nnsight-remote-interpretability
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documents the nnsight library, which is a legitimate tool for mechanistic interpretability. All code samples provide standard use cases such as activation analysis and patching.
- [SAFE]: Remote execution functionality is directed at the official NDIF service (ndif.us), which is the intended backend for this specific tool's remote inference capabilities.
- [SAFE]: Installation instructions utilize the standard Python package manager and official PyPI registry for nnsight and torch.
- [SAFE]: The skill uses clear placeholders like 'your_key' for API configuration, ensuring no sensitive data is exposed in the instructions.
Audit Metadata