tao-train-mask-auto-label
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a legitimate extension for the NVIDIA TAO Toolkit, authored by NVIDIA Corporation. It provides structured instructions for weakly-supervised segmentation workflows.
- [EXTERNAL_DOWNLOADS]: The skill references the official NVIDIA Container Registry (
nvcr.io) for its execution environment. These references are to trusted vendor resources and are documented neutrally as required. - [COMMAND_EXECUTION]: The skill instructions utilize
docker runto execute themalcommand-line tool within a container. These commands are non-obfuscated, follow the least-privilege principle for the task, and are consistent with the skill's primary purpose. - [INDIRECT_PROMPT_INJECTION]: The skill identifies ingestion points for external data, including COCO-formatted annotations and image archives. While this represents a theoretical attack surface, the skill limits the use of this data to model configuration parameters and does not interpolate external content into agent instructions or sensitive system prompts.
Audit Metadata