This is a fork of the OpenAI Codex CLI. For details about the original Codex CLI, its features, and usage, please refer to the main repository.

This fork introduces enhancements on the badpirate branch, focusing on "Unlocking Container Agent" capabilities. The goal is to enable a truly Full Auto agent that can:

• Run Docker containers for setting up and utilizing end-to-end tests.
• Install packages from npm without being blocked by a firewall.
• Provide a more flexible and powerful sandboxing mechanism for agent autonomy.

The run_in_container.sh script has been enhanced with additional flags to provide more flexibility and control over containerized workflows. Below are the new flags and their purposes:

• --dangerously-allow-network-outbound

  • Allows unrestricted outbound network traffic inside the container.
  • Use with caution as it bypasses the firewall restrictions.

• --dangerously-allow-sudo

  • Enables sudo execution inside the container via the sudonode user.
  • Useful for workflows requiring elevated privileges.

• --dangerously-allow-install

  • Combines the effects of --dangerously-allow-network-outbound and --dangerously-allow-sudo.
  • Also enables Docker-in-Docker (nested Docker daemon).

• --allow-docker-in-docker

  • Enables Docker-in-Docker by running the container in privileged mode and starting a nested Docker daemon.

• --notify

  • A flag for codex rather than run_in_container.sh, already exists on main, but on my build it will do two "beeps" when it requires your attention and on a non-mac machine. Because container is always on linux, this gives you a heads up that your agent is done. About as flaky as the main branch version but what are you gonna do?

1. Run a simple command inside the container:

   ./run_in_container.sh "echo Hello, world!"

2. Allow unrestricted outbound network traffic and sudo execution:

   ./run_in_container.sh --dangerously-allow-install --work_dir /path/to/project "npm install"

3. Enable Docker-in-Docker and list running containers:

   ./run_in_container.sh --allow-docker-in-docker "docker ps"

1. Clone this repository and switch to the badpirate branch:

   git clone https://github.com/badpirate/codex.git
   cd codex
   git checkout badpirate

2. Build the container:

   ./codex-cli/scripts/build_container.sh

3. Run the container with the desired flags:

   ./codex-cli/scripts/run_in_container.sh --dangerously-allow-install --allow-docker-in-docker

This fork is ideal for developers who need:

• A more autonomous agent capable of running complex workflows, including end-to-end tests.
• The ability to install dependencies and execute commands without being blocked by restrictive firewalls.
• Enhanced security and sandboxing for running commands in a controlled environment.

For any questions or contributions, feel free to open an issue or pull request. Happy hacking! 🚀