What problem does it solve? Coordinating multiple AI coding agents on one machine normally means digging through terminal history and guessing whether a reply is complete. This Skill lets agents send work to named tmux panes, receive durable stored replies, and recover outstanding requests after timeouts, detachments, or pane loss. ## Core Features & Use Cases - Durable agent messaging: Send requests with tmt talk and collect complete replies via tmt reply and tmt result, correlated by request ID and receipt rather than terminal markers. - Identity and room management: Create global identities, bind them to tmux panes, group them into rooms, and queue inbox requests for offline agents. - Attention recovery: Use tmt x commands to list, inspect, and acknowledge outstanding requests after interruptions or restarts. - Use Case: A coordinator agent asks a reviewer agent in another pane to audit a patch, detaches immediately, and later retrieves the full review with tmt result <request-id>. ## Quick Start Ask the agent to send a review request to the tmux pane named reviewer and wait for its complete reply.