🤖 AI Orchestration Workflow Template

Context Engineering in Action - A complete multi-agent AI development system for GitHub Copilot and Claude Code

🎯 What is This?

This repository provides a production-ready template for implementing Context Engineering with AI coding assistants. Instead of repeating the same instructions to AI every time, you build a persistent "knowledge system" that makes AI understand your project, follow your conventions, and work like a real development team.

The Problem

User: "Write an Angular component"
AI: *writes basic component*
User: "Don't forget standalone"
AI: *modifies*
User: "And signals"
AI: *modifies*
User: "And the new control flow"
... 😫 (repeat forever)

The Solution

Pre-defined agent personas with persistent context that remember your conventions:

# Frontend Developer Agent

## CORE PRINCIPLES
- Standalone components only - no NgModules
- Control flow: Use @if, @for, @switch
- Signals: Prefer input(), output(), viewChild()
- Change detection: OnPush required

Now the AI gets it right from the first try! ✨

---

🏗️ Architecture Overview

This template implements a multi-agent orchestration system where specialized AI agents collaborate like a real development team:

┌─────────────────────────────────────────────────────────────┐
│                    /orchestrate "task"                       │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  🪃 Orchestrator: Analyze task type, create TASK_ID         │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  📋 Project Manager: Create requirements & acceptance       │
│     criteria in BDD format                                   │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  ✋ USER VALIDATION: "APPROVED" or Feedback                  │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  🏗️ Software Architect: Evidence-based architecture design  │
│     (No hallucinated APIs - everything verified!)            │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  ✋ USER VALIDATION: "APPROVED" or Feedback                  │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  👥 Team Leader: Decompose into atomic tasks                 │
│     MODE 1: Create tasks.md                                  │
│     MODE 2: Assign → Verify → Commit (loop)                  │
│     MODE 3: Final verification                               │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  💻 Developer(s): Implement assigned tasks with git commits  │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  🧪 QA Phase: Testing + Code Review (user choice)           │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│  🎉 COMPLETE: All deliverables in task-tracking/TASK_ID/    │
└─────────────────────────────────────────────────────────────┘

---

📁 Repository Structure

.
├── .claude/                    # Claude Code configuration
│   ├── agents/                 # Agent persona definitions
│   │   ├── software-architect.md
│   │   ├── team-leader.md
│   │   ├── frontend-developer.md
│   │   ├── backend-developer.md
│   │   ├── project-manager.md
│   │   ├── senior-tester.md
│   │   ├── code-logic-reviewer.md
│   │   ├── code-style-reviewer.md
│   │   ├── researcher-expert.md
│   │   ├── ui-ux-designer.md
│   │   └── modernization-detector.md
│   └── commands/               # Custom slash commands
│       ├── orchestrate.md      # Main workflow command
│       ├── orchestrate-help.md
│       ├── review-code.md
│       ├── review-logic.md
│       └── review-security.md
│
├── .github/                    # GitHub Copilot configuration
│   ├── chatmodes/              # VS Code Copilot Chat personas
│   │   ├── workflow-orchestrator.chatmode.md
│   │   ├── software-architect.chatmode.md
│   │   ├── team-leader.chatmode.md
│   │   ├── frontend-developer.chatmode.md
│   │   ├── backend-developer.chatmode.md
│   │   ├── business-analyst.chatmode.md
│   │   ├── product-manager.chatmode.md
│   │   ├── senior-tester.chatmode.md
│   │   ├── code-reviewer.chatmode.md
│   │   ├── researcher-expert.chatmode.md
│   │   ├── ui-ux-designer.chatmode.md
│   │   └── modernization-detector.chatmode.md
│   ├── prompts/                # Phase-specific workflow prompts
│   │   ├── orchestrate.prompt.md
│   │   ├── phase1-project-manager.prompt.md
│   │   ├── phase2-researcher-expert.prompt.md
│   │   ├── phase3-ui-ux-designer.prompt.md
│   │   ├── phase4-software-architect.prompt.md
│   │   ├── phase5a-team-leader-mode1.prompt.md
│   │   ├── phase5b-team-leader-mode2.prompt.md
│   │   ├── phase5c-team-leader-mode3.prompt.md
│   │   ├── phase6-be-developer.prompt.md
│   │   ├── phase6-fe-developer.prompt.md
│   │   ├── phase7-code-reviewer.prompt.md
│   │   ├── phase8-modernization-detector.prompt.md
│   │   └── validation-gate.prompt.md
│   └── copilot-instructions.md # Global Copilot instructions
│
├── task-tracking/              # Generated task artifacts (created during use)
│   └── TASK_2025_XXX/          # Each task gets a folder
│       ├── context.md
│       ├── task-description.md
│       ├── implementation-plan.md
│       ├── tasks.md
│       └── ...
│
└── README.md                   # You are here!

---

🚀 Quick Start

For Claude Code Users

1. Copy the .claude folder to your project root:

   cp -r .claude /path/to/your/project/

2. Create a task-tracking folder in your project root:

   mkdir task-tracking

3. Start a workflow:

   /orchestrate "Add user authentication feature"
   

For GitHub Copilot Users (VS Code)

1. Copy the .github folder to your project root:

   cp -r .github /path/to/your/project/

2. Create a task-tracking folder in your project root:

   mkdir task-tracking

3. Open VS Code and use Copilot Chat with the workflow orchestrator chatmode

4. Start a workflow:

   /orchestrate "Add user authentication feature"
   

Using Both Together

The .claude and .github folders are designed to work together. Copy both to get the full experience with either tool!

---

🎭 Agent Personas

Each agent has a distinct personality and responsibility:

🪃 Workflow Orchestrator

Role: Coordinates the entire workflow

• Analyzes task type (FEATURE, BUGFIX, REFACTORING, etc.)
• Creates task tracking structure
• Manages agent handoffs

📋 Project Manager / Business Analyst

Role: Requirements specialist

• Creates SMART requirements
• Writes BDD acceptance criteria
• Validates scope

🏗️ Software Architect

Role: Evidence-based designer

• Key Principle: "Investigation, not assumption"
• Verifies every API exists in the codebase
• Creates implementation plans with file:line citations
• Anti-Hallucination Protocol: Never proposes unverified patterns

👥 Team Leader

Role: Task orchestration & verification

• MODE 1: Decomposes plans into atomic tasks
• MODE 2: Assigns tasks, verifies completion
• MODE 3: Final verification
• Golden Rule: "Don't trust - VERIFY"

💻 Frontend/Backend Developers

Role: Implementation specialists

• Follow established patterns from codebase
• Create git commits for each task
• Focus on real implementation (no stubs!)

🧪 Senior Tester

Role: Quality assurance

• Creates test plans
• Verifies acceptance criteria
• Documents test results

🔍 Code Reviewers

Role: Code quality guardians

• Style reviewer (patterns, conventions)
• Logic reviewer (business logic, edge cases)

---

💡 Key Concepts

Context Engineering vs Prompt Engineering

Prompt Engineering	Context Engineering
One simple prompt	Complete information system
For single LLM call	For multi-agent systems
No memory	Full memory management
No tools	Clear tool definitions
No validation	Defined quality gates

Anti-Hallucination Protocol

Every decision must be backed by evidence:

❌ AI might generate:
@Property({ primary: true }) // This doesn't exist!

✅ Our system requires:
// Investigation: Read entity.decorator.ts:145
@Neo4jEntity('User') // ✓ Verified at line 145

Atomic Tasks with Verification

❌ WRONG:
Task: "Implement entire landing page"
Result: AI claims it's done but 3/7 sections missing

✅ CORRECT:
Task 1: "Create HeroSection component"
Task 2: "Create AboutSection component"
...
→ Each task verified separately with git commits!

User Validation Checkpoints

The system pauses at key points for user approval:

1. After Requirements (Project Manager output)
2. After Architecture (Software Architect output)
3. QA Choice (Which quality checks to run)

---

📊 Results You Can Expect

Metric	Without Context Engineering	With This System
Hallucinated APIs	Many 😭	Zero ✅
Incomplete Tasks	40%+	<5%
Context Loss	Every conversation	Never
User Control	Minimal	Full
Traceability	None	Complete

---

🎬 Video Tutorial

This template is featured in our comprehensive video tutorial: "How to Build a Complete AI Development System: The Art of Context Engineering"

📺 Watch the video (Coming soon)

The video covers:

• What is Context Engineering and why it matters
• Setting up the multi-agent workflow
• How each agent works
• Real-world demonstration
• Best practices and anti-patterns

---

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Ideas for Contributions

• New agent personas for different domains
• Improved prompts for specific frameworks
• Translations to other languages
• Integration guides for other AI tools
• Bug fixes and improvements

---

📚 References & Credits

Research & Articles

• Context Engineering Guide
• Context Engineering for AI Agents
• LangChain: The Rise of Context Engineering

Key Quotes

"Prompt engineering is dead. Long live context engineering." — Andrej Karpathy

"Context engineering is the next phase, where you architect the full context, which in many cases requires going beyond simple prompting into more rigorous methods." — Prompt Engineering Guide

---

📄 License

MIT License - see LICENSE for details.

---

⭐ Star History

If this template helps you build better AI workflows, please give it a star! ⭐

---

Built with ❤️ by Hive Academy