Windows: agents silently and permanently archived after abnormal exit 1073807364 (0.5.7) #833

Version: Munder Difflin 0.5.7 (packaged), Windows 11 Pro 10.0.26200, provider claude (claude.exe via WinGet).

Symptom

Agents disappear from the roster without user action. They are archived automatically and stay archived across restarts (a message to one returns "has no live terminal (archived)"). Their folders and memory.md are intact, but there is no visible way to restore them.

Evidence (hive/log.jsonl + hive/crashes/)

4 agent-exit events in the whole log, all abnormal, all exitCode 1073807364 (0x4000013A, Windows console-close/forced termination), each immediately followed by an "archive" event (~5 ms later):

• All happen at night, with no app event for 5-18 min before, and the app is restarted many hours later.
• Crash logs show the claude TUI idle or mid-redraw, no error text.
• The 35 app-starts also archive other agents in batches, but those are restored on the next start; the 4 above never come back.

Hypothesis (unverified)

The PC was shut down / put to sleep / the console host was killed without quitting Munder Difflin first. The pty child gets exit 0x4000013A, the app records it as an abnormal exit and archives the agent permanently, instead of treating it as "interrupted, restore on next start".
User suspects they shut the PC down without closing the app; times (night) are consistent but not proven. Not every agent logged an exit, so the kill seems partial.

Related open issues

• 0.5.7: SIGABRT from node-pty exit-waiter ThreadSafeFunction teardown (quit-time throw and uv_cond_wait abort) #806 (node-pty SIGABRT at quit, 0.5.7, macOS): similar area, different OS/code.
• OpenCode agent fails at startup: restore passes a Claude session id to opencode --session #765 (restore passes wrong session id), Temp-worker task dispatch failure is archived as done and leaves the worker live #663 (failed dispatch archived as done).

Ask

1. Do not auto-archive on exit 0x4000013A / when the app itself is shutting down; restore those agents on next start.
2. Add a "restore archived agent" action in the UI.
3. Document what triggers auto-archive.

Crash log headers (no terminal output attached)

agent: atlas-mvio09k3      exitCode: 1073807364  captured: 2026-10-26T01:28:44.885Z
agent: creed-mviqkh39      exitCode: 1073807364  captured: 2026-10-31T04:22:46.275Z
agent: recruteur-mviua0p8  exitCode: 1073807364  captured: 2026-11-07T06:03:12.187Z
agent: magigos-mw5nges5    exitCode: 1073807364  captured: 2026-11-07T06:03:12.199Z
command: claude.exe (WinGet) for all four

In every log the agent was mid-turn (thinking / running Stop hook) with no error text; a Claude Code update notice was pending.