cmux

The cmux skill provides deep integration for controlling the cmux terminal environment, a Ghostty-based macOS terminal designed for AI coding agents. This tool allows automated workflows to manage complex UI states, such as window organization, workspace management, and precise pane routing. By offering a programmatic interface for terminal layouts, it enables developers and agents to maintain deterministic working environments across sessions, ensuring that context-heavy tasks remain organized and accessible.

• Perform full topology discovery including windows, workspaces, panes, and surfaces with JSON-formatted output for easy parsing.

• Execute dynamic layout commands such as creating workspaces, splitting panes (horizontal/vertical), and moving surfaces between containers.

• Focus routing and surface management to ensure that specific tasks or agent sessions appear in the correct visual location.

• Trigger visual attention cues, such as flashes on specific surfaces, to alert human users during long-running background tasks.

• Manage local application settings, including browser integration, sidebar behavior, and workspace configurations via CLI commands.

• Seamlessly integrate with external tools like Claude Code Teams, SSH remote sessions, and markdown viewer panels.

• The skill uses short handles like window:N or surface:N for direct targeting, though UUIDs are supported for precision in complex environments.

• Before editing configuration files, always use the built-in documentation commands (cmux docs settings) to verify schema and file paths.

• Prefer using cmux list- commands for state inspection prior to performing destructive move or reorder actions.

• Note that terminal-specific rendering properties like fonts, cursor styles, and themes remain under the control of Ghostty configuration files, while cmux manages the organizational topology.

• Use the identify command to clarify the caller's current context within the terminal environment.

• Designed for environments where CLI-driven window management outperforms manual GUI interaction, particularly in high-parallelism AI agent workflows.