The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a docker run command with --gpus all, --ipc=host, and --network=host. These flags provide the container with direct access to host network interfaces and shared memory, which are requirements for NVIDIA TAO high-performance GPU tasks but represent elevated execution privileges.
[INDIRECT_PROMPT_INJECTION]: The skill processes external ground truth and inference labels (in KITTI or COCO format) along with mapping files to compute metrics, which is an untrusted data ingestion surface.
Ingestion points: Annotations are loaded from paths specified in data.kpi_sources (e.g., ground_truth_ann_path and inference_ann_path) in SKILL.md.
Boundary markers: The validation script verify_kpi_analyze_spec.py ensures configuration paths exist, but no explicit markers are used to delimit the content of the labels themselves.
Capability inventory: The skill uses docker run for analysis and performs directory creation using Path.mkdir in the verification script.
Sanitization: The validation script scripts/verify_kpi_analyze_spec.py uses yaml.safe_load() for configuration and performs range checks on numerical parameters.
[EXTERNAL_DOWNLOADS]: The skill pulls a Docker container image from NVIDIA's official container registry (nvcr.io/nvidia/tao/tao-toolkit:7.2.0-data-services) to execute the analysis subtasks.

tao-analyze-detection-kpi