The Agent Skills Directory

[COMMAND_EXECUTION]: Provides workflows for executing nvdinov2 CLI tasks via Docker containers. These operations are restricted to the intended environment and official vendor images from nvcr.io.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to its requirement to process image datasets for self-supervised representation learning.
Ingestion points: Datasets are loaded from extracted folders specified by dataset.train_dataset.images_dir as documented in SKILL.md and references/skill_info.yaml.
Boundary markers: None identified; image content is processed directly by the model backbone.
Capability inventory: Utilizes the Bash tool to run training and inference commands within a containerized environment.
Sanitization: Content-level sanitization for image data is not provided, which is typical for the primary training task.

tao-train-nvdinov2