powercademy-skills

agent
Guvenlik Denetimi
Basarisiz
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Basarisiz
  • fs module — File system access in .github/workflows/lint.yml
  • child_process — Shell command execution capability in evals/run-evals.js
  • execSync — Synchronous shell command execution in evals/run-evals.js
  • fs module — File system access in evals/run-evals.js
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Opinionated Power Platform delivery skills for AI coding agents — methodology, conventions, and field-tested gotchas layered on Microsoft's official plugins. Works in Claude Code and GitHub Copilot CLI.

README.md

Powercademy Skills

An opinionated plugin marketplace for Power Platform delivery, for GitHub
Copilot CLI and Claude Code — every skill works in both.

Microsoft's official plugins
(power-platform-skills,
power-cat-skills) give your
agent capabilities — MCP servers, CLIs, scaffolding. This marketplace layers
judgement on top: a working delivery methodology, conventions, and
architecture thinking from real Power Platform engagements, encoded as skills.

Complementary, not competing. Nothing here duplicates what Microsoft ships —
setup-workstation installs both marketplaces side by side.

Microsoft ships the verbs; Powercademy ships the sentence. Their skills know
how to create a table or a flow; these skills carry the judgement about
whether, in what order, and what breaks if you don't — the delivery
methodology the capability-scoped Microsoft skills can't hold.

New here and using Copilot? Jump to the
GitHub Copilot setup guide.

Start here — two steps

  1. Bootstrap (one command, below) — installs the doctor plugin into
    GitHub Copilot CLI and/or Claude Code.
  2. Say "set up my workstation" — the setup-workstation skill asks what
    you'll build, then installs everything else: SDKs, CLIs, Microsoft's
    plugins, the workloads you pick. One plan, one approval, verified at the
    end. Proven on clean Windows, macOS and Linux machines.

Later, if anything breaks: "check my setup".

Prerequisites

Almost none by hand — the installer and the skills do the heavy lifting
(tool installs, marketplace registration, environment checks). What remains
for you is only what can't be automated: sign-ins, licences, permissions.
PREREQUISITES.md is the living who-handles-what list.

Install

One command, from any terminal with Node.js:

curl -fsSL https://raw.githubusercontent.com/Powercademy/powercademy-skills/main/scripts/install.js | node

Windows PowerShell:

iwr https://raw.githubusercontent.com/Powercademy/powercademy-skills/main/scripts/install.js -OutFile install.js; node install.js; del install.js

The installer is a bootstrap: it checks Node.js and git, detects Claude
Code and/or GitHub Copilot CLI, registers this marketplace and installs the
doctor plugin. Then start your agent and say:

set up my workstation

The setup-workstation skill takes it from there — asks what you'll build,
installs the SDKs, CLIs, Microsoft's plugins and the workloads you choose,
signs you in, and proves each tool works.

Manual bootstrap, inside a Claude Code or Copilot CLI session:

/plugin marketplace add Powercademy/powercademy-skills
/plugin install doctor@powercademy-skills

What's included

doctor (0.5.2) — start here

The answer to "why is setting this up such a pain?" — the first thing you run,
and the place you come back to when something breaks:

  • setup-workstation — takes a machine from nothing to fully equipped for
    end-to-end Power Platform work. Asks which workloads you'll build (flows,
    canvas, model-driven, Pages, code apps, Copilot Studio — standard or GitHub
    Copilot harness agents — PCF, mobile, process mining, ALM), inventories
    what's there, shows one plan for one approval, then installs — user-scoped first, with anything needing admin rights written
    up as a paste-ready IT request. Signs you in, verifies every tool actually
    runs, and ends with a readiness card. Re-run it to add a workload.
  • check-setup — diagnose and repair an existing toolchain: outdated or
    shimmed CLIs, auth profiles (who am I signed in as, where — the pain
    centre), the wrong tenant, plugins that won't load or drift between
    runtimes, MCP wiring. Say "check my setup" or "my auth is a mess".

Both read one versioned
toolchain manifest, and a
weekly upstream watch opens an issue
whenever Microsoft's plugins change what they need, a package in the manifest
is renamed or pulled, or a community skill lands in our lanes — so the setup
doesn't quietly rot. A fresh-machine
workflow proves it: clean Windows, macOS and Linux VMs run the bootstrap and
the whole manifest, installing into both Claude Code and Copilot CLI.

canvas-planner (0.2.0)

The decisions that determine whether a canvas app survives contact with real
data — made before anything is generated:

  • plan-canvas-app — is canvas even the right shape (or is this a
    model-driven app?), then the decision that ages worst: delegation and data
    strategy
    , verified per connector against Microsoft Learn rather than from
    memory. Plus screen and navigation architecture, and conventions set before
    generation. Produces a build plan, then hands execution to Microsoft's canvas
    authoring tooling — it generates far faster than you can by hand; it just
    can't tell you the app will silently return the wrong rows at 50,000 records.
  • Checkpoints run at production volumes, because a 50-row test proves
    nothing about delegation.

alm (0.1.0)

How the work actually ships — and the deployment that doesn't ruin your evening:

  • design-alm-process — grounds in the estate you really have (including
    the uncomfortable bits), interviews, then decides environment strategy,
    solution segmentation, publisher, the config boundary (what varies per
    environment and who owns it), promotion path and named approvers — written
    up as a process the team can follow.
  • promote-solution — runs a promotion end to end with checkpoints:
    pre-flights dependencies, exports managed, rehearses in test, verifies with
    evidence rather than exit codes, and treats production import as a 🔴 hard
    stop
    requiring explicit in-the-moment approval. Fixes go in dev, never
    downstream.

Routes execution to Microsoft's Dataverse solution skills, native pipelines and
pac — and hands Power Pages promotion off entirely, because their plugin
already covers it end to end.

conventions (0.3.0)

The judgement layer as a checklist — applied before anything gets built:

  • dataverse-conventions — living Dataverse schema and ALM conventions,
    each with its reasoning: statecode/statuscode over custom status
    columns, build-in-the-right-solution-with-the-right-publisher discipline,
    subcomponent and dependency packaging that survives import, modal dialogs
    for focused tasks. Grows every time a mistake is corrected.
  • flow-conventions — the Power Automate lessons that cost hours once:
    editing flows that already run (renames, child-flow contracts, regression
    checks), designer quirks, expressions, Graph permissions, testing without
    the real system (write gates, dry-run guards). Applied alongside whichever
    tool builds the flow — usually Microsoft's power-automate plugin.
  • pcf-conventions — whether PCF is even the right tool (often it isn't),
    which control type, and the PCF traps: the version bump or the cached
    bundle keeps rendering, dev push isn't deployment, the harness isn't the
    real host, and the premium-licensing trap. Applied alongside Microsoft's
    pcf plugin.
  • writing-skills — the house standard for authoring marketplace skills:
    dual-runtime constraints, trigger-description craft, redaction rules, and
    the review checklist. The contribution workflow, as a skill.

Roadmap

All tracked in issues with the roadmap label. Headlines:

  • audit-solution — inherited-solution auditor (grounding as a product)
  • advise-licensing — doc-verified licensing verdicts, never from memory
  • map-tenant — first-day cartography of an unfamiliar customer tenant
  • session-debrief — end-of-session retrospective that drafts gotcha
    contributions
  • Next — the orchestrator. A top-level "front door" skill that knows the
    whole Power Platform delivery surface (Dataverse, Power Automate, Power
    Pages, Code Apps, pac) and routes, sequences, and applies methodology
    across them. Skills can't wrap each other, but they can tell the agent
    which capability to reach for, in what order, with what checks — which is
    where the consolidation value actually lives. Ships alongside the first
    conventions skills (solution naming, environment strategy, ALM).

Tracked in issues.

Security

These skills instruct an agent that may act on a live tenant. The threat model
and safeguards are in SECURITY.md; a CI lint enforces the
portability and mechanics-quarantine constraints on every change.

Licence & trademarks

MIT.

Powercademy Skills is an independent, community project. It is not affiliated
with, endorsed by, or sponsored by Microsoft
. "Power Platform", "Power
Automate", "Dataverse", "Power Pages", and "Copilot" are trademarks of the
Microsoft group of companies; they are used here only to describe what the
software works with (nominative use).

Yorumlar (0)

Sonuc bulunamadi