Infrastructure AI Agent Integration#

The Edge Node Infrastructure software ships a set of agent skills that let you run platform workflows through natural language, using GitHub Copilot or Claude Code. Instead of manually running scripts and commands, describe the outcome you want and the agent takes care of the rest.

Some skills are meant to run from the developer host, while others are meant to run directly on the provisioned host.

Available Skills#

Skill

What it does

Execution target

create-image

Builds a host OS image using the Image Composer Tool or ISO based curation

Developer host

create-usb-installation-files

Packages a complete bootable USB artifact (usb-installation-files.tar.gz), optionally running create-image first

Developer host

validate-platform-config

Validates a provisioned edge node — checks k3s pod health, binary paths, cloud-init state, network/proxy settings, and device readiness (GPU VFs, NPU)

Provisioned host

update-install-packages

Updates Ubuntu package configuration and installs required packages on a provisioned system

Developer host

set-power-profile

Sets the platform power envelope — a named profile (LowPower 10 W, BalancedLow 15 W, BalancedHigh 20 W, Performance 25 W, MaxPerformance) or a custom PkgWatt (PL1) / SysWatt (psys) envelope with burst ratio and time window

Provisioned host

set-thermal-profile

Sets the thermald thermal escalation policy — generates, validates, applies, and verifies staged Fan/Processor/powerclamp trip points (cool, warm, hot, thermal-max, or custom)

Provisioned host

monitor-power-thermal

Monitors package temperature and RAPL power domains and logs results (pt_mon.txt)

Provisioned host

generate-platform-stress

Generates configurable CPU and integrated-GPU load with stress-ng — control worker count, per-CPU load percentage, GPU worker count, and duration

Provisioned host

generate-openvino-stress

Generates sustained AI inference load on CPU, GPU, or NPU using OpenVINO benchmark_app in a K3s pod or Docker container for realistic power/thermal qualification

Provisioned host

combined-power-thermal-profiling

Orchestrates a full profiling session — chains set-power-profile → set-thermal-profile → monitor-power-thermal → generate-platform-stress and emits one consolidated enclosure report with a throttle/headroom verdict

Provisioned host

How to Use Skills#

Open GitHub Copilot Chat or Claude Code in the repository workspace and describe what you want in natural language. The agent matches your request to the appropriate skill, asks for any missing inputs, then runs the workflow. These skills have been verified from a developer system and, where applicable, against a provisioned target system.

For skills marked as Provisioned host, run the agent CLI directly on the provisioned host after installing and authenticating it. The repository code and skill scripts are available on provisioned systems at /opt/edge/developer.

Quick setup example (GitHub Copilot CLI on target host):

curl -fsSL https://gh.io/copilot-install | sudo bash
copilot --version
# Login with github credentials
copilot
# then run /login

Example Prompts#

Build an OS image for my Intel Core Ultra system.

Create a bootable USB artifact using the ICT image at /path/to/image.raw.gz.

Validate this provisioned node — check pods, drivers, proxy settings, and GPU VFs.

Add sysbench and stress-ng to ict-template only for a debug image variant.

What the Agent Does#

For every skill invocation the agent follows this fixed sequence:

  1. Collects required inputs — prompts for anything not provided in the request

  2. Runs preconditions — verifies the environment is ready before making any changes

  3. Executes steps — runs the workflow, pausing for confirmation before any privileged or destructive action

  4. Validates results — confirms the expected artifacts or system state are present

  5. Reports outcome — returns artifact paths, validation status, and troubleshooting notes on failure

Note: The agent will never proceed past a failed precondition or skip the validation step.