The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the audiocraft library from a public repository hosted by Meta, along with other standard machine learning packages such as dora-search, transformers, and torch.\n- [COMMAND_EXECUTION]: The provided documentation includes shell commands for package management and environment setup, including instructions for system-level installation of FFmpeg using sudo in the troubleshooting section.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides examples of systems that process text-based descriptions to generate audio, creating a potential attack surface for indirect prompt injection.\n
Ingestion points: Text descriptions are ingested via prompt strings in the Python generation methods and through user-facing input fields in the Gradio and FastAPI deployment examples.\n
Boundary markers: The code snippets do not include explicit delimiters or safety instructions to separate untrusted user input from the model's generation context.\n
Capability inventory: The skill demonstrates the ability to write generated content to the file system using the torchaudio and scipy libraries.\n
Sanitization: No input validation, filtering, or sanitization of the provided prompt strings is implemented in the instructional examples.

audiocraft-audio-generation