🔥 AgentForge — Multi-Agent Code Review & Synthesis System

A production-grade, collaborative software engineering assistant built for the AI Agents: Intensive Vibe Coding Capstone Project (Freestyle Track).

AgentForge deploys a network of five specialized AI agents that review, critique, and co-develop software requirements. It showcases real-time agent coordination, OWASP-based security audits, an automated MCP static analysis tool, and a critique-and-refine loop — outputting a fully tested and documented codebase from a plain-English requirement.

---

📖 Table of Contents

1. Core Features
2. Orchestration Architecture
3. Course Concepts Demonstrated
4. Project Structure
5. Getting Started & Setup
6. Security & Compliance
7. Vercel Deployment Guide

---

🌟 Core Features

• Multi-Agent Orchestration: Autonomous, sequential agent pipelines coordinating through shared execution memory.
• Critique & Refinement Loop: The Code Generator updates its initial codebase in response to reviews from the Security and Architecture agents.
• Real-time Stream Pipeline: Built with Server-Sent Events (SSE) to stream live progress from individual agents straight to a modern, dark-themed dashboard.
• OWASP-based Audits: The Security Agent grades code against key secure coding criteria and vulnerability lists.
• Interactive Suggestion Cards: Quick-start templates to test the system immediately on standard scenarios.

---

📐 Orchestration Architecture

 User Input (Plain English Requirement)
                     │
                     ▼
             ┌───────────────┐
             │  Orchestrator │ ◄── [Server-Sent Events] ──► Dashboard UI
             └───────┬───────┘
                     │
           (1) Spawn CodeGen
                     │
                     ▼
             [CodeGen Agent] (Initial Draft)
                     │
          ┌──────────┴──────────┐
          ▼ (2)                 ▼ (3)
   [Security Agent]      [Architect Agent]
  (Vulnerability Scan)  (Scalability Check)
          └──────────┬──────────┘
                     │ (4) Feeds feedback back to
                     ▼
             [CodeGen Agent] (Performs Refactoring / Critique Loop)
                     │
                     ▼
          ┌──────────┴──────────┐
          ▼ (5)                 ▼ (6)
    [TestGen Agent]         [Doc Agent]
  (Unit/Integration)    (README & API Spec)

---

🎓 Course Concepts Demonstrated (Capstone Verification)

This project applies and strictly implements the core requirements for the Kaggle Capstone Project:

Requirement	Implementation Evidence in Codebase
Agent / Multi-agent system (ADK)	Implemented in lib/agents.js & lib/orchestrator.js. Includes a BaseAgent class handling API fallbacks and 5 specialized agents (CodeGen, Security, Architect, Test, Doc) operating in a coordinated critique loop.
MCP Server	Implemented in mcp_server.js & lib/mcp.js. A compliant Model Context Protocol server communicating over stdio via JSON-RPC 2.0. Exposes a check_code_quality tool that runs as a child process to perform static analysis.
Security features	Implemented in SecurityAgent & mcp_server.js. Code is evaluated against OWASP Top 10. The MCP tool actively scans for eval(), hardcoded secrets, and SQL injection patterns. Environment variables are strictly enforced.
Deployability	Implemented in pages/api/generate.js. A Vercel-ready Next.js app. Configured with "X-Accel-Buffering": "no" to ensure Server-Sent Events (SSE) stream perfectly through Vercel's edge proxies.

---

📁 Project Structure

AgentForge/
├── components/
│   ├── AgentCard.js       # Agent pipeline display component
│   ├── CodeBlock.js       # Markdown & syntax-highlighted code container
│   └── OutputTabs.js      # Tabs for switching results
├── lib/
│   ├── agents.js          # Defined LLM prompts & client thinking methods
│   ├── mcp.js             # MCP Client Integration (Spawns subprocess)
│   └── orchestrator.js    # Central orchestration loop logic
├── pages/
│   ├── _app.js            # Global wrapper
│   ├── index.js           # Interactive UI dashboard
│   └── api/
│       └── generate.js    # Streaming POST api endpoint
├── mcp_server.js          # Local MCP Server for Code Quality Tool
├── styles/
│   └── globals.css        # Premium dark glassmorphism styling
├── package.json
└── next.config.js

---

🛠️ Getting Started & Setup

Prerequisites

• Node.js (v18.x or above)
• GEMINI or Groq API Key

Installation

1. Clone the repository:

   git clone <your-repository-url>
   cd AgentForge

2. Install dependencies:

   npm install

3. Configure Environment Variables: Create a .env.local file in the root directory:

   cp .env.example .env.local

   Open .env.local and paste your Gemini API Key (get one at aistudio.google.com):

   GEMINI_API_KEY=your_gemini_api_key_here

4. Launch the development server:

   npm run dev

   Open http://localhost:3000 to view your local instance.

---

🛡️ Security & Compliance

• No API Keys in Source Code: The application strictly reads the Gemini API key via secure backend environment variables (process.env.GEMINI_API_KEY). Keys are validated for correct format before any LLM call is made.
• Input Sanitization: User requirements are enclosed inside strict, structured prompts, preventing agents from executing malicious or injected instructions.
• MCP Static Analysis: Every generated code block is scanned by the MCP server for eval() usage, hardcoded secrets, and SQL injection patterns before LLM-based review begins.

---

🚀 Vercel Deployment Guide

AgentForge is deployed live at https://multi-agent-code-review-plum.vercel.app — judges can test it directly without any local setup.

To self-host:

1. Deploy via Vercel CLI:

   npm install -g vercel
   vercel

2. Follow the prompt questions.
3. Add GEMINI_API_KEY (and optionally GROQ_API_KEY) to the project's Environment Variables in the Vercel Settings Panel.
4. Deploy to production:

   vercel --prod

Note: The X-Accel-Buffering: no header is already configured in pages/api/generate.js to ensure SSE streams correctly through Vercel's edge proxies.