The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests MIDI files and JSON music structures which may contain untrusted metadata, such as track names and instrument labels, that could influence agent behavior if interpreted as instructions.
Ingestion Points: midi_inventory.py (MIDI), render_electronic.py (MIDI), and midi_render.py (JSON).
Capability Inventory: The skill can write files to output directories, delete temporary files via os.remove, and spawn subprocesses for audio rendering.
Boundary Markers: None present to distinguish data from instructions.
Sanitization: Relies on third-party library parsing (music21, mido, json) without explicit sanitization of text fields.
[EXTERNAL_DOWNLOADS]: The skill includes an install.sh script that modifies the host system by installing packages (fluidsynth, ffmpeg) via apt-get and Python dependencies from the official package registry. While these are well-known and standard services, the execution of a system-modifying setup script is an inherent risk factor.
[COMMAND_EXECUTION]: The documentation and the render_electronic.py script utilize subprocess.run to orchestrate the audio synthesis pipeline by calling internal scripts. This is part of the skill's primary functional purpose.

music-generation