agent-launcher 3.0.1

Bilingual launcher for AI coding agents (Codex & Claude) with review and loop modes

Agent Launcher 🤖

Bilingual (English/Français) launcher for AI coding agents (Codex & Claude Code).

A powerful, user-friendly command-line tool that simplifies launching and managing AI coding agents with advanced features like review mode, loop mode, and configuration persistence.

✨ Features

• 🎯 Guided Selection: Interactive prompts to choose between Codex and Claude
• 🔒 Safety Controls: Configurable sandbox and approval bypass settings
• 🔄 Review Mode: Automatic cross-agent validation and debugging
• 🔁 Loop Mode: Iterative improvements with time-based iterations
• 💾 Persistent Configuration: Save your preferences for future use
• 🌍 Bilingual: Full English and French support
• 📝 Command History: Track all your agent invocations
• 🎨 Smart Terminal Detection: Automatic color support based on your terminal
• 🔧 Environment Variables: Configure via environment for automation
• 📦 Easy Installation: Self-installing with --install flag

📦 Installation

Option 1: pip install (Recommended)

pip install agent-launcher

Option 2: Direct Download

# Download the script
curl -o agent.py https://raw.githubusercontent.com/Actigence-Code/agent-launcher/main/agent.py

# Make it executable
chmod +x agent.py

# Install globally
./agent.py --install

Option 3: From Source

git clone https://github.com/Actigence-Code/agent-launcher.git
cd agent-launcher
pip install -e .

🚀 Quick Start

Basic Usage

# Interactive mode - will prompt for choices
agent

# Specify agent directly
agent --agent claude

# With bypass mode (be careful!)
agent --agent codex --bypass

# Dry run to see what would be executed
agent --dry-run --agent claude

Advanced Features

# Review mode: Second agent reviews the first agent's work
agent --agent claude --review

# Loop mode: Iterate improvements for 30 minutes
agent --agent claude --loop 30

# Combined: Loop with review
agent --agent claude --loop 60 --review

# Specify model and additional directories
agent --agent claude --model sonnet --add-dir /path/to/project

# Save your preferences as defaults
agent --agent claude --bypass --save-defaults

Environment Variables

# Configure via environment
export AGENT_TYPE=claude
export AGENT_BYPASS=true
export AGENT_MODEL=opus

# Now just run
agent

📖 Usage Examples

Example 1: Quick Interactive Session

$ agent
Agent ? (1) Codex | (2) Claude - tapez le numéro ou le nom: 2
Dangerously bypass approvals & sandbox ? [y/N | o/N] : n

Agent Ready / Agent Prêt
- Agent: Claude | Agent choisi : Claude
- Bypass: DISABLED | Bypass : DÉSACTIVÉ
WARNING: Read the official docs before enabling bypass.
Bon travail / Happy coding!

Launching Claude...

Example 2: Code Review Workflow

# Have Claude write the code, then Codex reviews it
agent --agent claude --review "Implement a binary search tree in Python"

# Or vice versa
agent --agent codex --review "Create a REST API with authentication"

Example 3: Iterative Improvement

# Work on optimizing code for 45 minutes
agent --agent claude --loop 45 --bypass "Optimize the database queries in src/models"

Example 4: Non-Interactive Automation

# Use in scripts or CI/CD
AGENT_TYPE=claude AGENT_BYPASS=false agent --quiet "Run tests and fix any failures"

🎛️ Command-Line Options

Agent Selection

• -a, --agent {codex,claude}: Choose the agent (Codex or Claude)
• --check-version: Check installed agent version

Safety Options

• --bypass: Enable dangerous sandbox/approval bypass ⚠️
• --no-bypass: Explicitly disable bypass
• -q, --quiet: Suppress styling (for scripts)

Advanced Modes

• --review: Enable review mode (other agent reviews output)
• --loop MINUTES: Enable loop mode for N minutes (default: 60)
• -m, --model MODEL: Specify model to use
• --add-dir DIR: Add additional working directory (repeatable)

Configuration

• --save-defaults: Save current options as defaults
• --dry-run: Show command without executing
• --install: Install launcher globally

Info Commands

• -V, --version: Show version information
• -h, --help: Show help message

🔧 Configuration Files

Config Location

• Linux/macOS: ~/.config/agent/config.json
• Configuration is automatically created on first use

Config Structure

{
  "default_agent": "claude",
  "default_bypass": false,
  "default_model": "opus"
}

Command History

All commands are logged to ~/.config/agent/history.log:

{
  "timestamp": "2025-11-23T12:34:56",
  "agent": "claude",
  "bypass": false,
  "model": "sonnet",
  "command": "claude --model sonnet 'Fix the bug'",
  "review_mode": false,
  "loop_duration": null
}

🎨 Review Mode

Review mode adds a quality assurance step by having the alternate agent review and improve the output:

1. Primary Agent Executes: Your chosen agent completes the task
2. Review Prompt: You're asked if you want to review
3. Secondary Agent Reviews: The other agent checks for bugs and improvements
4. Interactive: You can skip review or cancel at any time

# Example workflow
agent --agent claude --review "Implement user authentication"

# Output:
# [REVIEW MODE] Running primary agent: Claude
# [REVIEW MODE] The agent will execute your request first...
# (Claude implements authentication)
# [REVIEW MODE] Primary agent completed with exit code 0
# [REVIEW MODE] Launch review agent (Codex)? [Y/n | O/n] : y
# (Codex reviews and suggests improvements)

🔁 Loop Mode

Loop mode enables iterative improvements over a specified time period:

• Runs multiple iterations until time expires
• Optional review after each iteration
• Minimum 5 minutes required for next iteration
• User can cancel at any time

# Optimize code for 30 minutes
agent --agent claude --loop 30 "Optimize performance in src/api"

# With review after each iteration
agent --agent claude --loop 60 --review "Refactor legacy codebase"

🔐 Safety Notes

⚠️ Bypass Mode Warning

The --bypass flag disables safety sandboxes and approval prompts. Use with extreme caution:

• Codex: --dangerously-bypass-approvals-and-sandbox
• Claude: --dangerously-skip-permissions

Only use bypass mode when:

• You fully trust the code being executed
• You're working in a sandboxed/disposable environment
• You understand the security implications

🌍 Bilingual Support

All prompts and messages are displayed in both English and French:

Agent ? (1) Codex | (2) Claude - tapez le numéro ou le nom:
Dangerously bypass approvals & sandbox ? [y/N | o/N] :
Agent Ready / Agent Prêt
Bon travail / Happy coding!

🐛 Troubleshooting

Agent Not Found

# Check if agent is installed
agent --check-version --agent claude

# Install missing agent (requires npm)
# The launcher will offer to install if npm is available

Permission Errors

# For global installation
sudo agent --install

# Or install to user directory
AGENT_INSTALL_DIR=~/.local/bin agent --install

Clear Configuration

# Remove config to start fresh
rm -rf ~/.config/agent/

📚 Requirements

• Python: 3.8 or higher
• Operating System: Linux, macOS, or Windows (with WSL)
• Dependencies: None (pure Python, stdlib only)
• Optional: Node.js + npm (for agent installation)

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

Development Setup

git clone https://github.com/Actigence-Code/agent-launcher.git
cd agent-launcher
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -e ".[dev]"

Running Tests

pytest

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• OpenAI Codex for the amazing AI coding assistant
• Anthropic Claude Code for powerful code understanding
• All contributors who help improve this tool

📞 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: Wiki
• Email: [e-mail]
• Website: actigence.eu

🗺️ Roadmap

• Support for additional AI coding agents
• Plugin system for custom workflows
• Web UI for configuration
• Team collaboration features
• Cloud sync for configuration
• Integration with popular IDEs

---

Made with ❤️ in Tervuren, Belgium

Actigence Management Consulting | [e-mail]