Windows desktop: app crash mid-session kills background shell tasks with no completion record; preceded by MCP server disconnect/reconnect churn

Environment

• Claude Code 2.1.222, Windows desktop app
• Windows 11 Pro 10.0.26200
• Long-running session (several hours, 500+ assistant turns), ~25 MCP servers configured, several background Bash tasks (watch loops) running

What happened

1. Throughout the session, ambient notices showed repeated waves of MCP servers disconnecting and reconnecting (hundreds of deferred tools flapping unavailable/available in a single session, multiple times).
2. The app then crashed/exited entirely mid-session (twice today with the same signature - one earlier restart, one hard crash).
3. On resume, each background shell task reports: "No completion record was found for this background shell command from the previous session. It may have been stopped ... or it may have been running when the previous Claude Code process exited."
4. All background watch loops die silently with the process; the resumed session has to notice and re-arm them by hand. A file-based watcher that dies without a marker is indistinguishable from a quiet wait, so work signals can be missed.

Expected

• The app not to crash; failing that, a crash/exit marker for background tasks so a resumed session can distinguish "stopped by crash" from "still waiting" without guessing, and ideally automatic re-arm or a resume-time list of tasks that were alive at exit.

Notes

• The MCP churn and the crash may or may not be related; the churn preceded both exits.
• Happy to pull local logs if you tell me which files to look at on Windows.

🤖 Generated with Claude Code