AKS GPU Inference

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: Paid Azure usage, disclosed up front: using this skill requires an Azure account with an AKS cluster (AKS clusters and GPU node pools are billed by Microsoft under your own Azure agreement; GPU pools are among the most expensive AKS resources). It also requires kubectl and Azure CLI (az) access to your own cluster, from a host that authorizes those shell commands against the bound target, plus approved storage if evidence is collected to files. At use time the reads query your own cluster and Azure subscription (events, status, gpuProfile, DCGM exporter metrics on port 19400) and nothing is sent to any other service by the skill. Nothing is sent during installation itself. Mutations are never done on the back of a diagnosis: scaling, cordon or drain, deletion, add on enablement or monitoring changes each need your explicit authorization. Install "AKS GPU Inference" for me. Day 2 diagnosis of AKS GPU and KAITO inference incidents: GPU pods stuck Pending with Insufficient nvidia.com/gpu, model load OOM, DCGM and VRAM evidence, KAITO Workspaces not ready, and GPU autoscaling. The skill binds the subscription, cluster, kube context, pool and affected resource first, captures exact events and status and the observed gpuProfile, then routes to one of four reference files (GPU scheduling, GPU observability, KAITO workspaces, cost and scaling). It separates container or host OOM from device allocation failures, warns that deleting a KAITO Workspace leaves its GPU pools behind, and reports evidence, confidence, missing evidence and owner handoff. Mutations such as scaling, cordon or drain, deletion, add on enablement or monitoring changes require your explicit authorization. Repository: https://github.com/microsoft/azure-skills (the skill is the whole .github/plugins/aks-skills/skills/aks-gpu-inference/ folder at the pinned commit: SKILL.md, references/gpu-cost-and-scaling.md, references/gpu-observability.md, references/gpu-scheduling.md, references/kaito-workspaces.md) Pinned version: install the code at commit cf79ce8778ad063eeac19dbe8bf982d71b466ce0 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/microsoft/azure-skills into a temporary folder, check out commit cf79ce8778ad063eeac19dbe8bf982d71b466ce0, and copy the .github/plugins/aks-skills/skills/aks-gpu-inference/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-18: [FINGERPRINTS] - SKILL.md: SHA-256 0968850f574d452300a063c8b73f708a1cc54a7ba6adc103ce3c5a601f1a3f3e - references/gpu-cost-and-scaling.md: SHA-256 8c0751f64b8dd6ef3b182856860612d21ee08f15a97a94615cc4ad3abc5f7acd - references/gpu-observability.md: SHA-256 5b857b342d53070be96e79387385350c76622ca7cffb81abc4e6de480d1838fd - references/gpu-scheduling.md: SHA-256 23f0f447206cf39ba30e3cf87ce001ac61dd499fa7a651f80fca7a25c149e321 - references/kaito-workspaces.md: SHA-256 1fd74a9035f6824696e3767b18e5f350d02d93c70410f13f8d5ebbf617a13b6b If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full (and every other file in the folder) and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 5 files: SKILL.md plus four reference files in references/ (scheduling, observability, KAITO workspaces, cost and scaling). There is no executable code of its own: the reads are host advertised Azure reads or the read only queries described in the references. - Nothing is sent anywhere during installation itself; the only download is the skill folder from GitHub at the pinned commit. - At use time the evidence comes from your own AKS cluster and Azure subscription, through kubectl and az or equivalent host reads. SKILL.md forbids routing kubectl through Azure MCP and forbids bypassing host policy. - SKILL.md requires explicit authorization before scaling, cordon or drain, deletion, add on enablement or monitoring mutation, and tells the agent to leave conclusions unconfirmed when evidence is missing. 3. When I say "yes, install", install it as a skill: copy the whole folder into the agent's skills directory, in a folder named "aks-gpu-inference", keeping every file (SKILL.md, references/gpu-cost-and-scaling.md, references/gpu-observability.md, references/gpu-scheduling.md, references/kaito-workspaces.md). 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (for example: sign in or create an account, provide any API key, or install any local tool named in the prerequisites; using the skill is a separate step that only starts when I ask for it, and every paid or authenticated use disclosed above happens only then). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.