Azure Kubernetes

Azure Kubernetes Service

AUTHORITATIVE GUIDANCE — MANDATORY COMPLIANCE

This skill produces a recommended AKS cluster configuration based on user requirements, distinguishing Day-0 decisions (networking, API server — hard to change later) from Day-1 features (can enable post-creation). See CLI reference for commands.

Quick Reference

When to Use This Skill

Activate this skill when user wants to:

• Create a new AKS cluster
• Plan AKS cluster configuration for production workloads
• Design AKS networking (API server access, pod IP model, egress)
• Set up AKS identity and secrets management
• Configure AKS governance (Azure Policy, Deployment Safeguards)
• Enable AKS observability (Container Insights, Managed Prometheus, Grafana)
• Define AKS upgrade and patching strategy
• Understand AKS Automatic vs Standard SKU differences
• Get a Day-0 checklist for AKS cluster setup and configuration

Deploying an application to an existing cluster? This skill provisions and configures the cluster. To containerize an app and deploy it to a cluster that already exists (Dockerfile + manifests + Deployment Safeguards), use the azure-kubernetes-app-deploy sub-skill instead.

Rules

1. Start with the user's requirements for provisioning compute, networking, security, and other settings.
2. Use the azure MCP server and select mcp_azure_mcp_aks first to discover the exact AKS-specific MCP tools surfaced by the client. Choose the smallest discovered AKS tool that fits the task, and fall back to Azure CLI (az aks) only when the needed functionality is not exposed through the AKS MCP surface.
3. Determine if AKS Automatic or Standard SKU is more appropriate based on the user's need for control vs convenience. Default to AKS Automatic unless specific customizations are required.
4. Document decisions and rationale for cluster configuration choices, especially for Day-0 decisions that are hard to change later (networking, API server access).

Required Inputs (Ask only what’s needed)

If the user is unsure, use safe defaults.

• AKS environment type: dev/test or production
• Region(s), availability zones, preferred node VM sizes
• Expected scale (node/cluster count, workload size)
• Networking requirements (API server access, pod IP model, ingress/egress control)
• Security and identity requirements, including image registry
• Upgrade and observability preferences
• Cost constraints

Workflow

1. Cluster Type

• AKS Automatic (default): Best for most production workloads, provides a curated experience with pre-configured best practices for security, reliability, and performance. Use unless you have specific custom requirements for networking, autoscaling, or node pool configurations not supported by Node Auto-Provisioning (NAP).
• AKS Standard: Use if you need full control over environment configuration, which requires additional overhead to set up and manage.

2. Networking (Pod IP, Egress, Ingress, Dataplane)

Pod IP Model (Key Day-0 decision):

• Azure CNI Overlay (recommended): pod IPs from private overlay range, not VNet-routable, scales to large environments and good for most workloads
• Azure CNI (VNet-routable): pod IPs directly from VNet (pod subnet or node subnet), use when pods must be directly addressable from VNet or on-prem
  • Docs: https://learn.microsoft.com/azure/aks/azure-cni-overlay

Dataplane & Network Policy:

• Azure CNI powered by Cilium (recommended): eBPF-based for high-performance packet processing, network policies, and observability

Egress:

• Static Egress Gateway for stable, predictable outbound IPs
• For restricted egress: UDR + Azure Firewall or NVA

Ingress:

• App Routing addon with Gateway API — recommended default for HTTP/HTTPS workloads
• Istio service mesh with Gateway API - for advanced traffic management, mTLS, canary releases
• Application Gateway for Containers — for L7 load balancing with WAF integration

DNS:

• Enable LocalDNS on all node pools for reliable, performant DNS resolution

3. Security

• Use Microsoft Entra ID everywhere (control plane, Workload Identity for pods, node access). Avoid static credentials.
• Azure Key Vault via Secrets Store CSI Driver for secrets
• Enable Azure Policy + Deployment Safeguards
• Enable Encryption at rest for etcd/API server; in-transit for node-to-node
• Allow only signed, policy-approved images (Azure Policy + Ratify), prefer Azure Container Registry
• Isolation: Use namespaces, network policies, scoped logging

4. Observability

• Use Managed Prometheus and Container Insights with Grafana for AKS observability (logs + metrics).
• Enable Diagnostic Settings to collect control plane logs and audit logs in a Log Analytics workspace for security monitoring and troubleshooting.
• For other monitoring and troubleshooting tools, use features like the Agentic CLI for AKS, Application Insights, Resource Health Center, AppLens detectors, and Azure Advisors.

5. Upgrades & Patching

• Configure Maintenance Windows for controlled upgrade timing
• Enable auto-upgrades for control plane and node OS to stay up-to-date with security patches and Kubernetes versions
• Consider LTS versions for enterprise stability (2-year support) by upgrading your AKS environment to the Premium tier
• Fleet upgrades: Use AKS Fleet Manager for staged rollout across test to production environments

6. Performance

• Use Ephemeral OS disks (--node-osdisk-type Ephemeral) for faster node startup
• Select Azure Linux as node OS (smaller footprint, faster boot)
• Enable KEDA for event-driven autoscaling beyond HPA

7. Node Pools & Compute

• Dedicated system node pool: At least 2 nodes, tainted for system workloads only (CriticalAddonsOnly)
• Enable Node Auto Provisioning (NAP) on all pools for cost savings and responsive scaling
• Use latest generation SKUs (v5/v6) for host-level optimizations
• Avoid B-series VMs — burstable SKUs cause performance/reliability issues
• Use SKUs with at least 4 vCPUs for production workloads
• Set topology spread constraints to distribute pods across hosts/zones per SLO

8. Reliability

• Deploy across 3 Availability Zones (--zones 1 2 3)
• Use Standard tier for zone-redundant control plane + 99.95% SLA for API server availability
• Enable Microsoft Defender for Containers for runtime protection
• Configure PodDisruptionBudgets for all production workloads
• Use topology spread constraints to ensure pod distribution across failure domains

9. Cost Controls

• Use Spot node pools for batch/interruptible workloads (up to 90% savings)
• Stop/Start dev/test clusters: az aks stop/start
• Consider Reserved Instances or Savings Plans for steady-state workloads

Deep-dive scenarios — load only the relevant reference file:

Disambiguation: If a prompt matches multiple rows (e.g., "cheaper nodes" could suggest both Spot and autoscaler), prefer the most specific match. If ambiguous, ask the user to clarify their intent before loading a reference file.

Guardrails / Safety

• Do not request or output secrets (tokens, keys).
• Do not ask the user to paste subscription IDs. Discover subscription and resource scope via MCP tools (e.g., list subscriptions, list resource groups) or az account show / az account list so the agent can resolve context without exposing identifiers.
• If requirements are ambiguous for day-0 critical decisions, ask the user clarifying questions. For day-1 enabled features, propose 2–3 safe options with tradeoffs and choose a conservative default.
• Do not promise zero downtime; advise workload safeguards (PDBs, probes, replicas) and staged upgrades along with best practices for reliability and performance.

MCP Tools

Error Handling

Usage Notes

This supplement is maintained by the repository sync pipeline. It keeps the imported upstream skill usable inside this curated collection when the upstream source is intentionally concise.

Common Patterns

1. Confirm that the user's task matches the skill trigger.
2. Read the relevant project files or user-provided context before acting.
3. Choose the smallest reversible action that advances the task.
4. Run the verification command or manual check that proves the result.
5. Report the outcome, evidence, and any remaining risk.

Boundaries

• Prefer the upstream workflow for Azure Kubernetes; this section only adds local quality guardrails.
• Do not invent project facts when required files, vaults, services, or tools are unavailable.
• Stop and ask for clarification when the next action could overwrite user work, expose private data, or change production state.
• Treat skill selection as routing, not ceremony: invoke only the narrowest applicable workflow and keep user or repository instructions authoritative.