What problem does it solve?
Applying hardware-touching changes to live NVIDIA BlueField DPU and ConnectX NIC systems — firmware burns, BFB reflashes, mode flips, mlxconfig writes, kernel boot-parameter changes — can brick devices, drop management links, or leave systems unrecoverable when done without a captured baseline, out-of-band access, or a tested rollback path.
Core Features & Use Cases
- Change-application discipline: Walks operators through pre-flight inventory, out-of-band reachability, maintenance windows, apply/verify gates, and rollback ladders for every hardware-touching change class.
- Refuse-and-escalate rules: Hard-refuses changes lacking a documented rollback, out-of-band path for link-breaking changes, a maintenance window, or replica-first validation.
- Failure-mode prevention: Addresses bricked links, runaway firmware burns, silent mlxconfig mode changes (cold power cycle required), and discovered-during-failure rollbacks.
- Use Case: Before flipping a BlueField from NIC to DPU mode over SSH, the skill requires confirming an out-of-band console, capturing the PCIe topology and firmware baseline, rehearsing the change and rollback on a replica, and framing a maintenance window.
Quick Start
Ask your agent to help you safely apply an mlxconfig firmware parameter change or BlueField BFB reflash on a live system, and it will load this skill's change-application discipline before recommending any command.
