tao-train-segformer
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill defines a set of model management actions (train, evaluate, export, quantize, inference) that execute CLI commands such as
segformer trainwithin a containerized Docker environment. - [EXTERNAL_DOWNLOADS]: The skill references official vendor images hosted on the NVIDIA Container Registry (nvcr.io) for toolkit execution.
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion surface for external data by processing dataset paths and configuration overrides that parameterize the training workflow.
- Ingestion points: Dataset root directories specified in SKILL.md and configuration overrides provided during the training setup.
- Boundary markers: The skill utilizes structured YAML and JSON specifications to define parameter boundaries.
- Capability inventory: Shell command execution via Docker for model training, evaluation, and export operations.
- Sanitization: All configuration inputs are validated against action-specific JSON schemas located in the schemas/ directory to ensure they conform to expected formats.
Audit Metadata