tao-train-mask-grounding-dino
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: A thorough analysis of the skill's instructions, configuration templates, and metadata found no evidence of malicious patterns, prompt injections, or data exfiltration attempts.
- [EXTERNAL_DOWNLOADS]: The skill references official container images from the NVIDIA Container Registry (nvcr.io) and source code from NVIDIA's GitHub repositories. These references are documented neutrally as they originate from the skill's author and represent legitimate resources.
- [COMMAND_EXECUTION]: The skill describes shell commands used to invoke the TAO Toolkit CLI (e.g., mask_grounding_dino) within Docker containers. These commands are strictly tailored to the skill's intended purpose of model training and deployment.
- [DYNAMIC_EXECUTION]: The configuration schemas mention an optional data serialization mode using the Python pickle module. This is a standard optimization in the PyTorch and TAO frameworks for sharing data between subprocesses and is used here in a safe, local context.
Audit Metadata