[FEATURE] Show weekly and all-time usage in FleetView

Summary

FleetView (claude agents) shows what each session is doing but nothing about what the fleet is costing. When several background sessions are running, the burn rate is exactly the thing you want on screen — and it is the one number the view does not have.

Grepping the 2.1.263 bundle, no FleetView identifier carries cost or token data: the surface's fields are fleetViewGroupMode, fleetViewPeakConcurrent (telemetry only), fleetNudgeStore, fleetHostCall, fleetBackground, fleetFork. There is no per-row or aggregate usage field.

The data already exists — in another view

/usage already computes everything needed, across two separate pickers:

| Picker | Ranges | Backing |
|---|---|---|
| Limits | Day (d), Week (w) | settings:periodDay / settings:periodWeek |
| Stats | All time, Last 7 days, Last 30 days | stats-cache.json |

/usage also already attributes usage by subagents, MCP servers, and plugins ("What this plugin's skill descriptions add to the system prompt", "What's contributing to your limits usage?"), and knows the limit taxonomy: five_hour → session limit, seven_day → weekly limit, seven_day_opus, seven_day_sonnet, seven_day_overage_included.

Request

Surface weekly and all-time totals in the FleetView header — at minimum weekly-limit consumed vs. remaining, plus an all-time total from stats-cache.json. Per-session attribution in each row would be better still, since FleetView is where you decide which background session to stop, and that decision is currently made blind.

The coaching strings already in the bundle make the case: "These are often background/loop sessions. Continuous usage can add up quickly so make sure it is intentional." FleetView is the view where those sessions are managed, so it is where that warning should be actionable.

Environment

• Claude Code 2.1.263, macOS 25.6.0 (darwin), desktop app