🤖 AgentCorp: Autonomous AI Agent Office & Operating System

---

🌟 Overview

AgentCorp combines a Phaser 2D virtual office simulation (visual layer) with a Next.js runtime (execution layer) and real-time state management via Zustand.

Unlike standard mock agent runs, AgentCorp is an autonomous loop where specialized agents (CEO, PM, Engineer, Researcher, QA, HR) collaborate on software engineering tasks:

• 🧠 Task Decomposition: PlannerAgent breaks high-level tasks into logical subtask sequences (max 5 subtasks).
• 🤖 AI-Powered Execution: Worker agents generate deliverables via AI prompts (code, docs, reports, analyses).
• 🧪 Automated Validation: ValidatorAgent checks content quality with AI review and up to 3 retry attempts.
• 🔑 Intent-Aware Requests: IntentClassifier routes user requests to internal tasks or prompts for external credentials.
• 📈 Performance & Compliance: HR tracks efficiency, calculates morale, and manages agent promotions.

---

🎮 Virtual Office Live Simulation

The visual layout features a retro pixel-art office render:

• Workstations & Desks: Dedicated areas for PMs, Engineers, QA, and HR.
• The Coffee Machine: Agents automatically walk over to the coffee area to recharge when their energy falls below 40%.
• Meeting Room: Agents gather in the meeting room for status syncs and ambient chatter.
• Server Racks: Visually pulse to reflect system validation activities.

---

🏛️ System Architecture

graph TD
    User([User Request]) -->|classifyIntent| Intent[IntentClassifier]
    Intent -->|internal| Store[Zustand Game Store]
    Intent -->|needs_credentials| Cred[Credential Dialog]
    Cred -->|user provides keys| Store
    
    subgraph CentralOrchestrator[Central Orchestrator Tick]
        Phase1[CEO/PM: Prioritize & Assign] --> Phase2[Planner: Subtask Breakdown]
        Phase2 --> Phase3[Queue: Dispatch Jobs]
        Phase3 --> Phase4[Workers: Generate Deliverable via AI]
        Phase4 --> Phase5[Validator: Content + AI Review]
        Phase5 --> Phase6[HR: Performance Eval & Promotion]
    end

    Store --> Phase1
    Phase4 -->|deliverable| Store
    Phase5 -->|status pass/fail| Store
    Phase6 -->|logs & metrics| Store

---

🛠️ The Agent Roster

Role	Role Title	Description & Behavior
CEO	Chief Executive	Prioritizes tasks in the pending queue based on project needs.
PM	Project Manager	Delegates prioritized tasks to available engineers, researchers, or QA workers.
Engineer	Developer Agent	Generates code, documentation, and analysis content via AI prompts.
Researcher	Analyst Agent	Produces research reports with findings, analysis, and recommendations via AI.
QA	Quality Assurance	Reviews deliverables via AI and returns QA_PASS/QA_FAIL with reasoning.
HR	People Operations	Evaluates team morale, tracks task completion metrics, and processes promotions.

---

💻 Tech Stack

• Core Runtime: Next.js 16 (App Router), React 19 (Client-Side), TypeScript.
• Game Engine: Phaser 3 (2D Canvas Rendering, Agent A* Pathfinding).
• State Management: Zustand 5 (High-performance React/Phaser sync loop).
• Styling: Tailwind CSS + Framer Motion (Micro-animations, Glassmorphic overlays).
• AI Engine: Direct Google Gemini API Integration + OpenRouter Fallback.
• Streaming: Server-Sent Events (SSE) for execution console mirroring.

📂 Deliverables Directory

All deliverables generated by worker agents (reports, codebase analyses, validation outputs) are written directly to the ./deliverables/ directory in the filesystem, providing a transparent record of agent work.

Enjoy exploring AgentCorp! 🚀