Skip to content

Repository files navigation

🌐 English | 中文

Agent Harness Engineering on Azure

🚀 180-minute enterprise workshop on Agent Harness Engineering — from basic completions to multi-agent collaboration, master Copilot engineering practices end to end.

🆕 What's New in v2

Dimension v1 v2
Duration 120 min 180 min
Labs 3 Labs 5 Labs
Agents 3 Agents 10 Agents
Prompt Templates 1 Prompt 6 Prompts

📁 File Listing

File Description
README.md Project README (English)
zh/README.md Project README (Chinese)
docs/DESIGN.md Design Principles (English)
docs/enterprise-deployment-guide.md Enterprise Deployment Guide (Azure + VM)
docs/agent-security-sandbox.md Agent Security & Sandbox Isolation
docs/managed-harness-platforms.md Managed Harness platforms compared — AWS AgentCore vs. Azure AI Foundry (incl. Azure China gate)
zh/docs/ Chinese versions of all docs
.github/copilot-instructions.md Copilot global instructions
.github/agents/*.md 10 custom agent definitions
.github/prompts/*.prompt.md 6 reusable prompt templates
.github/workflows/ci.yml CI/CD pipeline (Azure Container Apps + VM)
SOUL.md Agent Soul — identity & personality definition
labs/lab1labs/lab5 Lab materials
src/ Sample application source code (Ticket Service)
package.json Node.js project configuration

🤖 Agent Table (10 Agents)

Agent Role Responsibility
code-reviewer Code Reviewer Review code quality, style, and best practices
architect Architect System design, tech selection, architecture decisions
investigator Investigator Deep-dive root cause analysis and context gathering
release-engineer Release Engineer Manage release process, versioning, and changelogs
product-reviewer Product Reviewer Evaluate feature completeness and UX from a product perspective
performance-engineer Performance Engineer Identify performance bottlenecks and provide optimizations
red-team Red Team Adversarial testing to uncover attack surfaces
security-reviewer Security Reviewer Check for security vulnerabilities and compliance
test-engineer Test Engineer Write and optimize test cases, improve coverage
doc-writer Doc Writer Generate and maintain technical and API documentation

📝 Prompt Template Table (6 Prompts)

Template Purpose Typical Scenario
add-endpoint Add API Endpoint Quickly generate RESTful routes, controllers, and tests
fix-bug Fix Bug Locate and fix bugs based on issue descriptions
code-review Code Review Structured review and feedback on PRs
design-feature Feature Design Output feature plan, data models, and API definitions
investigate-issue Issue Investigation Analyze logs and stack traces to identify root cause
ship-release Ship Release Generate changelog, tags, and release checklist

🧪 Lab Overview (5 Labs)

Lab Topic Duration Focus
Lab 1 Copilot Basics & Code Completion 30 min Inline completion, Chat panel, context awareness
Lab 2 Custom Instructions & Prompt Engineering 30 min copilot-instructions.md, prompt templates
Lab 3 Agent Collaboration (Agent Mode) 40 min Multi-agent invocation, role division
Lab 4 End-to-End Feature Development 40 min Full workflow from design to release
Lab 5 Security Review & Red Team 40 min Security scanning, adversarial prompts, defense strategies
Lab 4+ (Bonus) Spec-Driven Development with GitHub Spec-Kit 30–35 min Industrialize the hand-crafted multi-agent handoff from Lab 3 into a repeatable SDD workflow via Spec-Kitspecify → plan → tasks → implement

📎 Lab 4+ positioning: Lab 3 demonstrates hand-crafted multi-agent handoff (designer → reviewer → tester). Lab 4+ shows how to industrialize that same pattern with GitHub Spec-Kit's SDD toolchain — turning ad-hoc Agent chaining into reproducible spec-driven engineering. See labs/lab4-spec-driven-development.md.

✅ Prerequisites

  • VS Code latest version (1.90+ recommended)
  • GitHub Copilot active license (Individual / Business / Enterprise)
  • Node.js 18+ (for running the sample project)
  • Basic Git proficiency

🚀 Quick Start

# 1. Clone the repository
git clone https://github.com/<org>/copilot-workshop.git
cd copilot-workshop

# 2. Install dependencies
npm install

# 3. Open in VS Code
code .

# 4. Ensure the Copilot extension is enabled, then navigate to labs/lab-1/ to begin

📚 References

📄 License

MIT

About

A hands-on workshop for building AI agent systems with Azure + GitHub Copilot. Covers the 4-layer Harness model (Rule → Soul → Role → Tool), 10 specialized agents, multi-agent collaboration, enterprise deployment, and security sandbox. 120-min bilingual (EN/ZH) curriculum.

Resources

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages