The Agent Skills Directory

[SAFE]: The skill provides instructional content and structured schemas (JSON and YAML) for managing 3D object detection workflows. No malicious patterns such as prompt injection, credential theft, or unauthorized network operations were detected.
[EXTERNAL_DOWNLOADS]: The skill references an official NVIDIA container image (nvcr.io/nvidia/tao/tao-toolkit:5.5.0-pyt) for execution. This is a trusted source provided by the vendor for the intended purpose of the skill.
[COMMAND_EXECUTION]: The skill defines CLI subtasks like bevfusion train, bevfusion evaluate, and bevfusion inference to be executed within a container environment. These commands are standard for the TAO Toolkit workflow.
[INDIRECT_PROMPT_INJECTION]: As a development tool, the skill processes user-supplied dataset paths and configuration overrides. While this represents a standard attack surface, the skill includes guidance on using schemas and validation to mitigate risks associated with untrusted input.

tao-train-bevfusion