Hardware Profile Sweep Skill
Stateless atomic operation to query system resources, operating systems, and accelerator devices across all reachable nodes.
Prerequisites
systems-manager-mcp — for collecting low-level OS, CPU, RAM, storage, and GPU parameters.
tunnel-manager-mcp — for coordinating execution across multiple hosts.
Steps
Step 1: Query System Info
Query core system resources on each host via systems_manager:
- OS details (kernel version, distribution, architecture)
- CPU specs (model, cores, threads, clock speed)
- Physical Memory (total RAM, active/free space)
- Disk Storage (disk devices, partitions, mounting locations, free capacity)
- Physical disk health via
sm_storage_health (action report): SMART per drive incl.
RAID megaraid passthrough (model/serial/power-on-hours/reallocated/predicted-fail),
PERC/LSI physical-disk state, and BMC/IPMI drive-slot faults — correlated so a
BMC-flagged disk with clean SMART media reads as a link/aging fault. Surfaces failing
or fault-asserted drives the free-capacity check alone never sees.
Step 2: Detect Accelerator Hardware
Identify and probe any connected GPU or AI accelerator devices (e.g., using nvidia-smi, lspci | grep -i vga):
- GPU models (e.g., GTX 1080, RTX 3090, A100)
- VRAM capacity and bus interface type
- Driver version and execution status
Step 3: Format HW Profile Data
Export the structured metadata representing compute hardware. Data fields strictly align with ontology_infrastructure.ttl classes:
HardwareNode properties (CPU cores, RAM size, OS info)
GPUAccelerator properties (VRAM size, bus ID, model name)
1---2name: hardware-profile-sweep-23skill_type: skill4description: >5 Hardware Profile Sweep atomic skill. Collects CPU, memory, disk, OS info,6 and GPU/accelerator details across reachable hardware nodes using systems-manager-mcp.7domain: infrastructure8tags:9 - hardware10 - systems11 - gpu12 - probe13requires:14 - systems-manager-mcp15 - tunnel-manager-mcp16metadata:17 source: https://skillmd.com/skills/knuckles-team/hardware-profile-sweep-218 version: '1.6.0'19---2021# Hardware Profile Sweep Skill2223Stateless atomic operation to query system resources, operating systems, and accelerator devices across all reachable nodes.2425## Prerequisites2627- `systems-manager-mcp` — for collecting low-level OS, CPU, RAM, storage, and GPU parameters.28- `tunnel-manager-mcp` — for coordinating execution across multiple hosts.2930## Steps3132### Step 1: Query System Info33Query core system resources on each host via `systems_manager`:34- OS details (kernel version, distribution, architecture)35- CPU specs (model, cores, threads, clock speed)36- Physical Memory (total RAM, active/free space)37- Disk Storage (disk devices, partitions, mounting locations, free capacity)38- Physical disk health via `sm_storage_health` (action `report`): SMART per drive incl.39 RAID `megaraid` passthrough (model/serial/power-on-hours/reallocated/predicted-fail),40 PERC/LSI physical-disk state, and BMC/IPMI drive-slot faults — correlated so a41 BMC-flagged disk with clean SMART media reads as a link/aging fault. Surfaces failing42 or fault-asserted drives the free-capacity check alone never sees.4344### Step 2: Detect Accelerator Hardware45Identify and probe any connected GPU or AI accelerator devices (e.g., using `nvidia-smi`, `lspci | grep -i vga`):46- GPU models (e.g., GTX 1080, RTX 3090, A100)47- VRAM capacity and bus interface type48- Driver version and execution status4950### Step 3: Format HW Profile Data51Export the structured metadata representing compute hardware. Data fields strictly align with `ontology_infrastructure.ttl` classes:52- `HardwareNode` properties (CPU cores, RAM size, OS info)53- `GPUAccelerator` properties (VRAM size, bus ID, model name)