Fleet list — TASK-first columns, filter, scrolling

The first/primary column is TASK: a one-line summary of what each run is actually doing (extracted from the first step's prompt — the TASK: block for implement-task, or the first prompt line for other workflows). WORKFLOW is demoted to a dim type tag beside it. This is the fix for "I can't tell which run is which": two parallel implement-task runs read by their distinct task, not by an identical workflow name. When the fleet is taller than the terminal, the list windows to fit (the footer never gets pushed off-screen) and shows a ↑ N above … M below position line; the window edge-scrolls to keep the selected row visible. A row briefly flags with a ▶ gutter when its state changes, so a run finishing mid-list catches the eye.

Press / to filter: type a query and the list narrows in real time to runs whose task or workflow name matches (case-insensitive substring); Enter keeps the filter and returns to navigation, Esc clears it. A / <query> · N shown bar appears above the columns while filtering.

A preview card is pinned below the list and always reflects the row under the cursor: it shows the FULL task (the row cell truncates it; here it wraps) plus the run's facts — workflow, short run id, state, steps, age, tokens, cost — and the live detail line. Move the cursor (j/k) and the card updates instantly, so you can read what each run is doing without inspecting it.