Skills
skills/nvidia-nemo/megatron-bridge/nemo-mbridge-mlm-bridge-training/Gen Agent Trust Hub

nemo-mbridge-mlm-bridge-training

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions include commands to delete a local directory (rm -rf nemo_experiments) as part of a fresh training run setup. This is a standard procedure in machine learning development to ensure that training starts from scratch and does not automatically resume from stale or incompatible checkpoints.
  • [EXTERNAL_DOWNLOADS]: The skill references technical documentation hosted on docs.nvidia.com and provides instructions for updating submodules via git. These are routine operations targeting well-known services and official vendor infrastructure.
  • [SAFE]: The skill utilizes uv run for executing Python scripts, which is a best practice for managing isolated development environments and ensuring reproducible executions.
  • [SAFE]: All scripts and entry points referenced (such as run_recipe.py and pretrain_gpt.py) are localized to the repository or its official submodules, representing legitimate operational behavior for this domain.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 07:46 AM
Security Audit — agent-trust-hub — nemo-mbridge-mlm-bridge-training