Software Development Life Cycle Agency with Test-Driven Development (SDLC/TDD Agency)

An experimental AutoGen-based project exploring how multiple AI agents could potentially collaborate in a software development workflow. This prototype implements a Test-Driven Development (TDD) approach through agent orchestration.

Overview

The SDLC/TDD Agency investigates the possibilities of automating software development workflows using specialized AI agents. By implementing a basic Software Development Life Cycle (SDLC) with Test-Driven Development (TDD) principles, we explore how different agents might collaborate, learn, and potentially improve the development process.

Architecture

Core Components

1. Group Chat System

   • Manages communication between agents
   • Handles message routing and termination conditions
   • Maintains chat history and generates summaries

2. Agent System

   • User Proxy (Admin) Agent: Human interface for project oversight
   • Project Manager Agent: Orchestrates workflow and task management
   • Coder Agent: Implements features following TDD principles
   • Tester Agent: Manages test execution and quality assurance

3. Tools System

   • File management utilities
   • Documentation tools
   • Code execution capabilities
   • Project context management

Directory Structure

sdlc-tdd-agency/
├── main.py              # Main entry point
├── agency.md           # Agency definition and guidelines
├── src/
│   ├── agents.py       # Agent definitions and configurations
│   └── tools.py        # Utility functions and tool registrations
└── output/
    └── docs/           # Generated documentation

Current Capabilities

• Basic Workflow Automation (Experimental)

  • Task breakdown and assignment between agents
  • Initial progress tracking implementation
  • Basic conflict resolution patterns

• Test-Driven Development Exploration

  • Proof-of-concept test creation
  • Basic test execution and reporting
  • Initial code quality checks

• Documentation Management

  • Basic documentation generation
  • Simple version tracking
  • Project context experiments

• Learning Mechanisms (In Development)

  • Early-stage memory management through Mem0
  • Basic pattern recognition from interactions
  • Experimental workflow adaptations

Research Goals

This project aims to explore several key questions:

1. How effectively can AI agents collaborate in a software development context?
2. What are the limitations and challenges of current LLM-based agents in SDLC?
3. How might TDD principles be automated through multi-agent systems?

Usage

1. Environment Setup

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# or
.\venv\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements.txt

# Set up environment variables
cp .env.example .env
# Edit .env with your OpenAI API key

2. Running the Agency

python main.py

3. Interacting with the Agency The agency can be initiated with specific tasks through the UserProxyAgent:

chat_result = user_proxy.initiate_chat(
    manager,
    message="Your task description here"
)

Configuration

LLM Configuration

llm_config = {
    "model": "gpt-4o",
    "api_key": os.environ["OPENAI_API_KEY"],
    "temperature": 0,
}

Agent Configuration

Agents can be customized through their system messages and tool registrations. See src/agents.py for examples.

Tools and Utilities

The agency provides various tools for:

• File management
• Documentation creation and modification
• Code execution
• Project context management
• Test execution

Output

The agency generates:

• Chat summaries and history (chat_results.txt)
• Project documentation
• Code files
• Test reports

Current Limitations

• Agents may sometimes produce inconsistent or incorrect outputs
• Complex decision-making still requires human oversight
• Limited understanding of broader project context
• Experimental nature means workflows may need frequent adjustment

Next Steps

This experimental project is continuously evolving as we explore different approaches to AI-assisted software development. Current areas of investigation include:

1. Agent Team Expansion (Experimental)

Currently, the agency operates with four core agents: User Proxy (Admin), Project Manager, Coder, and Tester. We're exploring the addition of specialized agents to study how they might enhance the SDLC workflow:

1. Requirement Analyst Agent

   • Role: Gather and refine business requirements
   • Research Focus: Testing automated requirement analysis and user story generation
   • Current State: Early prototype stage

2. Design Architect Agent

   • Role: Develop system architecture and design
   • Research Focus: Exploring automated architecture decisions and trade-offs
   • Current State: Concept development

3. Docs Maintainer Agent

   • Role: Manage detailed and ongoing documentation
   • Research Focus: Testing automated documentation generation and maintenance
   • Current State: Basic implementation

4. Code Reviewer Agent

   • Role: Conduct code reviews to ensure quality
   • Research Focus: Studying automated code analysis patterns
   • Current State: Prototype testing

5. DevOps Agent

   • Role: Automate the deployment process
   • Research Focus: Investigating CI/CD automation patterns
   • Current State: Initial planning

2. Multi-Agent System Enhancements

To support the expanded agent team and improve overall functionality, we plan to investigate:

• Enhanced Collaboration: Improved communication protocols between agents
• Memory Management: Advanced implementation of Mem0 for better learning
• Workflow Optimization: Testing different process patterns
• Scalability Improvements: Exploring methods to handle larger projects

3. User Interface Development

We're evaluating several interface options for research purposes:

1. Streamlit

   • Description: Python-based framework for rapid prototyping
   • Research Value: Quick iteration on interaction patterns
   • Current Status: Initial testing

2. Gradio

   • Description: AI-focused interface framework
   • Research Value: Specialized AI model interactions
   • Current Status: Under evaluation

3. Dash

   • Description: Framework combining Flask, Plotly, and React
   • Research Value: Data visualization experiments
   • Current Status: Concept phase

4. TS/React

   • Description: Full-featured web application framework
   • Research Value: Complex interaction patterns
   • Current Status: Future consideration

Implementation Strategy

We'll follow an experimental approach:

1. Test new agents individually in controlled environments
2. Study multi-agent interactions through incremental integration
3. Prototype and evaluate interface solutions
4. Gather data on agent behaviors and effectiveness

License

MIT License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with AutoGen
• Inspired by modern software development practices and TDD principles