Tmux-Native Coding Agent

Use this skill when the user is reasoning about terminal-native AI coding workflows: Claude Code interactive mode, Codex, opencode, aider, tmux, panes, windows, sessions, capture-pane, long-running dev servers, logs, tests, or multiple agents coordinating through a terminal workspace.

Core Philosophy

The user is not merely asking for an AI chatbox inside a terminal. They are describing a durable development cockpit:

tmux + shell + editor + logs + dev servers + tests + coding agents

The important shift is that tmux becomes a persistent, observable workspace. A coding agent is not limited to the current prompt if it can inspect the surrounding terminal environment. It can see the running frontend server, backend logs, test output, migration failures, another agent session, or a scratch shell, as long as it has shell access and permission to inspect the relevant tmux pane.

This mindset treats the terminal as a context bus. Panes are live data sources. Windows are task lanes. Sessions are durable workspaces. The agent is a participant in that workspace, not a detached chatbot.

Mental Model

Use these distinctions when explaining the concept: