The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages authentication for NVIDIA services by checking for NGC_API_KEY and NVIDIA_API_KEY in shell environment variables and local .env configuration files (e.g., deploy/compose/.env). These keys are used for official registry login (nvcr.io) and model inference endpoints (integrate.api.nvidia.com).
[COMMAND_EXECUTION]: The instructions rely on numerous complex shell commands to perform environment auditing, manage Docker Compose stacks, and control Kubernetes clusters via Helm and kubectl. These commands include diagnostic scripts that inspect container health, logs, and environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external, untrusted content (such as PDFs, images, and audio files) to populate a knowledge base. This creates a surface for indirect prompt injection where malicious instructions embedded in documents could influence the agent's behavior during retrieval tasks.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill installs software from official registries, including the nvidia-rag Python package via pip and various Docker images from NVIDIA's container registry. These operations are standard for the described deployment workflow.
[DYNAMIC_EXECUTION]: The deployment process involves sourcing environment variable files (e.g., source deploy/compose/nvdev.env) to configure the RAG services, which is a common practice for Docker-based infrastructure.

rag-blueprint