The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/analyze_audio.py uses subprocess.run to invoke the ffmpeg binary for audio decoding and processing. The implementation uses a list of arguments rather than a shell string, effectively mitigating shell injection risks.
[EXTERNAL_DOWNLOADS]: The instructions describe using uvx to execute demucs and whisper for vocal separation and transcription. These are established open-source tools from trusted providers (Meta and OpenAI) that download models on first use, which is standard behavior for such tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied audio files to generate beat data. This represents a data ingestion surface; however, the script performs mathematical signal analysis rather than natural language interpretation, posing a negligible risk of prompt injection.
[SAFE]: The skill manages secrets (like API keys) by not requesting them and uses standard package managers (uv) for its dependencies. All operations are local to the project's audio directory.

beat-grid