Lab: Agentic Power Platform Development

Build a Power Platform app with AI agents, the TALXIS SDK and TALXIS components: a monorepo, ephemeral Dev/Test environments, trunk-based development, PR quality gates and GitHub Actions deployments.

Start here

Workshop attendees: open LAB-GUIDE.md (right-click → Open Preview) and follow it with Copilot Chat. This README is the reference for running the scripts yourself.

1. Fork this repo to your personal GitHub account (top-right Fork button).

2. First set your Codespaces Default idle timeout to 240 minutes at github.com/settings/codespaces (the default, 30 minutes, stops the Codespace during any longer pause; the setting only applies to Codespaces created afterwards). Then, on your fork (https://github.com/<you>/lab), click Code → Codespaces → Create codespace on main. All tools are preinstalled. The Codespace and free GitHub Actions minutes run on your account.

   ⚠️ Don't use a "one-click" badge that points at TALXIS/lab — that starts the Codespace on the parent repo, where you can't push and your free minutes won't apply. Always launch from your own fork.

3. Wait for VS Code to load in the browser. Open a terminal (Ctrl+`) — you're in PowerShell. The tools update for about a minute after the Codespace opens; wait until the terminal goes quiet.

4. Work through the Checkpoints below, starting with CP01. CP01 is the only one you truly must run first — it's the step that signs you in. After that, every checkpoint checks its own prerequisites (both .lab-state.json and this machine's live CLI sessions) before doing anything, so re-running a checkpoint that already succeeded is safe, and you can jump ahead or skip one you don't need. If a checkpoint can't find something it needs, it tells you exactly which earlier checkpoint to (re-)run. Not sure where you left off? Run .lab-scripts/lab-status.ps1 for a summary.

💡 Each checkpoint script is fully commented — open it, read what it does, then run the whole file.

Signing in (CP01)

CP01 signs you in to everything up front: GitHub (gh), Power Platform (txc), Azure (az) and Power Apps (pa). On your own computer a browser window opens; in Codespaces you get a code instead — open the printed link, enter the code, and approve. When prompted, allow the workflow scope for gh so later checkpoints can install GitHub Actions on your fork. Use the same account for Power Platform, Azure and Power Apps.

To sign in again later (new Codespace, expired sign-in), run .lab-scripts/lab-signin.ps1, or ask Copilot Chat to sign me in and it walks you through each step.

How each checkpoint works (PR flow)

Checkpoints don't push straight to main — they teach the real ALM loop. Each one:

1. Creates a checkpoint/cpNN branch and commits its changes.
2. Opens a Pull Request and prints the link.
3. Pauses — open the PR in your browser, review the diff and the running build check.
4. Press Enter to continue: it squash-merges, deletes the branch, and moves the cpNN tag to the merged commit for rollback. You can also click Merge pull request on GitHub first; pressing Enter afterwards just finishes the save. (Run without a terminal, for example by an agent, the script says "Review and merge the PR on GitHub" and carries on once you merge.) Once CP13 makes the build check required, the merge waits for it to pass — before that, GitHub accepts the merge right away, which is the point CP13 makes.

This is the slow, deliberate part — read the PR, watch the build go green, then merge. The first time a checkpoint enables GitHub Actions you may be asked to approve workflows on your fork — say yes.

Re-running a checkpoint is safe: with nothing new to commit it only makes sure the tag is there; with new changes it opens another PR for the same checkpoint and moves the tag; and a PR left open by an interrupted run is picked up again instead of duplicated.

Checkpoints

#	Script	Goal
01	CP01-check-machine-setup.ps1	Verify all tools are installed
02	CP02-create-repository-layout.ps1	Monorepo layout (solution, src, NuGet)
03	CP03-setup-continuous-integration.ps1	Branch protection — gated PRs into main
04	CP04-setup-runtime.ps1	Create Dev + Test Dataverse environments
05	CP05-setup-continuous-deployment.ps1	OIDC service principal + deploy workflow (identifiers live in deploy.yml, no secrets)
06	CP06-implement-data-model.ps1	Warehouse tables and columns
07	CP07-implement-backend.ps1	Plugins + logic solution
08	CP08-implement-security.ps1	Security roles
09	CP09-implement-ui.ps1	Model-driven app, sitemap, forms, views, grid PCF, warehouse picking code app (TALXIS base controls, Forms, phone-friendly Home quick actions)
10	CP10-deploy-and-sync.ps1	Deploy to Dev & pull changes back
11	CP11-integrate-external-data.ps1	External data integration: Product table, custom connector, code app wiring, barcode scan UI
12	CP12-move-configuration.ps1	Configuration data migration (CMT)
13	CP13-extend-branch-policies-build-checks.ps1	Require build check on PRs
14	CP14-automate-ui-testing.ps1	BDD UI test project + (manual) test workflow
15	CP15-implement-unit-tests.ps1	Plugin (FakeXrmEasy) + script (Jest) unit tests

Run a checkpoint:

.lab-scripts/CP01-check-machine-setup.ps1

Checkpoint tags

Every checkpoint commits, pushes, and tags its result; the tag marks the latest run of that checkpoint. Use the tags to look at how the repo looked at any point (git checkout tags/cp05, then git switch main to come back). They are not a rewind button: main on your fork keeps its history, because the branch protection from CP03 refuses force pushes to it, exactly as it would on a real team. To undo a checkpoint for real, open a PR that reverts its commit, or start over on a fresh fork. Re-running a checkpoint is the normal way to fix one that failed partway.

Your variables persist in .lab-state.json (committed), so you can resume on a fresh Codespace even if your terminal crashes.

Resuming on a different machine

.lab-state.json travels with the repo, but auth sessions and CLI profiles (gh/txc/az) live on whatever machine you're on right now — they don't come along with a git clone. To pick up the lab on a fresh Codespace, VM, or after a crash:

1. Clone the repo (or open a new Codespace on your fork). The checkpoint tags come with it.
2. Run .lab-scripts/lab-status.ps1 — it reports which checkpoints are done according to .lab-state.json and git tags, and whether this machine's own gh/txc/az sessions actually match what lab-state expects (a mismatch here is the most common reason a checkpoint fails partway through instead of at the top with a clear message).
3. Run .lab-scripts/lab-signin.ps1 if lab-status.ps1 flags missing/mismatched auth, then continue with whichever checkpoint it names as next. Each checkpoint starts by bringing main up to date with your fork, so a checkpoint whose PR you merged on GitHub from another machine is picked up as done.

Developing this lab

See LOCAL-DRY-RUN.md to run the checkpoints locally, without forking to GitHub or provisioning real infrastructure.