This entry is built around the tmux skill in the openclaw/skills repository. The agent behavior is precise: inspect existing tmux sessions, capture recent pane output, send text or control keys, and steer interactive terminal applications such as Claude Code, Codex, or other long-running TUIs. That makes the unit of value a concrete operator workflow, not a catalog card for tmux itself.

Use this when a user needs an agent to monitor or intervene in an already-running interactive CLI session. It is the right invocation when a plain shell command is not enough because the target process expects prompts, confirmations, navigation, or iterative input over time. In practice that means checking whether a coding agent is waiting for approval, reading the tail of a pane for status, or sending follow-up instructions without attaching manually.

The scope boundary keeps this from collapsing into a generic tmux or CLI listing. It is not a terminal multiplexer review, not a general process manager, and not a replacement for normal shell execution. The skill is specifically about controlling live tmux-backed sessions where interaction and screen scraping are the job. One-off commands should still use direct execution tools, and starting brand-new background processes belongs elsewhere.

Integration points include tmux session names, pane targeting, capture-pane output, send-keys flows, and agent orchestration patterns that rely on persistent interactive sessions. It fits naturally into runbooks, coding-agent supervision, and diagnostics workflows.