The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied datasets, which creates a potential surface for indirect prompt injection.
Ingestion points: Data enters the system via the dataset.segment.root_dir and dataset.segment.quant_calibration_dataset.images_dir paths specified in references/skill_info.yaml.
Boundary markers: The prompt instructions do not specify explicit delimiters or boundary markers for external data content.
Capability inventory: The skill uses the bash tool to execute segformer CLI commands and docker run as documented in SKILL.md and references/tao-deploy-segformer.md.
Sanitization: Input validation and data handling are managed by the underlying TAO Toolkit CLI and container environment.
[EXTERNAL_DOWNLOADS]: The skill references official NVIDIA container images to perform model lifecycle tasks.
Evidence: references/skill_info.yaml and references/tao-deploy-segformer.md reference nvcr.io/nvidia/tao/tao-toolkit images for training and deployment.
[COMMAND_EXECUTION]: The skill provides examples and instructions for running docker and segformer CLI tools to execute model training and optimization.
Evidence: Command templates are provided in references/skill_info.yaml and references/tao-deploy-segformer.md.
[SAFE]: No malicious patterns, hardcoded secrets, or unauthorized persistence mechanisms were identified.

tao-train-segformer