Your whole fleet, at a glance.

The Wall adds no new tracking — it aggregates stores that are already live, each card subscribed to its own source and refreshing as things change. Five of them tile a single grid built for a wide screen.

• Agents — every agent across all your workspaces; the ones needing attention bubble to the top, each with its time-in-state.
• Task Board — pending, in-progress, and done counts, urgent flags, and live Swarm progress when a run is active.
• Repo Health — worktrees ahead of and behind main, dirty and orphaned trees, and disk used.
• Live Activity — a rolling ticker of every agent tool call, newest first.
• Tokens · fleet — token throughput per agent, summed across the fleet.

This is a read-only surface — nothing on it is a control. It’s tuned to sit on a second monitor: one shared clock drives every “time ago” readout, and a single poll keeps the worktree health fresh for the whole wall, so there are no per-card timers fighting for the main thread.

• TV / full-screen mode covers the entire app shell; Esc drops back out.
• Type scales up in TV mode so the wall reads from across the room.
• One 1-second clock and one health poll feed every card.
• Docked narrow? The grid stacks into a single scrolling column.

The ticker prepends each agent tool call as it happens — tagged with the agent, the tool, and a one-line summary. It is capped at the last 100 events, so a wall left running for hours never grows unbounded, and the newest-row motion honours reduced-motion preferences.

Token throughput sits below it, summed across every live agent. Backends that report usage (like Claude Code) fill in exact counts; where a backend doesn’t, the agent shows a dash rather than a misleading zero.