tao-finetune-clip
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill downloads pretrained model weights from well-known and trusted services, specifically HuggingFace and TorchHub, which is standard practice for machine learning workflows.
- [SAFE]: Documentation and scripts reference official NVIDIA container registries (nvcr.io) for execution environments, ensuring that software is sourced from a recognized vendor.
- [SAFE]: No instances of prompt injection, data exfiltration, or unauthorized privilege escalation were detected. The use of shell commands is limited to the official TAO Toolkit CLI tools.
- [SAFE]: The skill contains no obfuscated content, hidden URLs, or persistence mechanisms. All parameters and operations align with the stated purpose of vision-language model training and deployment.
Audit Metadata