popbones/tmux.skill

This repository, tmux.skill by popbones, packages a small skill for OpenAI's Codex coding agent that teaches it how to use tmux as a place to run long-running commands. tmux is a Unix tool that keeps terminal sessions alive in named windows and panes, so the idea here is to give an AI agent a way to start a dev server, a file watcher, a build, a log tail, or any interactive command, and then come back to it later instead of running everything in one short-lived shell. The repo is intentionally tiny. It contains three pieces: SKILL.md, which holds the instructions Codex itself reads to learn the workflow; a shell helper at scripts/tmux-send-and-capture, which sends text into a chosen tmux pane and reads back the most recent lines of output; and agents/openai.yaml, which carries the metadata Codex uses to show the skill in its interface. To install it, the README says to clone or copy the folder into the Codex skills directory, which lives at $CODEX_HOME/skills/tmux, or under ~/.codex/skills/tmux if that variable is not set. After a fresh Codex session, the new skill is picked up. The helper script takes a tmux target in the usual session:window.pane format, an optional number of lines to capture, and the text to send. One example sends the string status please to a coordination pane and grabs the last 80 lines. A second example uses a --codex-queue flag to push text into a pane where Codex is already busy, so the message waits until the current task finishes before being read. The README does not describe a license, tests, or any wider configuration. It is meant as a building block for agent workflows that need durable shells, and it expects the user to already have tmux installed and a working Codex setup.