The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute the re_identification command-line interface for training, evaluation, inference, and export actions, as defined in references/skill_info.yaml. These commands are specific to the NVIDIA TAO Toolkit and are restricted to the provided container environment.
[DYNAMIC_EXECUTION]: In SKILL.md, the instructions recommend setting the environment variable TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 to handle specific checkpoint loading failures. This enables potentially unsafe Pickle deserialization (loading full Python objects rather than just weights). However, the skill explicitly warns users: 'Do not use this env var for untrusted checkpoints,' which is the recommended mitigation for this known PyTorch/Lightning behavior.
[DATA_EXPOSURE]: The skill schemas and templates include a field for an encryption_key. No hardcoded keys are present; the field is consistently initialized as an empty string ('') or a placeholder (key), indicating it is intended as a user-provided configuration at runtime.

tao-train-reid