kyler505/codex-axi

codex-axi gives one AI agent a compact command line tool for controlling OpenAI's Codex from the outside. It lets an agent start a piece of work in Codex, keep a stable reference to that thread so it can check on it later, watch its progress, redirect it while it is still running, and resume it after a pause. It builds on top of the official openai-codex Python SDK and the managed Codex app server, rather than replacing Codex's own safety checks or how it actually stores threads. The project is currently in an alpha stage and targets a specific narrow range of Codex versions, so running the built in doctor command after installing is recommended to check that everything lines up correctly on your system. Installation needs Python 3.10 or newer and an already authenticated Codex command line tool, and it works on macOS or Linux for its preferred way of running. Once installed, an agent can start a foreground task that waits for a result, or start a background worker that returns immediately so the caller can check on it, send it new instructions, interrupt it, or close it later. There are also commands for viewing live events from a running turn, inspecting real parent to child Codex agent relationships, and installing optional hooks that show a quick workspace summary at the start of a session in Claude Code, Codex, or OpenCode. All output uses a compact text format by default, though a full JSON format is available for tools that need it, and errors come back in a structured way so a calling agent can understand and react to them without extra parsing work.