Agent Kit

🚀 The Ultimate AI Agent System for Modern Development

22 Agents • 112 Skills • 30 Workflows • Multi-Platform Support

⚡ Quick Install

# Step 1: Install Agent Kit
npm install -g @musashishao/agent-kit

# Step 2: One-click setup (Venv + Deps + AI Hosts + MCPs)
ak setup

This command automatically creates an isolated environment, installs dependencies, and configures Claude Desktop, Cursor, and VS Code for your project.

This installs the .agent folder containing all templates into your project.

📦 What's Included

| Component | Count | Description | |-----------|-------|-------------| | Agents | 22 | Specialist AI personas (frontend, backend, security, etc.) | | Skills | 112 | Domain-specific knowledge modules | | Intelligence | 5 | RAG, Knowledge Graph, Memory, Verification Gate, and NotebookLM | | Workflows | 32 | Slash command procedures | | Templates | 5 | Project templates (web, mobile, backend, game) |

🔗 Multi-Platform Support

Agent Kit works seamlessly with multiple AI coding assistants:

| Platform | Status | Config File | |----------|--------|-------------| | Codex CLI | ✅ Full Support | AGENTS.md | | Gemini CLI | ✅ Full Support | GEMINI.md | | Cursor | ✅ Compatible | .cursor/rules | | Claude Code | ✅ Compatible | CLAUDE.md | | Aider | ✅ Compatible | AGENTS.md |

🎯 Codex CLI Integration

Quick Setup

# Step 1: Install Agent Kit
npx @musashishao/agent-kit init

# Step 2: Generate AGENTS.md for your project
npx @musashishao/agent-kit codex --template web

# Step 3: Sync workflows as slash commands
npx @musashishao/agent-kit setup-codex

# Step 4: Verify installation
npx @musashishao/agent-kit doctor

Available Templates

| Template | Command | Use Case | |----------|---------|----------| | Web | codex --template web | Next.js, React, Tailwind | | Mobile | codex --template mobile | React Native, Flutter | | Backend | codex --template backend | Node.js, Express, APIs | | Default | codex | Generic projects |

Using Workflows in Codex CLI

After running setup-codex, use Agent Kit workflows as slash commands:

# Create new features
/kit-create blog with authentication

# Debug issues
/kit-debug why login returns 401

# Design premium UI
/kit-ui-ux-pro-max dashboard with glassmorphism

# Quality optimization
/kit-quality create production-ready checkout

🛠 CLI Commands

| Command | Description | |---------|-------------| | setup | Full One-Click Setup (Venv + Deps + MCPs) | | init | Install .agent folder into your project | | codex | Generate AGENTS.md for Codex CLI | | setup-codex | Sync workflows with Codex CLI slash commands | | agents | List all available agents | | skills | List all skills by category | | workflows | List all workflows | | status | Check installation status | | sync | Advanced Sync (Graph + RAG + Memory) | | memory | Manage JSON Memory & Session state | | set-lang | Set Agent & Workflow language (vi/en) | | setup-notebooklm | Setup NotebookLM MCP integration | | setup-intel-verifier | Setup Intel Stack (Brave Search) | | setup-memory-mcp | Setup Memory MCP (Knowledge Graph) | | setup-notion-mcp | Setup Notion MCP integration |

Options

# Codex template options
agent-kit codex --template web      # Web application
agent-kit codex --template mobile   # Mobile app
agent-kit codex --template backend  # Backend API
agent-kit codex --force             # Overwrite existing

# Init options
agent-kit init --force              # Overwrite existing .agent folder
agent-kit init --path ./myapp       # Install in specific directory

# Setup-codex options
agent-kit setup-codex --prefix my-  # Custom prefix for slash commands

# Memory management
agent-kit memory get --type session # View current session state
agent-kit memory update-task --task "X" --status "completed"

# Language management
agent-kit set-lang vi               # Switch to Vietnamese
agent-kit set-lang en               # Switch to English (default)

🌐 Localization & Multi-language

Agent Kit now supports full localization for the AI experience.

Supported Languages

• 🇺🇸 English (en) - Default
• 🇻🇳 Vietnamese (vi) - Professional translation

How it works

Running ak set-lang vi performs three actions:

1. AI Instruction Update: Updates GEMINI.md logic to ensure the AI responds in your chosen language.
2. Workflow Swapping: Physically updates the description field in all .agent/workflows/*.md files so that CLI tools (Codex, Antigravity) display localized help text.
3. Internal Config: Saves preference to .agent/memory/user.json.

Note: If ak command is outdated, use npx @musashishao/agent-kit@latest set-lang vi

🔌 MCP Integration

Agent Kit includes built-in MCP (Model Context Protocol) servers to extend your AI assistant's capabilities.

Quick Setup

Agent Kit uses the Intel Stack to extend your AI assistant's capabilities.

# Setup Intel Stack
ak setup-intel-verifier

# Setup NotebookLM (Deep Research)
ak setup-notebooklm

# Restart your AI client

Built-in Servers (MCP Stack)

| Server | Tools | Description | |--------|-------|-------------| | NotebookLM | 42 | Research, RAG, Audio/Video gen | | Brave Search | 1 | Real-time web search | | Memory MCP | 10 | Long-term Knowledge Graph | | Notion | 15 | Workspace integration |

Available Templates

| Template | Description | |----------|-------------| | api | REST API integration | | database | SQLite/PostgreSQL connector | | web-scraper | Web page scraper | | custom | Minimal starter template |

📚 NotebookLM Integration (New)

Agent Kit deeply integrates with Google NotebookLM to provide Source-Grounded Reasoning (Reasoning based on actual facts).

Key Features

• Deep Research: Scours 40+ trusted web sources for project planning.
• Context Offloading: Offloads long documents (PDF, Docs) to NotebookLM to free up the Context Window.
• Evidence-Based: AI always provides source citations from your Notebook.
• Content Generation: Automatically generates Audio Podcasts, Slides, and Briefing Docs from your project.

Quick Setup

# Step 1: Run the automated setup script
ak setup-notebooklm

# Step 2: Run the Auth command (displayed after Step 1)
# ./venv/bin/notebooklm-mcp-auth

# Step 3: Test the research workflow
/research AI Agentic Workflow 2026

🔍 Intel Stack (New)

The Intel Stack provides real-time search for Vibe Coding:

| MCP Server | Purpose | API Key | |------------|---------|---------| | Brave Search | Real-time technical web search | Required |

Quick Setup

# One command to set up Brave Search
ak setup-intel-verifier

# Or via npx
npx @musashishao/agent-kit setup-intel-verifier

# Restart Codex/Antigravity

# Verify
gemini mcp list

Usage Examples

# Brave Search - Real-time research
"Search for 'Next.js 15 release notes'"

Multi-Client Config

The setup script writes to ~/.gemini/antigravity/mcp_config.json. For other clients, copy the printed JSON to:

| Client | Config Path | |--------|-------------| | Cursor | ~/.cursor/mcp.json | | VS Code | ~/.vscode/mcp.json | | Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json |

📁 Project Structure

.agent/
├── agents/          # 16 Specialist Agents
├── skills/          # 42 Skills
├── memory/          # 🧠 AI Session & State
├── workflows/       # 17 Slash Commands
├── rules/           # Platform-specific rules
│   ├── CODEX.md    # Codex CLI rules
│   └── GEMINI.md   # Gemini CLI rules
├── templates/       # Project templates
└── ARCHITECTURE.md  # Full documentation

🎯 Usage

Using Agents

Mention an agent by name to invoke specialized expertise:

Use the @security-auditor to review authentication
Use the @frontend-specialist to analyze React components
Use the @debugger to fix the login issue

Using Skills

Skills are loaded automatically based on task context. The AI reads skill descriptions and applies relevant knowledge.

Using Workflows

Invoke localized workflows with slash commands:

| Command | Description | |---------|-------------| | /autofix | Autonomous self-healing loop for failed commands | | /brainstorm | Structured brainstorming for projects and features | | /context | Auto-generate optimal context for complex tasks | | /create | Create new applications with AI-Ready infra | | /enhance | Add or update features in an existing app | | /spec | Create technical specifications before planning | | /plan | Create detailed project plans and task breakdowns | | /tasks | Generate executable task lists from plans | | /orchestrate | Coordinate 3+ specialized agents for complex tasks | | /debug | Systematic problem investigation and root cause analysis | | /ui-ux-pro-max | Professional UI/UX planning with 50+ styles | | /test | Generate and execute comprehensive test suites | | /quality | Optimize context and output for high-quality results | | /status | Display project and agent health status | | /deploy | Production deployment with pre-flight checks | | /preview | Manage local development and preview servers | | /saas | Build SaaS features (billing, landing, etc.) | | /marketing| Growth marketing and campaign planning | | /pentest | Run security audits and penetration tests | | /ai-agent | Build specialized AI agents and loops | | /gamekit-init| Initialize game project structure | | /dashboard | Create and view visual project dashboards | | /compact-docs| Compact and distill decisions into ADRs | | /next | Suggest the next logical steps for development |

💡 Tip: Use ak set-lang vi to see these descriptions in Vietnamese!

Example:

/kit-create landing page with hero section
/kit-debug why login fails
/kit-ui-ux-pro-max dashboard with glassmorphism

🌐 Platform Setup Guides

Gemini CLI

# Install Agent Kit
npx @musashishao/agent-kit init

# Configure Gemini settings
echo '{ "contextFileName": ".agent/rules/GEMINI.md" }' > .gemini/settings.json

Cursor IDE

# Install Agent Kit
npx @musashishao/agent-kit init

# Copy rules to Cursor
mkdir -p .cursor/rules
cp .agent/rules/GEMINI.md .cursor/rules/agent-kit.md

Claude Code

# Install Agent Kit
npx @musashishao/agent-kit init

# Copy rules for Claude
cp .agent/rules/GEMINI.md CLAUDE.md

📊 Quality Validation

Ensure your setup is optimal:

# Run health check
npx @musashishao/agent-kit doctor

# Run quality validator
python3 .agent/skills/context-engineering/scripts/quality_validator.py .

Expected output:

📊 Summary
  Passed: 32/32 (100%)
  ✓ Excellent! Agent Kit is fully configured for Codex CLI.

📚 Documentation

• Website: agent-kit.musashi.works
• AGENTS.md: Project root
• ARCHITECTURE.md: .agent/ARCHITECTURE.md
• Platform Guides: /docs/platforms/

🤝 Contributing

We welcome contributions! Please see .agent/ARCHITECTURE.md for guidelines.

License

MIT © Musa (musashishao)