🚧 Work in Progress

This CLI orchestrates three AI agents (Architect, Developer, Reviewer). Currently untested with real API calls. Requires funded API keys.

Code Forge — CLI that orchestrates multiple LLM agents (Architect, Developer, Reviewer) to build production-ready code.

AI-powered code analysis and development system using Claude/OpenAI/Deepseek LLM.

CodeForge is a command-line tool for code analysis and development using Large Language Models (LLM). It automates code reviews, architectural design, and component generation through an interactive workflow with roles: Architect, Developer, Reviewer.

❓ What is LLM?

LLM (Large Language Model) - AI neural network models that understand and generate text. CodeForge uses Claude/OpenAI/Deepseek for:

• Code analysis and review
• System architecture design
• Automatic TypeScript code generation
• Quality and security checks

🎯 Two Modes

REVIEW - Code Analysis

npm run dev -- review src/api.ts

Check code quality, security, and architecture.

DEVELOP - Interactive Development

npm run dev -- develop monitoring-system \
  --requirements "Monitoring system for 1000+ hosts"

Automatic component creation: Architect → Developer → Reviewer → Approval.

---

🚀 Installation & Setup

1. Install Dependencies

npm install

2. Configure API Key

# Copy example
cp .env.example .env

# Add Claude API key
export ANTHROPIC_API_KEY=sk-ant-...

3. Run

npm run dev -- review src/app.ts
npm run dev -- develop my-project --requirements "Your requirements"

4. Build for Production

npm run build

---

⚙️ Configuration

CodeForge looks for config.json in these locations (priority order):

1. Current directory: ./config.json
2. User home: ~/.codeforge/config.json

Example config files are provided:

• config.example.json - Simple mode configuration
• config.advanced.example.json - Advanced mode with all three models

Simple Mode: config.json

{
  "mode": "simple",
  "roles": {
    "model": "claude",
    "modelType": "claude-sonnet-4-5",
    "temperature": 0.7,
    "maxTokens": 4096
  },
  "tools": {
    "enabled": ["read_file", "write_file", "test_runner", "code_search"],
    "securityRules": {
      "allowedPaths": ["./src", "./tests", "./"],
      "forbiddenPaths": ["./.git", "./node_modules"]
    }
  },
  "apiKeys": {
    "claude": "sk-ant-..."
  }
}

Advanced Mode: config.json

{
  "mode": "advanced",
  "roles": {
    "architect": {
      "model": "claude",
      "modelType": "claude-opus-4-1",
      "temperature": 0.5,
      "maxTokens": 4096
    },
    "developer": {
      "model": "gpt",
      "modelType": "gpt-4-turbo",
      "temperature": 0.7,
      "maxTokens": 4096
    },
    "reviewer": {
      "model": "deepseek",
      "modelType": "deepseek-chat",
      "temperature": 0.3,
      "maxTokens": 2048
    }
  },
  "tools": {
    "enabled": ["read_file", "write_file", "test_runner", "code_search"],
    "securityRules": {
      "allowedPaths": ["./src", "./tests", "./"],
      "forbiddenPaths": ["./.git", "./node_modules"]
    }
  },
  "apiKeys": {
    "claude": "sk-ant-...",
    "openai": "sk-...",
    "deepseek": "sk-..."
  }
}

Environment Variables (.env)

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
DEEPSEEK_API_KEY=sk-...
LOG_LEVEL=info              # debug, info, warn, error
EDISON_MODE=advanced         # simple, advanced

Note: The config file in the root directory should not be committed to git. Use the example files (config.example.json, config.advanced.example.json) as templates.

---

📋 Command Examples

# Simple code review
npm run dev -- review src/handler.ts

# Advanced review (all roles)
npm run dev -- review src/api.ts --mode advanced

# Development with requirements
npm run dev -- develop payment-system \
  --requirements "Payment processing API"

# Output as Markdown
npm run dev -- review src/app.ts --format markdown --output analysis.md

# Development mode with verbose logs
npm run dev -- develop service --requirements "..." --verbose

---

🧪 Testing

npm test              # Run tests
npm test -- --watch   # Watch mode
npm run type-check    # Type checking

---

📚 Samples & Examples

Hello World Go Workflow Sample

Demonstrates the complete 3-agent orchestration workflow:

npx ts-node samples/hello-world-go-workflow.ts

This sample shows:

• Architect designing system architecture with 2 components
• Developer implementing components (with intentional errors in Core component)
• Reviewer analyzing code and finding critical issues (division by zero)
• Developer fixing issues based on reviewer feedback
• Reviewer approving fixed code
• Token tracking across all phases
• Cost estimation using model-specific pricing
• JSON output with metrics and workflow history

Output Files Generated:

• metrics-hello-world.json - Token usage and cost breakdown by agent and phase
• workflow-history-hello-world.json - Complete workflow execution history with task details

Features Demonstrated:

• Error detection and feedback loop
• Multi-iteration development with reviewer approval
• Different component behaviors (Service approves on first try, Core requires fixing)
• Aggregated metrics across all agents
• Structured task execution with proper state management

---

📝 Project Structure

src/
├── cli/              # Command-line interface
├── models/           # LLM adapters (Claude, GPT, DeepSeek)
├── agents/           # Roles: Architect, Developer, Reviewer
├── config/           # Configuration loading
├── tools/            # Tools (read_file, write_file, etc.)
└── core/             # Core logic

---

❓ How to Run as Global Command (codeforge)?

To run codeforge instead of npm run dev --:

Method 1: npm install -g (Recommended)

# In project folder
npm install -g .

# Now use anywhere:
codeforge review src/app.ts
codeforge develop system --requirements "..."

Method 2: Add bin to package.json

Edit package.json:

{
  "name": "edison-cli",
  "bin": {
    "codeforge": "./dist/cli/cli.js"
  }
}

Then:

npm install -g .
codeforge review src/app.ts

Method 3: Create alias

# In .bashrc or .zshrc:
alias codeforge="npm run dev --"

# Reload shell
source ~/.bashrc

I recommend Method 1 - the cleanest and most professional approach.

---

📄 License

MIT

---

Made for developers who care about code quality.