nnsight-remote-interpretability
Pass
Audited by Gen Agent Trust Hub on Jun 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as educational documentation for the nnsight library and the National Deep Inference Facility (NDIF). All code samples demonstrate legitimate research use cases for model interpretability.
- [EXTERNAL_DOWNLOADS]: The skill references official project resources, including documentation at nnsight.net, the NDIF service at ndif.us, and the official GitHub repository for the ndif-team. These are recognized sources in the AI research community.
- [REMOTE_CODE_EXECUTION]: The documentation transparently explains the library's 'remote=True' feature, which allows offloading computation graphs to NDIF servers. This is an intended functional capability for analyzing large-scale models without local hardware.
- [COMMAND_EXECUTION]: Includes standard installation commands for the nnsight package via pip and instructions for setting environment variables for API authentication.
- [CREDENTIALS_UNSAFE]: Uses generic placeholders such as 'your_key' and 'your_key_here' in examples demonstrating how to configure API access via environment variables or configuration objects.
Audit Metadata