A sophisticated multi-agent orchestration system for Claude Code, featuring parallel execution, multiple workflow patterns, and advanced coordination strategies.

Ready to use: A clean, domain-agnostic multi-agent framework for any software development project.

This repository provides a complete multi-agent development framework that allows you to:

• 🚀 Execute agents in parallel for 40-75% faster development
• 🎯 Choose optimal workflow patterns (Sequential, Parallel, Hybrid, Adaptive)
• 🤝 Coordinate specialized agents for complex projects
• 📊 Track metrics to optimize your workflow
• 🔄 Synthesize results from concurrent agent work

/orchestrator-v2 <your-requirements-file.md>

The orchestrator will:

1. Analyze your project characteristics
2. Recommend the best workflow pattern
3. Launch agents (in parallel when beneficial)
4. Synthesize results and manage human validation gates
5. Track time saved and quality metrics

Not sure? Start with Hybrid - it provides the best balance.

Located in .claude/commands/:

• orchestrator-v2.md - Enhanced orchestrator with parallel execution
• architect.md - System design and architecture
• developer.md - Code implementation
• qa-tester.md - Quality assurance and testing
• documentation.md - Technical writing
• project-manager.md - Planning and coordination

Located in .claude/commands/patterns/:

• sequential-workflow.md - Traditional waterfall approach
• parallel-workflow.md - Maximum concurrency
• hybrid-workflow.md - Sequential phases, parallel tasks
• adaptive-workflow.md - Dynamic per-phase optimization

Located in .claude/commands/coordinators/:

• parallel-coordinator.md - Manages concurrent agents
• task-synthesizer.md - Merges parallel outputs

• HOW_TO_USE_MULTI_AGENT.md - Complete usage guide
• docs/agentic-patterns/WHEN_TO_USE_WHAT.md - Pattern selection guide
• docs/agentic-patterns/PARALLEL_EXECUTION_GUIDE.md - Parallel execution details

• docs/agentic-patterns/AGENTIC_PATTERNS.md - Comprehensive pattern guide

/orchestrator-v2 my-project-requirements.md

Select "Sequential" when prompted. Agents run one at a time.

/orchestrator-v2 my-project-requirements.md

Select "Hybrid" when prompted. Sequential phases with parallel tasks:

Phase 1: Planning (Sequential)
    ↓
Phase 2: Architecture (PARALLEL)
    ├─→ Data Model Agent
    ├─→ API Design Agent
    └─→ Integration Agent
    ↓ [Synthesis]
    ↓ [Human Gate]
Phase 3: Development (PARALLEL)
    ├─→ Feature A Agent
    ├─→ Feature B Agent
    └─→ Feature C Agent
    ↓ [Integration]

/orchestrator-v2 my-project-requirements.md

Select "Parallel" for maximum speed. All independent work runs concurrently.

Best for: Learning, highly coupled systems, simple projects

How it works: Agents run one at a time in waterfall fashion

Time: Baseline (100%)

Best for: Research, exploration, independent features

How it works: All agents run concurrently, then synthesize

Time: ~25-40% of sequential (60-75% faster)

Best for: Most real-world projects

How it works: Sequential phases, parallel tasks within each phase

Time: ~50-60% of sequential (40-50% faster)

Example:

Sequential: 10 hours
Hybrid: 5-6 hours (save 4-5 hours)
Parallel: 2.5-4 hours (save 6-7.5 hours, but harder to manage)

Best for: Experienced users wanting per-phase optimization

How it works: Chooses best strategy for each phase dynamically

Time: ~40-50% of sequential (50-60% faster)

The system includes human validation checkpoints:

🚦 HUMAN VALIDATION REQUIRED

Phase: Architecture
Pattern Used: Hybrid
What Needs Approval: System design

Summary:
- Designed data model
- Defined API interfaces
- Created integration architecture

Deliverables:
- [Link to architecture doc]

To Proceed:
- ✅ Approved - continue
- 📝 Feedback - [adjustments needed]
- ❌ Rejected - [concerns]

The orchestrator tracks:

1. Create a new file in .claude/commands/:

touch .claude/commands/my-custom-agent.md

2. Follow the template structure:

---
description: Brief description of agent role
---

# Agent Name

You are a **[Role]** for [project type].

## Your Responsibilities

- Responsibility 1
- Responsibility 2

## Instructions

Step-by-step instructions for the agent...

## Output Format

Expected deliverables...

3. Invoke your custom agent:

/my-custom-agent <context>

1. Create a new file in .claude/commands/patterns/:

touch .claude/commands/patterns/my-pattern.md

2. Define the workflow structure
3. Reference it in your orchestrator calls

✅ Good candidates for parallel:

• Independent features
• Different architectural areas
• Separate test suites
• Different documentation types

❌ Poor candidates for parallel:

• Tightly coupled code
• Core engine/kernel
• Sequential algorithms
• Single coherent document

When merging parallel work:

1. Review all outputs thoroughly
2. Resolve conflicts with clear rationale
3. Document integration decisions
4. Test integrated result

Place validation gates at:

• Phase boundaries
• Before irreversible decisions
• After parallel synthesis
• Major deliverables

Solution: Use task-synthesizer or switch to hybrid/sequential pattern

Solution: Reduce number of parallel agents or provide clearer specifications

Solution: Switch patterns mid-project (it's okay!)

This multi-agent system is designed to be extended. Contributions welcome:

1. New agent types
2. New workflow patterns
3. New coordinators
4. Documentation improvements
5. Example projects

[Specify your license]

Built using Claude Code's multi-agent capabilities with the Task tool for parallel execution.

Ready to build something?

/orchestrator-v2 your-project-requirements.md