The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of dataset directories and annotation files (.pkl), creating a potential surface for indirect prompt injection if malicious data is provided.
Ingestion points: Processes external files through parameters like dataset.root_dir and annotation files defined in the training specification templates (e.g., kitti_person_infos_train.pkl).
Boundary markers: No explicit delimiters or boundary markers are instructed for the data ingestion process to prevent the agent from following instructions embedded within the data.
Capability inventory: The skill possesses shell execution capabilities via the bevfusion and torchrun commands, allowing for model training, conversion, and evaluation.
Sanitization: The instructions do not specify sanitization or validation routines for the content of annotation files or dataset paths before they are used in command construction.
[EXTERNAL_DOWNLOADS]: The skill utilizes a vendor-managed container image from a trusted registry.
Evidence: References nvcr.io/nvidia/tao/tao-toolkit:5.5.0-pyt for executing BEVFusion tasks.
[COMMAND_EXECUTION]: The skill prescribes the execution of shell commands to manage the machine learning lifecycle.
Evidence: Uses bevfusion convert, bevfusion train, and torchrun to execute Python-based training logic within the container environment.

tao-train-bevfusion