Agentic-Warden

Universal AI CLI Management Platform with Google Drive Sync

Agentic-Warden is an intelligent platform for managing AI CLI tools (Claude, Codex, Gemini) with MCP routing, process tracking, and Google Drive configuration synchronization.

🎯 Features

1. AI CLI Management (`REQ-001`, `REQ-002`, `REQ-006`, `REQ-011`)

Process Tree Tracking: Monitor AI CLI processes and their children
Provider Management: Switch between AI providers (OpenAI, Anthropic, Google, etc.)
Capability Detection: Auto-detect installed AI CLI tools
Wait Mode: Block until AI CLI tasks complete
Update Management: Install and update AI CLI tools

Usage:

# Launch AI CLI with process tracking
aiw claude "explain this code"
aiw codex "write a python script"
aiw gemini "translate to french"

# Wait for all AI CLI tasks to complete
aiw wait --timeout 300

# Manage providers
aiw provider list
aiw provider add openrouter "sk-xxx"

2. Google Drive Sync (`REQ-003`)

Synchronize AI CLI configurations across devices using Google Drive.

Setup OAuth:

export GOOGLE_CLIENT_ID="your_client_id.apps.googleusercontent.com"
export GOOGLE_CLIENT_SECRET="your_client_secret"

Commands:

# Push AI CLI configurations to Google Drive
aiw push [config_name]
# Example: aiw push claude
# Example: aiw push codex
# Example: aiw push gemini

# Pull configurations from Google Drive
aiw pull [config_name]

# List available backups on Google Drive
aiw list

Note: Currently supports only 3 predefined AI CLI directories:

~/.claude
~/.codex
~/.gemini

OAuth Flow:

Uses OAuth 2.0 Device Flow (RFC 8628)
Authentication tokens stored in ~/.aiw/auth.json
Detailed setup instructions printed during first use

3. Intelligent MCP Routing (`REQ-012`, `REQ-013`)

Route user requests to the best MCP tool with 98% token reduction.

How it works:

LLM ReAct Mode: Uses LLM reasoning to analyze requests
Vector Search Mode: Semantic similarity fallback
Dynamic Registration: Registers tools on-demand (98% token savings)
JavaScript Orchestration: Auto-generates workflows for complex tasks

Usage:

# Start MCP server
aiw mcp

# Configure in Claude Code (~/.config/claude-code/mcp.json)
{
  "mcpServers": {
    "agentic-warden": {
      "command": "aiw",
      "args": ["mcp"]
    }
  }
}

# Available MCP tools:
# - intelligent_route: Auto-route to best tool
# - get_method_schema: Get tool schemas
# - execute_tool: Execute tools with negotiation

4. MCP Server Management (`REQ-007`)

Manage external MCP servers with hot-reload support.

Commands:

# Add MCP server
aiw mcp add filesystem npx --description "File operations" \
  -- -y @modelcontextprotocol/server-filesystem /home/user

# List servers
aiw mcp list

# Enable/disable servers
aiw mcp enable brave-search
aiw mcp disable filesystem

# Edit configuration
aiw mcp edit

# Hot-reload: Changes apply instantly without restart

Configuration File: ~/.aiw/.mcp.json

🚀 Quick Start

Installation

# Clone and build
git clone https://github.com/putao520/agentic-warden.git
cd agentic-warden
cargo build --release

# Install binary
cargo install --path .

Initial Setup

# Verify installation
aiw --version

# Check available AI CLI tools
aiw status

# Configure MCP servers (optional)
aiw mcp add filesystem npx -- -y @modelcontextprotocol/server-filesystem $HOME

# Set up Google Drive sync (optional)
export GOOGLE_CLIENT_ID="your_client_id.apps.googleusercontent.com"
export GOOGLE_CLIENT_SECRET="your_client_secret"
aiw push claude

📊 Feature Matrix

Feature	REQ-ID	Status	Description
Process Tree Tracking	001	✅	Monitor AI CLI processes
Provider Management	002	✅	Switch AI providers
Google Drive Sync	003	✅	Backup/sync configurations
Wait Mode	005	✅	Block until tasks complete
Tool Detection	006	✅	Auto-detect AI CLI tools
MCP Servers	007	✅	External MCP integration
Intelligent Routing	012	✅	LLM-based tool routing
JS Orchestration	013	✅	Auto-generate workflows
AI CLI Updates	011	✅	Install/update tools
Role System	014	✅	AI persona management

Overall Status: ✅ All core features implemented and tested

🛠️ Configuration

Environment Variables

# AI CLI Configuration
export CLI_TYPE=claude              # claude, codex, or gemini
export CLI_PROVIDER=llmlite         # Any provider from provider.json

# MCP Server Configuration
export MCP_CONFIG_PATH=~/.aiw/.mcp.json

# Google Drive OAuth
export GOOGLE_CLIENT_ID=xxx
export GOOGLE_CLIENT_SECRET=xxx

# LLM Routing (optional)
export OPENAI_TOKEN=sk-xxx          # For Ollama/local LLM mode
export OPENAI_ENDPOINT=http://localhost:11434
export OPENAI_MODEL=qwen2.5:7b

Configuration Files

MCP Servers: ~/.aiw/.mcp.json
Providers: ~/.aiw/providers.json
AI CLI Auth: ~/.claude/config.toml, ~/.codex/config.json, etc.
Google Drive Auth: ~/.aiw/auth.json

🧪 Testing

# Run all tests
cargo test

# Run specific test module
cargo test capability_detection
cargo test sync::smart_oauth
cargo test mcp_routing

# Build and check
cargo check
cargo build --release

Test Coverage: 131/131 unit tests passing

📖 Documentation

Requirements: SPEC/01-REQUIREMENTS.md
Architecture: SPEC/02-ARCHITECTURE.md
API Design: SPEC/04-API-DESIGN.md
Testing Strategy: SPEC/06-TESTING-STRATEGY.md

🤝 Contributing

Contributions welcome! Please:

Check GitHub Issues
Follow existing code style
Add tests for new features
Update documentation

📜 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

MCP Protocol: modelcontextprotocol.io
Claude Code: Anthropic's official CLI for Claude
Anthropic Code Execution: Inspired by Code Execution with MCP

Agentic-Warden - Universal AI CLI Management Platform v6.0.1

aiw 0.5.4