lean-spec2cloud (copilot plugin)

A lean, opinionated copilot plugin for shipping features to Azure. Five skills, one loop:

specify → plan → implement → verify → deploy

Plus a spec2cloud orchestrator that runs the whole loop end-to-end from a single prompt.

How it works

Each skill is a focused step that reads/writes a single doc under ./docs/ and (when relevant) ./.azure/:

Skill	Trigger	Inputs	Outputs
specify	Starting or refining requirements	User prompt	./docs/spec.md, .github/copilot-instructions.md
plan	Spec exists, need design + AZD template choice	spec.md	./docs/plan.md, ./.azure/deployment-plan.md
implement	Plan exists, need code + IaC	spec.md, plan.md	./src/, ./infra/, azure.yaml
verify	Implementation ready, exercise locally	plan.md, ./src/	./docs/verify.md, provisioned Azure deps
deploy	Local verification passed	All of the above	Deployed Azure endpoint, ./docs/deploy.md
spec2cloud	Run the full loop autonomously	User prompt	All of the above

Skills auto-chain: implement will run plan if missing; verify will run implement; deploy requires verify. Each pauses on genuine ambiguity rather than guessing.

Install

Add the Spec2Cloud marketplace (or update it if already added):

copilot plugin marketplace add Azure-Samples/Spec2Cloud
# or, to update:
copilot plugin marketplace update Spec2Cloud

Then install the plugin (or update it):

copilot plugin install lean@Spec2Cloud
# or, to update:
copilot plugin update lean@Spec2Cloud

Prerequisites for verify / deploy: az, azd, and bicep CLIs installed; azd auth login completed.

Examples

One-shot: prompt → deployed endpoint

  copilot -p "/fleet lean:spec2cloud Build a simple stateless AI chat application that allows users to interact with a large language model, using React + Vite with a component library for the frontend and Python with the Microsoft Agent Framework for the backend API, with no data persistence. Deploy both the frontend and backend as Azure Container Apps, host the models on Microsoft Foundry, and enable monitoring. Authentication and private networking are out of scope." --no-ask-user --yolo

Runs all five stages, records assumptions, and returns the deployed URL.

copilot -p "/fleet lean:spec2cloud Build a todo web app with a C# backend deployed on App Service, using Cosmos DB for NoSQL as the data persistence layer." --no-ask-user --yolo

Step-by-step

/lean:specify Build a todo web app with a C# backend deployed on App Service, using Cosmos DB for NoSQL as the data persistence layer.
/lean:plan
/lean:implement
/lean:verify
/lean:deploy

Reverse-engineer a spec from existing code

/lean:specify reverse

Generates ./docs/spec.md from current ./src/ and ./infra/.

Sync the spec with recent chat decisions

/lean:specify sync

Validate implementation against spec

/lean:specify validate

Alternative: Use with APM

As an alternative to the Copilot plugin marketplace, you can distribute the Lean plugin through APM so teams get a pinned, reproducible setup.

1. Install APM:

# Windows PowerShell
irm https://aka.ms/apm-windows | iex

# macOS / Linux
curl -sSL https://aka.ms/apm-unix | sh

2. In your project, create or update apm.yml with a dependency that points to this repo and plugin path:

name: my-project
version: 1.0.0
dependencies:
  apm:
    - git: github.com/Azure-Samples/Spec2Cloud/plugins/lean-spec2cloud

3. Install dependencies into your Copilot target:

apm install --target copilot

APM will resolve and lock content in apm.lock.yaml, then deploy primitives into your local Copilot harness files.

To refresh later:

apm update

For CI or deterministic restores:

apm install --frozen

Related

• speckit-spec2cloud — full spec-kit variant (10 skills: constitution, clarify, tasks, analyze, checklist, …) for heavier workflows.