tao-train-visual-changenet

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches pretrained weights and model assets from established platforms including HuggingFace (nvidia/C-RADIOv2-B) and NVIDIA's NGC registry. It also specifies downloading datasets from S3 buckets (e.g., pdx.s8k.io, nvcf-storage-handling) for use during evaluation and training.
  • [COMMAND_EXECUTION]: The instructions involve executing shell commands for Docker container management, environment setup, and data mirroring using utility tools like s5cmd and inline python3 staging scripts.
  • [REMOTE_CODE_EXECUTION]: The segmentation workflow requires compiling custom CUDA kernels (MultiScaleDeformableAttention) from source provided within the environment using python setup.py develop. This runtime compilation of binary extensions is a standard requirement for the specific model architectures supported by the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 04:38 PM
Security Audit — agent-trust-hub — tao-train-visual-changenet