The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to execute docker run with specialized configurations, including --gpus, --shm-size=8g, and --network=host. These flags are required to run the NVIDIA TAO Data Services container for object detection analysis.\n- [PRIVILEGE_ESCALATION]: Documentation recommends against mapping the current user's UID via the --user flag, as the TAO container requires a root environment for certain startup tasks (e.g., getpass.getuser()). This results in processes running with root privileges inside the container.\n- [COMMAND_EXECUTION]: A troubleshooting workflow involves running an alpine Docker container to perform a recursive chown on host-mounted directories. This is a standard procedure to restore ownership of files created by the root-privileged container.\n- [SAFE]: The container image nvcr.io/nvidia/tao/tao-toolkit:7.2.0-data-services is hosted on the official NVIDIA container registry and is a trusted resource for this skill's use case.\n- [EXTERNAL_DOWNLOADS]: The skill pulls the specialized TAO toolkit image from NVIDIA's official container registry at nvcr.io.\n- [SAFE]: The scripts/verify_object_detection_spec.py helper script follows security best practices by using yaml.safe_load() for parsing configuration data.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: Annotation directories and JSON files defined in the YAML configuration paths ground_truth_ann_path and inference_ann_path.\n
Boundary markers: Input paths are provided in a structured YAML configuration, but the skill does not use explicit delimiters or instructions to ignore potential instructions embedded within the datasets.\n
Capability inventory: Access to docker and bash tools for container execution and ownership management.\n
Sanitization: The scripts/verify_object_detection_spec.py script validates the configuration schema but does not perform sanitization on the dataset contents.

tao-analyze-gaps-od-map