AzureAutoLab

License: MIT

Hands-on AZ-104 exam prep that grades itself. Open a lab, complete the assigned tasks, click Verify — a deterministic bash script runs az CLI checks against your real subscription and then reports [PASS] / [FAIL] per task.

• 154 labs spanning all 5 AZ-104 domains (Identity & Governance, Storage, Compute, Networking, Monitoring)
• 100 "build" labs — given a scenario, create the resources from scratch
• 54 "troubleshoot" labs — the app pre-provisions broken Azure resources for you, you fix them, then verify and clean up with one click
• Local web dashboard (Next.js 16, App Router) that lists every lab, runs the verify script, and stores per-user results in a gitignored sidecar
• Setup wizard that gates the app until your environment is ready (Azure CLI installed, signed in, has create/update permissions on a subscription)
• Optional AI chat (bring your own OpenAI key) for asking questions while you work

git clone https://github.com/kevthefoo/AzureAutoLab
cd AzureAutoLab
npm install
npm run build
npm start

Use npm run dev instead if you're hacking on the app itself — it adds hot reload at the cost of slower per-request rendering.

Open http://localhost:3000. You'll be redirected to /setup, which checks:

1. Azure CLI is installed (az --version)
2. You're signed in (az account show) — if not, the wizard shows you the exact az login command
3. You have write permissions on the active subscription (Owner / Contributor / User Access Administrator) — if you have multiple subs, a dropdown lets you switch
4. OpenAI API key (optional — only needed if you want the chat panel)

Once steps 1–3 are green, click Continue to the app and pick a lab.

Requires: Node.js 20+, Azure CLI, and an Azure subscription where you can create and delete resources.

Recommended: A Microsoft Entra ID P1 or P2 license on your tenant. A handful of identity labs (Conditional Access, group-based licensing, PIM, B2B guest collaboration) need features that aren't on the free tier. Most labs work fine on a free tenant — Entra ID P1/P2 unlocks the rest.

1. Read the Scenario and Tasks
2. Go to the Azure portal and create the requested resources
3. Click Verify in the dashboard — the server runs the lab's bash ## Verify block against your subscription
4. The dashboard shows [PASS] Task 1: ... / [FAIL] Task N: ... per task, and saves the result locally
5. Clean up the resource group yourself when done

1. Click Start — the app provisions a deliberately misconfigured Azure environment (everything tagged AutoLabId=<NN> for traceability)
2. Read the scenario, find the misconfiguration, fix it in the portal
3. Click Verify — same deterministic bash check as build labs
4. Click Cleanup — deletes the dedicated RG-TS-<NN> resource group plus any stray tagged resources

Troubleshoot labs will cost real money while they're provisioned (typically a few cents per hour). Always click Cleanup when you finish a lab.

Each lab markdown file ends with a fenced bash ## Verify block that runs az CLI commands and emits one line per task:

[PASS] Task 1: storage account exists in eastus
[FAIL] Task 2: blob container is not set to private
[PASS] Task 3: lifecycle policy archives blobs after 90 days

The web app spawns that block with child_process, streams stdout over SSE to your browser, parses the [PASS] / [FAIL] lines, and writes the result to labs/.state/lab-NN.json (gitignored, per-user).

No LLM is involved in grading. The chat panel uses an LLM, but pass/fail is pure az.

labs/
 lab-001-resource-groups-rbac.md # build lab — scenario, tasks, ## Verify
 lab-101-troubleshoot-rdp.md # troubleshoot lab — adds ## Setup + ## Cleanup
 .state/ # per-user results (gitignored)
app/ # Next.js 16 App Router routes
components/ # React UI
lib/ # lab loading, bash spawn, setup checks
middleware.ts # gates the app behind /setup

Everything stays local:

• Your subscription ID, signed-in user, and lab results never leave your machine
• Results are written to labs/.state/ which is in .gitignore
• OpenAI keys (if you provide one) are written to .env.local (also gitignored) and only used for the chat panel
• The app talks to Azure via the CLI you're already signed into — no service principals, no stored credentials

Not built yet, in roughly the order I'd like to ship them:

• Finish the quiz mode. A 100-question bank lives in data/quiz-questions.json and /quiz is wired up, but the flow needs polish — score tracking, review wrong answers, per-domain filtering, retry-missed mode.
• Grow the question bank. Bring it to 300+ questions covering every AZ-104 sub-objective, with citations back to Microsoft Learn so you can dig into what you missed.
• Tutorial section. Short walkthroughs that introduce each Azure service before you hit the first lab on it — Azure CLI basics, RBAC mental model, VNet/subnet, storage tiers, etc. Aimed at people new to Azure, not just new to AZ-104.

MIT — see LICENSE.