heartmula
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches project source code from GitHub and model checkpoints from the HuggingFace registry. These downloads originate from well-known developer platforms.
- [DYNAMIC_EXECUTION]: The instructions require the user to manually modify Python source code files (
src/heartlib/heartmula/modeling_heartmula.pyandsrc/heartlib/pipelines/music_generation.py) to inject code for RoPE cache initialization and compatibility fixes before running the generation script. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content via external text files for lyrics and tags, which are then processed by the underlying AI model.
- Ingestion points: Files passed to the
--lyricsand--tagsarguments inrun_music_generation.py(e.g.,SKILL.md). - Boundary markers: Absent; there are no delimiters or instructions to treat the file content as purely data.
- Capability inventory: The skill executes Python scripts (
run_music_generation.py) which utilize file I/O and subprocesses for audio generation. - Sanitization: Absent; the instructions do not provide methods for filtering or validating the contents of the lyrics or tags files.
Audit Metadata