@maestroai/cli

Orchestrate multiple Claude Code agents working as a team

Installation

# Install globally
npm install -g @maestroai/cli

# Or use npx (no installation needed)
npx @maestroai/cli init

Prerequisites

• Node.js >= 20.0.0
• Claude CLI - Install from https://claude.ai/code

Installing Claude CLI

# macOS / Linux
npm install -g @anthropic-ai/claude-code

# Or with Homebrew (macOS)
brew install anthropic/claude/claude-code

Usage

Initialize a new project

maestro init

This will:

• Check that Claude CLI is installed
• Create a maestro.yaml configuration file
• Create a .maestro/ directory for messages and logs
• Create a .agents/ directory with customizable agent prompts
• Generate a todo.md with planned tasks (optional)

Start the orchestration

maestro start

Launches the TUI (Terminal User Interface) for monitoring agents.

Check status

maestro status

Shows running agents and their states.

Stop orchestration

maestro stop

Gracefully stops all running agents.

Configuration

Edit maestro.yaml to customize:

• Number and types of agents
• Working directory
• Model selection
• Budget limits
• Feedback mode (supervised/autonomous)

Agent Roles

Maestro includes specialized agent roles:

• Orchestrator - Coordinates the overall workflow
• Project Manager - Breaks down work and manages tasks
• Architect - Designs system architecture
• Developer - Writes implementation code
• Designer - Creates UI/UX designs
• QA Engineer - Tests and validates work
• DevOps - Handles deployment and infrastructure
• Technical Writer - Creates documentation
• Code Reviewer - Reviews and validates changes

Process Safety

By default, agents follow process safety rules to keep your machine stable:

Agents will:

• Write production code, tests, configs, and docs
• Run lightweight commands (linters, type-checks on single files)
• Run targeted test files
• Install individual packages

Agents will not:

• Run builds (npm run build, cargo build, etc.)
• Run Docker commands
• Start dev servers
• Run full dependency installs
• Run full test suites

You can customize these rules by editing the .agents/*.md files.

License

MIT