The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions include installing the audiocraft library from the official Meta GitHub repository (facebookresearch/audiocraft). This is a legitimate source and represents standard practice for using open-source research models.\n- [COMMAND_EXECUTION]: The troubleshooting documentation provides standard system administration commands such as sudo apt-get install ffmpeg to install necessary audio backend dependencies. These commands are typical for the skill's primary purpose and do not suggest malicious intent.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text descriptions to generate audio, creating a potential surface for indirect prompt injection.\n
Ingestion points: User-supplied text prompts via the generate and generate_with_chroma methods in SKILL.md and advanced-usage.md.\n
Boundary markers: Absent; prompts are interpolated directly into model calls.\n
Capability inventory: The skill environment includes file system writing via torchaudio.save and scipy.io.wavfile.write, as well as package installation via pip.\n
Sanitization: No explicit sanitization or filtering of prompt text is implemented.

audiocraft-audio-generation