The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a codebase from a GitHub repository and download multiple model checkpoints from Hugging Face. \n
Repository source: https://github.com/HeartMuLa/heartlib.git. \n
Model sources: HeartMuLa/HeartMuLaGen, HeartMuLa/HeartMuLa-oss-3B-happy-new-year, and HeartMuLa/HeartCodec-oss-20260123 from Hugging Face.
[COMMAND_EXECUTION]: The skill requires the execution of several shell commands for environment setup and music generation. \n
Setup commands: git clone, uv venv, uv pip install -e ., and uv pip install --upgrade for datasets and transformers. \n
Execution command: python ./examples/run_music_generation.py with various arguments for paths and generation parameters.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data in the form of lyrics and tags to generate music, which constitutes an indirect prompt injection surface. \n
Ingestion points: The skill reads from ./assets/lyrics.txt and ./assets/tags.txt during the generation process in SKILL.md. \n
Capability inventory: The primary capability is audio generation and saving output to a file (./assets/output.mp3) via the run_music_generation.py script. \n
Boundary markers: The instructions provide structural tags for lyrics (e.g., [Intro], [Chorus]), but no specific sanitization or delimiters are mentioned to prevent the model from interpreting content as instructions. \n
Sanitization: No explicit sanitization or filtering of the input text is described in the skill instructions.

heartmula