← All architecture labs
AI-Ready Repo Lab #001Ready to run
AI EngineeringCopilot workflow

Build an AI-Ready Repository with GitHub Copilot

Can repository-native instructions, specialized agents, skills, tool boundaries, verification, and behavioral evals make agentic development more repeatable and governable?

Lab boundaryThis is a hands-on reference workflow, not proof that a repository is universally safe or that one configuration works for every team. The Node application is only the runnable demo; the agent layer is stack-aware.

Repository
agentic-development-starter
Primary tool
VS Code + GitHub Copilot
Demo runtime
Node.js 20+
Supported patterns
React/TS · Spring Boot · Python · mixed stacks
Human gates
Plan · implementation · review
Verification
Repository-native

What you will inspect

The repository as an agent execution environment

3Specialized agent roles
5Reusable Agent Skills
3Stack guidance baselines
1Human-supervised workflow

Lab procedure

Run the workflow end to end

01

Inspect the repository contract

Open AGENTS.md and identify the universal engineering rules, stack detection behavior, verification expectations, and security boundaries.

02

Use Planner

Ask Planner to plan Exercise 2 from TASKS.md. Confirm it remains read-only, detects the Node demo toolchain, and identifies the correct verification command without editing or executing.

03

Approve the plan

Review scope, architecture impact, tests, risks, and verification before handing work to Implementer.

04

Use Implementer

Run the Start Implementation handoff. Confirm Implementer stays within the approved scope, adds tests, and runs repository-native verification.

05

Use Reviewer

Run Review Changes. Reviewer may inspect and execute verification, but cannot edit. Material findings must be handed back to Implementer.

06

Inspect skills and evals

Review .agents/skills and evals/ to see how repeatable procedures and agent-behavior regression scenarios complement the three agent roles.

07

Adapt to another stack

Apply the same agent layer to a React/TypeScript, Spring Boot, or Python repository while preserving that repository’s native architecture and toolchain.

Success criteria

What a healthy run should demonstrate

  • Planner plans from repository evidence without editing or executing.
  • Implementer uses the existing architecture, toolchain, tests, and verification gates.
  • Reviewer reports material findings without modifying source.
  • Skills load only when relevant rather than becoming permanent prompt noise.
  • Required verification failures are reported honestly, not bypassed.
  • Human approval remains visible between agent stages.

Run and adapt

Follow the lab into the working artifacts