The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the project source code and model checkpoints from well-known platforms.
Fetches repository: https://github.com/HeartMuLa/heartlib.git via git clone.
Downloads model checkpoints from HuggingFace: HeartMuLa/HeartMuLaGen, HeartMuLa/HeartMuLa-oss-3B-happy-new-year, and HeartMuLa/HeartCodec-oss-20260123.
[COMMAND_EXECUTION]: Installation and usage involve several shell commands.
Uses uv for environment management and package installation (uv venv, uv pip install).
Uses hf for model management.
Executes Python scripts for music generation (python ./examples/run_music_generation.py).
[DYNAMIC_EXECUTION]: The skill instructs the user to manually patch Python source files within the cloned repository to fix dependency and compatibility issues.
Patch 1: Modifies modeling_heartmula.py to re-initialize RoPE caches.
Patch 2: Modifies music_generation.py to add ignore_mismatched_sizes=True to model loading calls.
[INDIRECT_PROMPT_INJECTION]: The skill processes lyrics and tags which are untrusted external inputs, creating a potential injection surface.
Ingestion points: --lyrics and --tags arguments in run_music_generation.py.
Boundary markers: None specified in the usage instructions.
Capability inventory: The process involves model inference and file I/O within the local project directory.
Sanitization: No explicit sanitization or validation of the input text is described.

heartmula