Skip to content

portolasas/ppp

v0.2.2

ppp: builds Power Platform solutions (Power Automate, Power Apps, SharePoint, Power BI, Copilot Studio) offline from solution exports; the user carries out tenant steps guided by cards.

GitHub

ppp

ppp is an agent plugin for VS Code (GitHub Copilot) and Claude Code that builds Power Platform solutions with you: Power Automate flows, Power Apps canvas screens, SharePoint lists, Power BI reports and Copilot Studio prompts.

It works offline from solution exports. The agent has no tenant access, no connector and no live API. It reads the exported files, writes flow JSON, screen YAML and list schemas, checks them and packs an importable solution. Everything in the tenant (creating a scaffold, exporting, importing, testing) you do yourself, guided by short cards with the expected result.

What you get

  • Agent ppp with the core rules: questions as cards with a recommendation first, no silent assumptions, a commit after each confirmed step, stopping points with a model recommendation.
  • 13 skills, loaded when a task needs them:
    • phases: ppp-intake, ppp-clarify, ppp-prototype, ppp-plan, ppp-package, ppp-diagnose, ppp-docs, ppp-golive;
    • platform knowledge, each statement marked observed, documented, open or version-dependent: ppp-power-automate, ppp-power-apps, ppp-sharepoint, ppp-power-bi, ppp-copilot-studio.
  • Command line ppp (Python standard library only): unpack and pack solutions with pac, inventory flows, check flows before packing (ppp verify, checks C1–C14), check screen YAML before "paste code" (ppp canvas), convert App.Formulas to de-DE, simulate flows, create project files from templates, record usage. ppp --help lists every command.
  • Hooks for session start, turn end and compaction. They run only in ppp projects and never block.

Requirements

  • VS Code with GitHub Copilot and agent plugins enabled, or Claude Code.
  • Python 3.10 or newer.
  • Power Platform CLI pac (in PATH, from the VS Code extension Power Platform Tools, or in ~/.dotnet/tools).
  • git; a GitHub repository per project is recommended.
  • In VS Code a model that follows the agent text: GPT-6.1 Sol for building, Claude Opus 5.5 for clarification, planning and diagnosis. Not GPT-6 Luna.

Quick start

1. Install

VS Code. Add to your user settings (JSON):

"chat.plugins.enabled": true,
"chat.plugins.marketplaces": ["portolasas/ppp-plugin"]

Install ppp in the Extensions view under Agent Plugins, run "Developer: Reload Window" and choose the agent ppp once in the agent picker; it stays chosen.

Claude Code. Once per computer:

/plugin marketplace add portolasas/ppp-plugin
/plugin install ppp@ppp

Then start a new session.

2. Start a project

Open a folder for the project and tell the agent what you want, for example:

  • "New solution: an app to request training slots, with an approval flow."
  • "Take over the existing solution; the export is in exchange/."

The agent asks name, prefix, modules, mode (small, medium, large) and handbook language on one card and creates the project with ppp init. For an existing solution it unpacks the export, inventories every flow and drafts the specs.

3. Work in packages

  1. Clarify the requirements in card rounds (specs/, decisions/), optionally with a click prototype.
  2. Plan packages in ROADMAP.md.
  3. Build each package in changes/NNN-name/: you confirm the proposal, the agent builds and checks offline, packs the solution into exchange/, you import it and test the critical scenarios on cards. App screens reach Power Apps Studio by "paste code".
  4. Go live: full test, move to test and production, hypercare.

After a break the agent continues from STATE.md.

4. Updates

VS Code fetches new versions itself (with extensions.autoUpdate, about every 24 hours); reload the window afterwards. In Claude Code:

/plugin marketplace update ppp
/plugin uninstall ppp@ppp
/plugin install ppp@ppp

What changed is in CHANGELOG.md.

Project layout

project.json          name, prefix, modules, mode, solutions
STATE.md              where the work stands; read first in every session
AGENTS.md             project rules for the agent
ROADMAP.md            all packages with goal and status
specs/, decisions/    confirmed requirements and decisions
changes/NNN-name/     one folder per package (proposal, tasks); finished ones in changes/archive/
solution/<Name>/      working copy of the solution (pac unpack format)
components/<module>/  inventory, screens, list schema, flow documents, generators
exchange/             exports in, packages out (not in git)
inputs/, docs/, ...   material as handed over; handbook and training; created when needed

Data

ppp keeps everything in your project folder and its git repository. Exports can contain personal data and tenant details; keep project repositories private. The plugin itself holds no tenant, personal or customer data.

More