Powercademy/powercademy-skills / plugins/workstation/skills/setup-workstation
setup-workstation
VerifiedTake a machine from nothing to fully equipped for building Power Platform solutions with AI agents - installs the CLIs, SDKs, runtimes, Microsoft and Powercademy plugin marketplaces, workload plugins and recommended MCP servers, signs in, verifies everything works, and ends with a readiness card. The first thing to run on any new machine. Trigger whenever the user says "set up my machine", "set up my workstation", "setup workstation", "new laptop", "new machine", "first-time setup", "get me started", "install everything I need", "what do I need installed", "install the tools", "onboard this machine", "prepare my PC for Power Platform", or "add canvas / Pages / Copilot Studio / mobile to my setup". Covers the whole machine across every plugin - not one plugin's own first-run. If the task is installing or provisioning the toolchain on a machine, trigger. When in doubt, trigger.
Setup Workstation — from nothing to ready
The first thing anyone runs. It installs what a Power Platform agent workstation
needs, in the right order, with consent — then proves it all works. Afterwards,
check-setup is where they go when something breaks.
The single source of truth is ${PLUGIN_ROOT}/shared/toolchain-manifest.json.
Read it before doing anything. Every item, tier, check command, install command,
admin requirement and workload mapping lives there — never improvise an install
command from memory. If the manifest's lastVerified date is more than 90 days
old, say so up front: commands may have moved.
Posture
- The user never types a command. Anything runnable, you run.
- Plan once, consent once. Inventory everything first, show one plan, get one "yes" — not install-ask-install-ask.
- Offer before installing on a customer's machine. Say whose machine it looks like if unsure, and ask.
- User-scoped first. Install everything that needs no admin rights before touching anything that does.
- Never claim a side effect you cannot observe — no "a browser has opened", no "installed" without a passing check.
- A tool is installed when its check passes, not when its installer exits 0.
Step 1: Detect the machine
Establish, and state back in one short block:
- OS (Windows / macOS / Linux) — picks the install column in the manifest.
- Agent runtime(s) present — Copilot CLI, Claude Code, or both. Plugins get installed into every runtime present. The Copilot desktop app can't be detected from a terminal — ask whether they use it.
- Admin rights — can this user elevate? (Windows: check membership of the
local Administrators group.) If not, every
admin-scoped item becomes part of the IT request in Step 5. - Managed device signals — Intune/MDM enrolment, Entra join. A managed corporate device changes the plan: expect elevation to fail and Conditional Access to block device-code sign-in.
Step 2: Ask what they'll build
Offer the manifest's workloads as a short multi-select — flows, canvas apps,
model-driven apps, Power Pages, code apps, Copilot Studio, PCF, mobile, process
mining, ALM. "Not sure yet" is a valid answer: install the required tier only and
say they can re-run this any time to add a workload.
A workload with variants gets one follow-up question (its question, offering
its options); use its default if they don't mind. Its items are the union of
the chosen options' items.
Then offer the recommended tier as an opt-in, one line each with its why. Default answer: yes to Microsoft Learn MCP (it underpins "verify, never recall"), the user's choice for the rest.
Then ask each of the manifest's integrations questions — today, "Which
project-management tool does your team use?" Nothing in an integration is
installed by default (askBeforeInstall):
- Azure DevOps → say what the Azure DevOps MCP server gives them (work items,
pull requests, pipelines, wikis and test plans from the conversation), ask
whether they want it, and if so ask for the organisation name (the
{organization}in the install command — just the name, e.g.contoso, not the URL). Respect the item'sruntimeNote: the method differs by runtime. - Anything else, or none → nothing to install; carry on.
Step 3: Inventory before installing
Resolve the plan — required items + every chosen workload's items + opted-in
recommended items, plus everything they dependsOn. If a plugin in the
manifest's retired list is already installed, add its removal to the plan
(with its replacedBy as the reason). Then run each item's check from the
manifest and classify it:
| State | Meaning |
|---|---|
| ✅ ready | Check passes and meets min |
| ⬆ outdated | Present but below min — upgrade with the item's update command (else its install command), don't uninstall first |
| 🧩 needs repair | Resolves but doesn't function — output matches failIfOutputMatches (shim) or a diagnose entry; run the repair it names, applying any setEnv in-session |
| ⬜ missing | Not present |
| 🔁 bundled | Recommended MCP already provided by an installed plugin (bundledIn) — skip, don't double-register |
Step 4: Show one plan, get one yes
Workstation plan
Item Now Action Needs admin Node.js LTS ⬆ 22.14 upgrade to 24 yes .NET 10 SDK ⬜ install yes pac CLI 🧩 shim pac install latestno Python 3.12 ⬜ install no … Already fine: git, Copilot CLI 1.0.80 Will need IT: .NET 10 SDK, Azure CLI (no admin rights on this machine)
Go ahead?
Order follows dependencies: bootstrap → required → workload → recommended, and
dependsOn within each tier.
Step 5: Install
Work the approved plan:
- User-scoped items first, each followed immediately by its check.
- Admin items — if the user can elevate, run them (they'll see an elevation prompt; say so before it appears). If they can't, or an install fails with UAC exit code 1602, do not retry and do not imply it was their mistake: collect every admin item into an IT request (see below).
- PATH refresh. A fresh install isn't on the current session's PATH. Refresh
it in-session where the platform allows (see
${PLUGIN_ROOT}/shared/microsoft-refs.md), re-run the check, and only if that still fails tell the user to open a new terminal. - Marketplaces and plugins — register each marketplace, then install each
plugin, into every runtime present. Honour the manifest's runtime fields:
an item with
runtimesapplies only to those runtimes, andmarketplaceByRuntimenames a different marketplace for a given runtime (itsruntimeNotesays why). Exact commands per runtime are in${PLUGIN_ROOT}/shared/microsoft-refs.md. New plugins and MCP servers load only after the agent restarts — say so, and say it once, at the end. - MCP servers machine-wide. Prefer the MCP servers plugins already bundle. When registering one directly, register it for the user, not the folder (the manifest's commands do) — a folder-scoped server silently vanishes in every other project folder.
The IT request — when admin items can't be installed here, write
it-request.md into the working directory: one line per item with its exact
package ID from the manifest, the reason it's needed, and the endpoints the tools
must reach (nuget, npm, PyPI, GitHub, Microsoft Learn, Microsoft sign-in). Make it
paste-ready. Then carry on with everything that doesn't depend on those items.
Step 6: Sign in
Two sign-ins matter, and both are the user's to complete:
- pac — work the auth ladder in
${PLUGIN_ROOT}/shared/microsoft-refs.md: device code first, interactive when Conditional Access blocks it. Ask which environment in plain language; prefer a dev environment. - Azure CLI — interactive by default (managed tenants often block device code), allowing for no subscriptions.
Finish with a connection card: user, tenant, environment, profile. Confirm it's the right place.
Step 7: Verify and hand over
Re-run every check in the plan. Then generate the workstation health report — run it, don't hand-write it, so the score is the same every time:
node ${PLUGIN_ROOT}/scripts/workstation-report.js --out-dir <working directory>
It is read-only. It writes workstation-report.md and a rendered
workstation-report.html: a health score out of 100 with a one-line
summary at the top, then the top three fixes, everything they have (with
versions), their workloads, the workloads they don't have yet — each with
what it unlocks, the phrase to add it and links to its docs — and the
recommended extras and integrations. Give the user both paths and the score
in one line, and suggest they open the page. If the score isn't 100, say what
the top fix is.
Then give the readiness card in the chat (same shape as check-setup
uses), and write a short setup record — what was installed, what's waiting
on IT — into the working directory as workstation-setup.md:
Workstation ready
Layer Status Core ✅ node 24 · git · .NET 10 · pac 2.13 · az 2.81 · Python 3.12 Identity ✅ user@contoso.com · Contoso Dev Runtimes ✅ Copilot CLI 1.0.80 · ✅ Claude Code Plugins ✅ dataverse · power-automate · workstation · conventions · design Recommended ✅ Learn MCP (bundled) · ✅ Playwright (bundled) Waiting on IT ⏳ Azure CLI — see it-request.mdRestart your agent now so the new plugins load. Try first: "spec a flow that emails me when a Dataverse case is escalated"
Give one "try first" prompt per chosen workload. Tell them that from now on, if anything stops working, they say "check my setup".
Re-running
Re-running is how workloads get added. Skip everything already ✅, show only the delta in the plan, and never reinstall something whose check passes.
Calibration
"Set up my new laptop" → Full flow, Steps 1–7.
"Add Copilot Studio to my setup" → Skip to Step 3 for that workload's items only.
"Is my machine set up properly?" / something's broken → That's check-setup, not this skill. Hand over.
Customer's VM → Ask before installing anything, prefer user-scoped items, and expect the IT request.
Reference files
${PLUGIN_ROOT}/shared/toolchain-manifest.json— the source of truth: items, tiers, checks, install commands, workloads.${PLUGIN_ROOT}/shared/microsoft-refs.md— marketplace/plugin commands per runtime, PATH refresh, auth ladder.${PLUGIN_ROOT}/scripts/workstation-report.js— the read-only health report (score, summary, what's ready, what to add).${PLUGIN_ROOT}/shared/artifact-style.md— house style the report's page follows.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |