Multi-Agent Vibecoding Framework

A multi-agent development framework using Claude Code's native subagent capabilities for autonomous project development.

Overview

This framework provides a team of specialized AI agents that work together to build projects with minimal human intervention:

Agent	Role	Model
Product Owner	Defines requirements, user stories, and validates deliverables	opus
Project Manager	Coordinates work, breaks down tasks, tracks progress	sonnet
Software Architect	System design, API design, technology decisions	opus
Senior Engineer	Implements features, fixes bugs, and writes code	sonnet
Tester	Creates tests, runs test suites, and ensures quality	sonnet

Getting Started

Prerequisites

• Claude Code CLI installed
• Node.js 18+ (optional, for JavaScript/TypeScript projects)
• Python 3.10+ (optional, for Python projects)

Setup

1. Clone this repository:

   git clone https://github.com/YOUR_USERNAME/project_agent_framework.git
   cd project_agent_framework

2. The agents are automatically available in Claude Code. Verify with:

   /agents
   

Usage

Starting a New Feature

1. Define Requirements (Product Owner):

   Use the product-owner agent to define requirements for [feature description]
   

2. Plan & Coordinate (Project Manager):

   Use the project-manager agent to break down the requirements into tasks
   

3. Design (Software Architect):

   Use the software-architect agent to design the system architecture
   

4. Implement (Senior Engineer):

   Use the senior-engineer agent to implement [task description]
   

5. Write Tests (Tester):

   Use the tester agent to create tests for [feature]
   

Quick Reference

Action	Command
View agents	/agents
Define requirements	Use the product-owner agent to...
Coordinate tasks	Use the project-manager agent to...
Design architecture	Use the software-architect agent to...
Implement feature	Use the senior-engineer agent to...
Create tests	Use the tester agent to...

Project Structure

project_agent_framework/
├── .claude/
│   ├── agents/           # Agent definitions
│   └── CLAUDE.md         # Project configuration
├── docs/
│   ├── tasks/            # Task tracking
│   ├── architecture/     # Architecture Decision Records
│   ├── requirements/     # Requirements documentation
│   └── guides/           # Usage guides
├── templates/            # Task and documentation templates
├── src/                  # Source code
├── tests/                # Test files
├── package.json          # Node.js configuration
└── pyproject.toml        # Python configuration

Workflow

1. Product Owner defines clear requirements and user stories
2. Project Manager breaks down work into tasks and coordinates
3. Software Architect designs system architecture (for significant features)
4. Senior Engineer implements the solution
5. Tester creates comprehensive tests
6. Product Owner validates against original requirements

Task Tracking

Tasks are tracked in docs/tasks/ as markdown files. Each task includes:

• Status (TODO, IN_PROGRESS, IN_REVIEW, DONE, BLOCKED)
• Assigned agent
• Priority and dependencies
• Acceptance criteria

Expanding the Team

Additional agents can be added when needed:

• QA Reviewer: Final quality gate and security audits

See docs/guides/ for instructions on adding new agents.

License

MIT