tao-train-mask2former
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a standard implementation for the NVIDIA TAO Toolkit, providing configuration schemas and command templates for model training and deployment.
- [EXTERNAL_DOWNLOADS]: Fetches pretrained Swin Transformer weights from a public GitHub repository. This is a common and expected practice for model initialization and uses a well-known, established source.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for external data, which is necessary for its primary purpose of training and evaluating machine learning models.
- Ingestion points: Dataset directories and annotation files defined in
SKILL.mdandreferences/skill_info.yaml(e.g.,dataset.train.img_dir,dataset.train.instance_json). - Boundary markers: Absent; data paths are interpolated directly into configuration files and shell commands.
- Capability inventory: Execution of the
mask2formerCLI through Docker commands as documented inreferences/tao-deploy-mask2former.mdandreferences/skill_info.yaml. - Sanitization: Not explicitly present in the skill's documentation; validation is assumed to be handled by the underlying TAO Toolkit software.
Audit Metadata