codex-autopilot

What This Skill Does

Codex Autopilot is a sophisticated automation framework designed to manage multiple concurrent Codex CLI sessions. By leveraging tmux for session multiplexing and macOS launchd for persistent background execution, the skill transforms static terminal windows into a dynamic, watchdog-driven development environment. It constantly monitors your active AI coding sessions, automatically intervening to keep them productive. Whether a session is idly waiting for input, stuck on a permission prompt, or ready for a new task from the internal queue, the Autopilot handles the interaction without manual intervention. It includes robust logging, notification integration for platforms like Discord and Telegram, and intelligent state management to ensure that your coding agent remains active and focused throughout your development lifecycle.

Installation

To install, first ensure you have the required dependencies: macOS, tmux, Python 3, yq, and jq installed via Homebrew. Begin by cloning the repository to your local machine using git clone https://github.com/imwyvern/AIWorkFlowSkill.git ~/.autopilot. Navigate to the directory, copy the configuration template with cp config.yaml.example config.yaml, and populate it with your specific project paths, API tokens, and notification webhooks. Finally, run the provided ./install.sh script to register the persistent launchd agent, which will ensure your watchdog loop remains active across system restarts.

Use Cases

• Managing large-scale software projects that require simultaneous AI-assisted development across multiple codebases.
• Automating the execution of long-running refactoring or testing tasks that require periodic input or verification.
• Orchestrating complex workflows where the output of one coding session informs the initialization of another.
• Ensuring zero-downtime development by keeping AI agents 'awake' during long-duration coding cycles.

Example Prompts

1. "Autopilot, initialize the current tmux session with the project in ~/dev/backend-api and add it to the active watchdog monitoring list."
2. "Watchdog, check the status of all current Codex sessions; if the 'frontend-ui' project is idle, inject a prompt to refactor the navigation bar."
3. "Codex automation, enqueue a new task to generate unit tests for the authentication module in the primary project and notify me on Discord once complete."

Tips & Limitations

For optimal performance, regularly audit your project configuration files to prevent session bloat. Ensure that your terminal environment variables are correctly inherited by the launchd process to avoid permission errors. Note that while this skill is highly effective for multitasking, it is strictly designed for macOS environments using tmux. Exercise caution when granting automatic permission approvals for security-sensitive operations; always review the logs generated by the watchdog service to ensure the agent is operating within your intended parameters.