🧠 AgentMind

Collaborative multi-agent orchestration for ANY AI IDE

AgentMind transforms your AI coding assistant into a collaborative workspace where multiple specialist agents work together — planning, coding, reviewing, testing, and debugging — all coordinated by an intelligent Team Lead.

Works with Claude Code, VS Code Copilot, Cursor, Windsurf, and any AI-powered IDE.

---

🚀 Installation

One-Line Install

macOS / Linux:

git clone --depth 1 https://github.com/AbdulRahmanAzam/AgentMind.git && bash AgentMind/install.sh

Windows (PowerShell):

git clone --depth 1 https://github.com/AbdulRahmanAzam/AgentMind.git; powershell -ExecutionPolicy Bypass -File AgentMind\install.ps1

The installer auto-detects your IDEs and copies agent/skill files to the right locations:

IDE	Agents Directory	Skills Directory
Claude Code	~/.claude/agents/	~/.claude/skills/agentmind/
VS Code Insiders	%APPDATA%/Code - Insiders/User/prompts/ (Win) · ~/.config/Code - Insiders/User/prompts/ (Linux) · ~/Library/Application Support/Code - Insiders/User/prompts/ (macOS)	~/.agents/skills/agentmind/
VS Code	%APPDATA%/Code/User/prompts/ (Win) · ~/.config/Code/User/prompts/ (Linux) · ~/Library/Application Support/Code/User/prompts/ (macOS)	~/.agents/skills/agentmind/
Cursor	~/.cursor/agents/	~/.cursor/skills/agentmind/
Windsurf	~/.codeium/windsurf/agents/	~/.codeium/windsurf/skills/agentmind/

Uninstall

# macOS / Linux
bash AgentMind/uninstall.sh

# Windows (PowerShell)
powershell -ExecutionPolicy Bypass -File AgentMind\uninstall.ps1

---

✨ How It Works

You give AgentMind a request. The Team Lead assembles specialists, creates a task plan, and coordinates execution:

User: "Build a REST API with auth, tests, and docs"
         │
         ▼
┌──────────────────┐
│   🧠 Team Lead    │  Understands → Plans → Delegates → Monitors → Verifies
└────────┬─────────┘
         │
    ┌────┼────┬────────┐
    ▼    ▼    ▼        ▼
  ⚙️     🧪    🔒       📝
 Backend Test Security  Docs
    │    │    │        │
    └────┴────┴────────┘
         │
   Shared Workspace (.agentmind/)
   Tasks · Mailbox · Status

The 5 Phases

1. Onboarding — Lead understands your request, asks clarifying questions
2. Planning — Decomposes into tasks with dependencies, assembles the right team
3. Execution — Each agent deep-researches, then implements step by step
4. Verification — Build, test, lint, typecheck — debug until everything passes
5. Completion — Summary of what was built, files changed, architecture decisions

Agent Deep Research

Every agent must research before coding:

• Read existing codebase thoroughly
• Understand patterns and conventions
• Check what other agents have built
• Plan their approach step by step
• Ask questions when confused (via mailbox)

Inter-Agent Communication

Agents talk to each other through .agentmind/mailbox/:

{"from": "agentmind-backend", "to": "agentmind-test", "content": "Auth endpoints ready at src/routes/auth.ts. Test /register, /login, /refresh.", "type": "direct"}

The Lead monitors all communication, answers questions, and ensures smooth handoffs.

---

🎭 Available Agents

Agent	Icon	Role	Access
agentmind-lead	🧠	Team Lead — orchestrates everything	Full
agentmind-backend	⚙️	APIs, databases, server logic, auth	Full
agentmind-frontend	🎨	UI components, styling, state management	Full
agentmind-test	🧪	Unit, integration, and e2e tests	Full
agentmind-security	🔒	OWASP audits, vulnerability scanning	Read-only
agentmind-reviewer	👁️	Code quality, patterns, readability	Read-only
agentmind-devops	🚀	CI/CD, Docker, deployment	Full
agentmind-docs	📝	README, API docs, architecture guides	Full
agentmind-perf	⚡	Profiling, caching, optimization	Full

---

💬 Quick Start

Claude Code

claude
# then type:
/agentmind Build a full-stack todo app with React, Express, and PostgreSQL

VS Code Copilot

Open the Chat panel and type:

@agentmind-lead Build a full-stack todo app with React, Express, and PostgreSQL

Cursor / Windsurf

Reference agentmind-lead in chat:

@agentmind-lead Build a full-stack todo app with React, Express, and PostgreSQL

What Happens Next

1. The Lead asks you to confirm the team (e.g., Backend + Frontend + Test + Security)
2. You say "yes" or adjust
3. Agents start working — each in their own session
4. They communicate, build, test, and debug collaboratively
5. Lead verifies everything passes and gives you the summary

---

🏗️ Architecture

AgentMind/
├── agents/                         # Agent definitions (.agent.md format)
│   ├── agentmind-lead.agent.md     # Team Lead — the orchestrator
│   ├── agentmind-backend.agent.md  # Backend Developer
│   ├── agentmind-frontend.agent.md # Frontend Developer
│   ├── agentmind-test.agent.md     # Test Engineer
│   ├── agentmind-security.agent.md # Security Reviewer (read-only)
│   ├── agentmind-reviewer.agent.md # Code Reviewer (read-only)
│   ├── agentmind-devops.agent.md   # DevOps Engineer
│   ├── agentmind-docs.agent.md     # Documentation Writer
│   └── agentmind-perf.agent.md     # Performance Optimizer
├── skills/
│   └── agentmind/
│       └── SKILL.md                # Main orchestration skill & protocol
├── install.sh                      # Unix/macOS installer
├── install.ps1                     # Windows installer
├── uninstall.sh                    # Unix/macOS uninstaller
├── uninstall.ps1                   # Windows uninstaller
├── LICENSE
└── README.md

Workspace Communication Files

When agents are working, they create .agentmind/ in your project:

your-project/
└── .agentmind/
    ├── tasks.json                  # Task list with statuses & dependencies
    ├── mailbox/
    │   ├── broadcast.jsonl         # Team-wide announcements
    │   ├── lead.jsonl              # Messages to the Lead
    │   ├── agentmind-backend.jsonl # Messages to Backend Dev
    │   └── ...                     # One file per agent
    └── state/
        └── team.json               # Current team roster & phase

---

⚙️ Task System

Tasks have statuses and dependency tracking:

Status	Meaning
pending	Ready to start (no unmet dependencies)
blocked	Waiting for dependency tasks to complete
in-progress	Agent is actively working
completed	Done successfully
failed	Failed — Lead will create fix tasks

Priorities: critical → high → medium → low

---

🔒 Security

• Read-only agents: Security Reviewer and Code Reviewer cannot modify files
• Path traversal protection: Agents validate file paths within the workspace
• Dangerous command blocking: Recursive deletes, disk format, fork bombs are blocked
• No secrets in code: Agents use environment variables for sensitive config

---

🗺️ Roadmap

• Custom agent definitions (bring your own specialist)
• Persistent team configurations across projects
• Web dashboard for monitoring agent progress
• MCP server integration for external tools
• Agent memory — learn from past projects

---

📄 License

MIT © 2025 AgentMind Contributors

---

🙏 Acknowledgments

• Inspired by claude-seo for the agent/skill architecture
• The multi-agent collaboration pattern from software engineering teams
• The open-source AI tooling community