The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run to interact with the Docker daemon and Git binary. For example, scripts/verify_docker_compose_access.py manages disposable containers to verify environment health, and scripts/nva_generic_defaults.py executes Git commands to identify the repository state. These calls use list-based arguments to prevent shell injection and are necessary for the skill's setup and verification functions.
[DYNAMIC_EXECUTION]: The skill performs dynamic module loading using importlib.import_module in scripts/apply_generic_agent_template.py and scripts/run_expected_conversation.py to initialize databases and load healthcare tool handlers. Additionally, scripts/apply_generic_agent_template.py uses the ast module and string manipulation to patch Python and YAML files in the target NVA repository. This behavior is the core mechanism used to apply healthcare overlays to the base NVIDIA blueprint.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user conversation history, which is interpolated into system prompts for live validation tests in scripts/run_expected_conversation.py. While the skill includes instructions to the agent to follow strict formatting and uses semantic judging to verify behavior, the ingestion of untrusted conversational data represents a persistent attack surface for indirect prompt injection.
[EXTERNAL_DOWNLOADS]: The skill clones the Nemotron Voice Agent repository from github.com/NVIDIA-AI-Blueprints. This is an official repository owned by the skill's vendor (NVIDIA) and is the intended target for customization.

ambient-healthcare-agent-with-nemotron-voice-agent