[BUG] Cowork jobs stuck spinning indefinitely — persists through full quit and uninstall/reinstall; backend confirms isRunning: false and TaskList empty

Environment

• OS: Windows 11 Home 10.0.26200
• - Claude Code: Desktop app (Windows)
• - - Feature: Cowork multi-agent panel

Description

Cowork jobs show a perpetual spinning/in-progress state in the UI. The spinner never resolves. This is not a transient display glitch — it has persisted through all of the following recovery attempts:

1. Fully quitting Claude via the taskbar icon (not just closing the window)
2. 2. Quitting via the system tray / menu icon
3. 3. Uninstalling and reinstalling the Claude Code desktop app entirely

Backend State Confirms No Live Processes

Using mcp__ccd_session_mgmt__list_sessions, every session returns "isRunning": false. Using TaskList, the result is "No tasks found". Using TaskStop with the session IDs returns "No task found with ID: <session_id>". The backend is completely idle — the spinning is purely a frontend UI state problem.

Related Issues

This appears to be part of the same cluster as:

• #26805 — "Working through complex response" hangs indefinitely
• - #59532 — User messages remain pulsating after session-switch
• - - #45531 — Cowork stuck on "Starting session" for days
• - - - #62897 — UI incorrectly shows Claude working

The key distinction: this bug survives a full uninstall/reinstall cycle, which suggests stale state is being persisted to disk or rehydrated from a server-side source rather than being purely in-memory UI state.

Steps to Reproduce

1. Run one or more Cowork jobs
2. 2. Jobs fail to complete or are interrupted
3. 3. Cowork panel shows spinning indicators on those jobs
4. 4. Quit Claude completely (taskbar + system tray)
5. 5. Relaunch Claude — spinners still present
6. 6. Uninstall and reinstall Claude Code desktop app
7. 7. Relaunch — spinners still present
8. 8. Query backend via session management MCP: isRunning: false for all sessions, TaskList returns empty

Expected Behavior

Cowork job spinners should resolve when the backend confirms no active processes, and should not survive an app reinstall.

Actual Behavior

Spinners persist indefinitely regardless of backend state, through full app restart and reinstall. Backend tooling (TaskStop) also cannot kill the jobs because they do not exist as live tasks.

Impact

High — the stuck UI state cannot be cleared by the user through any self-service means.