Fleet

Run six agents at once and the canvas stops answering the question that actually matters: which of them needs you right now. One tile is waiting on a permission prompt three screens to the left, another finished ten minutes ago, a third died. Scanning tile-by-tile does not scale past two or three agents.

Mission Control — the fleet panel — answers that question directly. It groups every agent tile by state, puts the ones waiting on you first, and shows enough of their output to act on without leaving the panel.

What you can do with it

• See every agent tile grouped by state — Needs input, Working, Done, Failed, Idle, Exited — always in that order, with empty groups skipped.
• Read the pending prompt without visiting the tile — each row in Needs input shows the last lines of that tile's output, almost always the actual permission question, so you can decide whether it is safe to handle before switching tiles.
• Click a row to fly to the tile — the fastest path from "agent X needs me" to actually looking at agent X's terminal, with no visual hunting across the canvas.
• Glance at the collapsed badge — when the panel is closed, the pill shows the count of tiles waiting on input (amber) so you know when to interrupt, or the count of all non-exited agents (muted) when nothing is waiting.
• Watch Dynamic Workflows progress per row — when a tile is running a Claude Code Dynamic Workflow, the panel shows a workflow · 3/5 agents line under the row, so you can see fan-out progress without flying to the tile.

How to use it

1. Mission Control lives as a collapsible panel pinned to the right edge of the screen. Collapsed, it is a small pill showing a glyph and a count. Expanded, it is a scrollable list grouped by state.
2. Click the pill to expand the panel. Click again to collapse it.
3. Read the Needs input group first — those tiles are waiting on you.
4. Click any row to pan and zoom the canvas directly to that tile.
5. Handle the prompt in the tile, then come back to the panel for the next one.

The six groups, in display order:

If there are no agent tiles on the canvas, the panel does not render at all — nothing to triage, nothing to show.

Common use cases

• "Which agent is blocked on me?" — open the panel, read the Needs input group, click the top row to handle it.
• "Did the refactor agent finish?" — scan the Done group instead of hunting for the tile.
• "Why did that worker die?" — the Failed and Exited groups show those tiles; click to fly over and read the output.

Tips and best practices

• Keep the panel collapsed when you are heads-down; the amber badge will tell you when something needs you.
• Read the prompt tail under a Needs input row before clicking — it tells you whether the question is safe to auto-handle or needs a careful look.
• The panel's expanded/collapsed state resets to collapsed on the next launch. It does not persist across restarts.
• Mission Control and the right drawer share the right edge — opening one collapses the other.

Related features

• Right Drawer — the other right-edge panel; the two coordinate so only one is open at a time.
• Context Gauge — see how full a tile's context window is at a glance.
• Dynamic Workflows — the Claude Code feature whose per-row progress the panel surfaces.

Limitations

• Only terminal and team-pane tiles that MadoHub has classified as agents appear in the panel — a bare shell you just opened may take a moment to show up.
• The panel's expanded/collapsed state is in-memory only and resets to collapsed on the next launch.
• The panel and the right drawer cannot be open at the same time.