The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use sudo nvpmodel and sudo jetson_clocks. These are standard administrative commands for managing power profiles and clock frequencies on NVIDIA Jetson hardware to ensure consistent performance during model serving.
[EXTERNAL_DOWNLOADS]: The instructions reference several external container images, including vllm/vllm-openai:latest (upstream vLLM), ghcr.io/nvidia-ai-iot/vllm:latest-jetson-orin (NVIDIA-AI-IOT package), and nvcr.io/nvidia/sglang:26.01-py3 (NVIDIA NGC registry). These are well-known and vendor-owned repositories appropriate for the task.
[CREDENTIALS_UNSAFE]: The skill mentions the use of HF_TOKEN for accessing gated models on Hugging Face. Notably, the documentation includes a security warning explicitly advising against passing tokens as environment variables in shared environments and suggests safer alternatives like mounted credential files or Docker secrets.
[COMMAND_EXECUTION]: The skill suggests sourcing a local script skills/jetson-diagnostic/scripts/detect_jetson.sh to determine the hardware generation. This is a standard practice within the Jetson ecosystem to tailor configuration to specific hardware (Thor vs Orin).

jetson-llm-serve