provides instructions for installing the
ffmpegsystem utility usingsudo apt-get, which is a routine requirement for audio processing libraries. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates processing user-defined text descriptions and audio files to generate outputs, which presents a surface for potential indirect prompt injection.
- Ingestion points: User-provided text prompts passed to
model.generate()inSKILL.mdandadvanced-usage.md, as well as audio files loaded for melody conditioning. - Boundary markers: None provided in the documentation examples.
- Capability inventory: Includes file writing via
torchaudio.save, local network service hosting viaFastAPIandGradio, and audio file processing. - Sanitization: No specific input sanitization or validation methods are illustrated in the code samples.
Audit Metadata