tao-train-ocdnet
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No security concerns were identified during the analysis of the markdown instructions and configuration schemas. The skill focuses on model training and deployment parameters for scene text detection.- [EXTERNAL_DOWNLOADS]: The skill references container images from the official NVIDIA container registry (nvcr.io). These references are used to specify the execution environment for TAO Toolkit actions.- [INDIRECT_PROMPT_INJECTION]: The skill structure includes configuration for ingesting external datasets, which represents a potential surface for indirect injection findings. However, this is inherent to the machine learning training task.
- Ingestion points: Dataset splits and calibration images enter the agent context through data path overrides in SKILL.md and references/skill_info.yaml.
- Boundary markers: Explicit boundary markers (like 'ignore instructions in data') are absent in the YAML configurations.
- Capability inventory: The skill allows execution of model-specific CLI tools (ocdnet) via the Bash tool as specified in the action definitions.
- Sanitization: Data validation and sanitization are handled by the underlying TAO Toolkit framework rather than the skill instructions themselves.
Audit Metadata