REPO PACKAGE
microsoft GitHub-Copilot-for-Azure
OFFICIAL40skills
6ready to use
LABSCO SUMMARY
It splits roughly into two halves. One half operates on a subscription you already have: provisioning and deploying with azd, cost and quota queries, AKS setup and Automatic-migration readiness, storage and messaging troubleshooting, compliance and reliability audits, cross-cloud migration from AWS, and Entra ID / Agent Identity registration, with many of these calling the Azure MCP server (@azure/mcp) that the plugin installs via npx on first use. The other half is AI Foundry model work: capacity discovery across regions, guided or preset model deployment, fine-tuning (SFT, DPO, RFT), and full agent build/deploy/evaluate/optimize through microsoft-foundry.
This is for engineers already on Azure who want an agent that queries real resources instead of guessing at API shapes — most of it is useless without a subscription to point it at. Eight of the 44 skills aren't about Azure development at all: they're the repository's own tooling for authoring and testing skills (skill-authoring, skill-reviewer, sensei), investigating its CI test runs, and filing bugs against failing ones — material for someone contributing to this repository, not for someone using the plugin.
READ THE FULL ANALYSIS
Where azure-skills' copy comes from. Microsoft's own microsoft/skills package ships an azure-skills bundle that its README says is copied from this repository and should be fixed upstream here rather than there; that bundle carries thirty-two skills, noticeably fewer than the 44 live here today, so a chunk of what this catalogue now offers is already ahead of whatever microsoft/skills is still showing.
What it costs to start. Twenty-three of the 44 skills are labeled account_key (they need Azure credentials before they run), ten more call the Azure MCP server directly, and only eight ship as pure guidance that runs on nothing — a small minority next to the rest, so "clone it and try it" undersells what most of this actually asks for.
Three of the 47 skills we have ever tracked in this repository are no longer live; nothing in what we captured says why, so we have left the reason unstated rather than guess.
ALSO IN THIS PACKAGE
PLUGINS
4 itemsPLUGIN
azure-kusto-graph-skills
Kusto graph analysis, IRQL security hunting pipelines, and graph visualization skills for Azure Data Explorer.
PLUGIN
azure-local-skills
Azure Local skills covering standard deployments (Azure Stack HCI, 1-16 node hyperconverged, rack-aware clusters) and multi-rack (rack scale) deployments built on Network Fabric Controller, Cluster Manager, and managed network fabric.
PLUGIN
azure
Microsoft Azure MCP and Skills integration for cloud resource management, deployments, and Azure services. Manage your Azure infrastructure, monitor applications, and deploy resources directly from Claude Code.
PLUGIN
foundry-iq-skills
Create and retrieve from a Foundry IQ knowledge base using File, Azure Blob, or ADLS Gen2, then connect an existing Prompt or Hosted Agent.
MCP SERVERS
3 itemsMCP SERVER
azure
All Azure MCP tools in a single server. The Azure MCP Server implements the MCP specification to create a seamless connection between AI agents and Azure services.
stdio · npx -y @azure/mcp@latest server startMCP SERVER
azure
All Azure MCP tools in a single server. The Azure MCP Server implements the MCP specification to create a seamless connection between AI agents and Azure services.
stdio · npx -y @azure/mcp@latest server startMCP SERVER
azure
All Azure MCP tools in a single server. The Azure MCP Server implements the MCP specification to create a seamless connection between AI agents and Azure services.
stdio · npx -y @azure/mcp@latest server startWHAT'S INSIDE
40 showing · 40 totalReady to use
Nothing else to set up — install it and go.
40 of 40
SETUP
Showing all 40 skills
airunway-aks-setup
Takes a group of servers you already run on Azure and sets them up step by step, until an AI model is loaded onto them and answering requests.
analyze-skill-issues
Digs out why one of this repository's own skills keeps failing its automated tests, by fetching and reading the result files stored from the last week of runs.
analyze-test-run
Takes one automated test run of this repository and reports what passed, what failed, and how often the skill being tested was actually picked up — then opens a bug for each failure with the likely cause.
appinsights-instrumentation
What it takes to make a web app on Azure report its own errors, requests and timings to Microsoft's monitoring service, laid out per language and hosting setup.
azure-ai
Searches your own indexed content, turns speech into text or text into speech, and points you to the right reference for the rest of Microsoft's AI services.
azure-aigateway
Puts a checkpoint in front of your AI models and tools, so every request that goes through it can be capped, cached, counted and screened for harmful content.
azure-cloud-migrate
Looks at an app you run on Amazon or Google's cloud, writes up what moving it to Azure would involve, and then rewrites the code so it runs there.
azure-compliance
Scans everything you are running on Azure for security and best-practice problems, ranks what it finds by how urgent it is, and flags the stored keys, passwords and certificates that are about to expire.
azure-compute
Picking, pricing and switching on rented computers in Microsoft's cloud — a single machine or a group that grows and shrinks with demand — and reserving that capacity ahead of time.
azure-deploy
The last step of shipping: an app that has already been prepared and checked gets pushed live on Azure, past the failures that usually interrupt it, and you are handed the web address it now answers on.
azure-diagnostics
Something you run on Azure has broken or gone slow, and this works through the health data, logs and metrics in a set order until it can say what caused it and what to try next.
azure-enterprise-infra-planner
For a whole company's cloud footprint rather than a single app — it turns a description of what you need into a network, identity and security design, writes and security-scans the code that builds it, and deploys only once you have signed off.
azure-kubernetes
Before you build a cluster of machines to run containers on Azure, this settles the decisions that are painful to undo later — networking, who can reach the controls, security and monitoring.
azure-kubernetes-automatic-readiness
Azure has a hands-off version of its container service that enforces much stricter rules, and this checks whether what you run today would be turned away by them — with a suggested fix for each thing that would.
azure-kusto
Azure Data Explorer holds billions of log and telemetry records; this puts your question to it in the query language it expects and brings back the rows, the table layout or the chart you asked for.
azure-messaging
Messages have stopped moving between the parts of your system, or keep turning up twice — this reads the error the Azure queue and event-stream libraries threw and traces it back to what caused it.
azure-prepare
Everything that has to happen before an app can go to Azure: a written plan you approve, then the configuration and infrastructure files generated to match it, with nothing yet deployed.
azure-quotas
Azure caps how much of each thing you may create in a region, and hitting that cap is what stops a deployment dead; this shows the cap, how much of it is left, and files the request to raise it.
azure-reliability
Checks whether your web apps would survive a datacentre outage, names the single points of failure it finds, and applies the fixes with your say-so at each step.
azure-resource-lookup
Answers the question of what you actually have running in Azure — everything of a given type, everything missing a required tag, and the leftovers nobody is using any more.
azure-resource-visualizer
Draws the picture nobody ever got round to drawing: it walks one group of Azure resources, works out what is connected to what, and writes it up as a diagram with notes.
azure-storage
Azure will hold your data in half a dozen different ways, and the one you pick decides what it costs and how safely it is kept; this matches the choice to what you are storing and shows how to read and write it.
azure-upgrade
Moving something you already run on Azure onto a newer plan, tier or service — checked for readiness first, then carried out in confirmed steps you can stop and pick up again.
azure-validate
Between writing a deployment plan and actually running it, this tests everything that would make it fail — the config files, the permissions, a preview of what would change — while nothing has happened yet.
capacity
You want to run an AI model on Azure, but the spare capacity for it is not in every region; this ranks the places that have enough of it right now, and changes nothing itself.
customize
Putting an AI model on Azure with every dial set by hand — the exact version, the purchasing option, how much throughput, which content filter — instead of letting it choose for you.
deploy-model
The front door for putting an AI model on Azure: it reads what you asked for and sends you down the quick route, the fully configurable one, or a search for somewhere with room to spare.
entra-agent-id
Gives each AI agent its own login to your organisation's Microsoft directory, with its own permissions and its own audit trail, instead of every agent sharing one.
entra-app-registration
Before an app can let people sign in with their work Microsoft account, it has to be registered with that directory first; this walks through the registration and the settings it needs.
file-test-bug
The paperwork after an automated test breaks: it pulls the failing line and the log of what the assistant actually did, works out why the two disagree, and opens the bug report with that reasoning already written in.
finetuning
Training a general AI model on your own examples so that it answers your way — preparing the data, running the training, and checking afterwards whether it really came out better.
investigate-integration-test
Starts from the bug report for a broken test, fetches the logs it points at, and comes back with why it broke and what a fix would involve — stopping short of changing anything.
markdown-token-optimizer
Every word in an instruction file is paid for again each time an AI model reads it; this measures what one file is costing, points out the padding, and proposes the cuts without editing anything itself.
microsoft-foundry
Microsoft Foundry is where AI agents and models get built and run on Azure, and this is the single door into all of it — setting up, building, deploying, testing and fixing — passing each job to the right specialist.
preset
The quick way to get an AI model running on Azure: it tries where you already are, moves to wherever there is room if that is full, and makes the remaining choices for you.
python-appservice-deploy
Gets a Python website live on Azure with barely a question asked — it needs a name, creates whatever is missing around it, and works out for itself how the app should be started.
sensei
Rewrites a skill's own description until it matches house style, then keeps running the tests that check an assistant picks that skill for the right questions until they pass.
skill-authoring
The rules this project sets for writing one of its skills: what the description has to tell the assistant, how long each file may be, and what belongs in which folder.
skill-reviewer
Reviews a proposed change to a skill the way a maintainer would — what breaks the house rules, how serious each finding is, and whether its wording would pull in questions meant for a different skill.
vally-eval
Every skill here has to prove it actually helps, and this writes the test that does it — real questions put to a live assistant, graded on whether the skill got it to the right outcome.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (40 skills)
npx skills add microsoft/GitHub-Copilot-for-AzureSINGLE SKILLOr install a single skill
npx skills add microsoft/GitHub-Copilot-for-Azure --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.