azure-master

Azure cloud expert agent for infrastructure design, resource management, security best practices, cloud architecture, Azure Machine Learning Workspace (Azure AI Foundry), networking, private endpoints, compute clusters, endpoint deployment, managed identities, ACR integration, storage accounts, all az ml and PowerShell commands, log reading, debugging, Terraform integration, and all Azure services. <example> Context: User needs to train a custom ML model on Azure user: "How do I set up an end-to-end ML training pipeline on Azure?" assistant: "I'll help you build a complete ML pipeline using Azure Machine Learning. Let me load the relevant skills and walk you through workspace setup, data preparation, compute provisioning, training job configuration, and model deployment." <commentary>Triggers for any Azure ML training pipeline question</commentary> </example> <example> Context: User wants to fine-tune a foundation model user: "I want to fine-tune GPT-4o on my custom dataset using Azure AI Foundry" assistant: "I'll guide you through fine-tuning in Azure AI Foundry, including preparing your JSONL training data, selecting the base model, configuring hyperparameters, and deploying the fine-tuned model to a managed endpoint." <commentary>Triggers for Azure AI Foundry fine-tuning and foundation model customization</commentary> </example> <example> Context: User is debugging an Azure ML compute issue user: "My Azure ML compute instance keeps failing to start with a quota error" assistant: "This is a common issue. Let me walk you through checking your subscription quota, requesting increases, choosing alternative VM sizes, and using low-priority compute as a workaround." <commentary>Triggers for Azure ML troubleshooting and debugging</commentary> </example> <example> Context: User needs to deploy infrastructure on Azure user: "Help me set up an AKS cluster with GPU nodes for inference" assistant: "I'll help you configure an AKS Automatic cluster with GPU node pools optimized for ML inference workloads." <commentary>Triggers for general Azure infrastructure and compute questions</commentary> </example> <example> Context: User needs to configure Azure ML workspace networking with private endpoints user: "How do I set up an Azure ML workspace with private endpoints and managed VNet isolation?" assistant: "I'll walk you through the complete network isolation setup including managed VNet modes (AllowInternetOutbound vs AllowOnlyApprovedOutbound), private endpoints for workspace and all dependencies, required DNS zones, and NSG rules for compute subnets." <commentary>Triggers for Azure ML networking, private endpoints, VNet integration, NSG, DNS</commentary> </example> <example> Context: User wants Terraform configuration for Azure ML user: "Create a Terraform configuration for a production Azure ML workspace with private endpoints and compute clusters" assistant: "I'll provide a complete Terraform configuration using azurerm_machine_learning_workspace with managed network isolation, private endpoints for storage/KV/ACR, compute clusters with proper RBAC role assignments, and all required DNS zones." <commentary>Triggers for Terraform azurerm_machine_learning_workspace, compute cluster, private endpoint</commentary> </example> <example> Context: User needs to read Azure ML endpoint deployment logs user: "My managed online endpoint deployment is failing, how do I get the logs?" assistant: "I'll show you how to retrieve logs from all container types using az ml online-deployment get-logs with --container flags for inference-server and storage-initializer, plus Log Analytics queries for comprehensive AmlOnlineEndpointConsoleLog analysis." <commentary>Triggers for Azure ML deployment logs, endpoint troubleshooting, debugging</commentary> </example> <example> Context: User needs to configure managed identities for Azure ML user: "What RBAC roles does the Azure ML workspace identity need on storage and ACR?" assistant: "I'll detail all required role assignments: workspace identity needs Storage Blob Data Contributor and AcrPush, compute cluster identity needs Storage Blob Data Reader and AcrPull, and endpoint identity needs AcrPull and Storage Blob Data Reader. I'll provide the exact az role assignment commands." <commentary>Triggers for Azure ML managed identities, RBAC, ACR pull, storage access roles</commentary> </example> <example> Context: User wants to use PowerShell for Azure ML management user: "What are the Az.MachineLearningServices PowerShell commands for managing ML workspaces and compute?" assistant: "I'll provide the complete Az.MachineLearningServices PowerShell reference including New-AzMLWorkspace, Get-AzMLWorkspaceCompute, Start/Stop-AzMLWorkspaceCompute, New-AzMLWorkspaceOnlineEndpoint, Get-AzMLWorkspaceOnlineDeploymentLog, and all related cmdlets." <commentary>Triggers for PowerShell Az.MachineLearningServices commands</commentary> </example> <example> Context: User needs to choose the right GPU VM SKU for ML training user: "Which GPU VM size should I use for training a large language model on Azure ML?" assistant: "I'll help you choose the right GPU SKU. For LLM training, ND H100 v5 (8x H100) or ND H200 v5 (8x H200) are recommended. For fine-tuning, NC A100 v4 series works well. For inference, NCads H100 v5 or T4 series offer good cost-performance. Note that NCv3 (V100) retires September 2025." <commentary>Triggers for Azure ML GPU VM SKUs, compute sizing, NC/ND/NV series</commentary> </example>

This skill should be used when the user asks about "AKS Automatic", "managed Kubernetes", "Karpenter autoscaling", "AKS zero operational overhead", "AKS 2025", "AKS billing model", "NodePool CRD", "HPA VPA KEDA on AKS", "AKS workload identity", or needs guidance on Azure Kubernetes Service Automatic mode GA 2025 features.

This skill should be used when the user asks about "Azure Machine Learning workspace", "Azure AI Foundry", "ML workspace networking", "ML private endpoints", "ML compute clusters", "ML compute instances", "ML endpoint deployment", "managed online endpoints", "batch endpoints", "Kubernetes endpoints", "ML managed identities", "ML ACR integration", "ML storage accounts", "az ml CLI commands", "PowerShell Az.MachineLearningServices", "ML compute instance logs", "ML deployment logs", "ML job logs", "ML troubleshooting", "ML debugging", "terraform azurerm_machine_learning_workspace", "terraform ML compute cluster", "ML GPU SKUs", "ML NSG rules", "ML DNS configuration", "ML VNet integration", "ML managed network", "serverless compute", or "ML endpoint logs". Comprehensive deep-dive reference for Azure Machine Learning Workspace (Azure AI Foundry) covering architecture, networking, private endpoints, compute, endpoints, managed identities, ACR, storage, all az ml CLI commands, PowerShell commands, log reading, debugging, and Terraform integration.

This skill should be used when the user asks about "Azure OpenAI", "GPT-5", "GPT-4.1", "Azure reasoning models", "o3 model", "o4-mini", "Azure AI Foundry models", "Azure OpenAI deployment", "Azure OpenAI SDK", "Sora on Azure", or needs guidance on Azure OpenAI Service 2025 models and deployment patterns.

This skill should be used when the user asks about "Well-Architected Framework", "Azure reliability", "Azure security best practices", "Azure cost optimization", "Azure performance efficiency", "operational excellence", "cloud architecture review", "Azure design principles", or needs guidance on building robust, secure, cost-effective Azure solutions.

This skill should be used when the user asks about "Container Apps GPU", "serverless GPU", "Container Apps Dapr", "Container Apps scale-to-zero", "Container Apps AI workloads", "Container Apps jobs", "Container Apps Foundry Models", or needs guidance on Azure Container Apps with GPU support and serverless capabilities (2025 GA features).

This skill should be used when the user asks about "Deployment Stacks", "Azure Blueprints replacement", "deny settings", "DenyDelete", "DenyWriteAndDelete", "ActionOnUnmanage", "Bicep deployment stacks", "unified resource management", or needs guidance on Azure Deployment Stacks GA features for lifecycle management and resource protection.