Multi-agent AI development pipeline using GitHub Copilot agents + OpenSpec.
.github/agents/— 8 specialized AI agents (orchestrator, advisor, devs, testers, writer)openspec/— Spec-driven development framework configAGENTS.md— Instructions for AI assistantsproject.md— Project context template (fill this first!)
.devcontainer/— GitHub Codespaces config (auto-installs Python, Node, OpenSpec)
- GitHub Copilot with agent mode enabled
- VS Code (local or Codespaces)
- Node.js 20.19+ (auto-installed in Codespaces)
Fill in openspec/project.md with your project details.
- Edit manually, OR
- Ask an agent:
@orchestrator Help me fill in openspec/project.md for a [describe project]
openspec initThis creates the openspec/specs/ and openspec/changes/ directories.
Tell the orchestrator what to build:
@orchestrator I want to build [describe your feature]
The orchestrator will:
- Call
@advisorto analyze requirements and propose a plan - Call
@writerto write specifications (if needed) - Route implementation to
@junior-dev,@senior-dev, or@ui-dev - Verify with
@testeror@ui-tester
| Agent | Role |
|---|---|
@orchestrator |
Central coordinator — routes work through the pipeline |
@advisor |
Analyzes requirements, designs architecture, debugs complex issues |
@writer |
Documentation: OpenSpec proposals, specs, changelogs, reports |
@senior-dev |
Complex implementation: architecture-heavy, cross-cutting code |
@junior-dev |
Routine implementation: pattern-following, scaffolding, bulk coding |
@tester |
User proxy: runs code, inspects outputs, reports honestly |
@ui-dev |
Frontend implementation: HTML/CSS/JS components |
@ui-tester |
Visual QA: screenshot-based verification |
→ See .github/agents/README.md for full details, workflow diagrams, and cost breakdown.
The pipeline follows: Analyze → Implement → Test → Document.
The @orchestrator coordinates everything — you just tell it what you want built.
MIT