audiocraft-audio-generation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the AudioCraft library and pre-trained models from official and trusted repositories.
- Code is fetched from Meta's GitHub repository: https://github.com/facebookresearch/audiocraft.git.
- Models are downloaded from HuggingFace: facebook/musicgen-*.
- [COMMAND_EXECUTION]: Provides instructions for environment setup and dependency management.
- Includes standard pip install commands for libraries like audiocraft, torch, and transformers.
- Mentions standard system package installation via apt-get or brew for FFmpeg in the troubleshooting guide.
- [INDIRECT_PROMPT_INJECTION]: The skill implements a surface for processing natural language prompts to generate audio content.
- Ingestion points: Text prompts are accepted via MusicGen.generate(), FastAPI endpoints, and LangChain tool interfaces.
- Boundary markers: None present in the example prompts.
- Capability inventory: The skill performs file writing (torchaudio.save, scipy.io.wavfile.write) and network downloads for models.
- Sanitization: None present; however, the impact is limited to the generated audio waveform, posing no significant security risk.
Audit Metadata