## What happened

After ~3 weeks of uptime, every application launch (terminals via `omarchy-launch-terminal`, and anything else going through `uwsm-app`) failed with a cryptic `Invalid arguments: ` error toast.

Root cause: `/run/user/1000` (752M tmpfs) was at **100%**. The systemd user manager could no longer create transient scope units, so every app launch failed. Even plain `systemd-run --user --scope` failed with `Failed to start transient scope unit: Unit ... not found`.

The space was consumed by `/run/user/1000/quickshell/by-id/`: the Omarchy shell (quickshell) creates a log directory per shell instance and never prunes old ones. On this system there were **77 stale instance dirs** dating back weeks, including a single **463MB `log.log`** and an 18MB `log.qslog` in one dead instance. Total quickshell usage: ~704M of the 752M tmpfs.

Deleting the stale `by-id` dirs (keeping the live instance) immediately restored app launching.

## Expected

Old quickshell instance log dirs under `/run/user/1000/quickshell/by-id/` (and their `by-pid` symlinks) should be pruned automatically — e.g. on shell startup, or capped/rotated so a few MB of logs can't fill the runtime tmpfs and break the session.

## Steps to reproduce

1. Use Omarchy normally for a few weeks; restart the shell occasionally (`omarchy restart shell`, crashes, etc.). Each instance leaves a log dir in `/run/user/1000/quickshell/by-id/`.
2. Observe `du -sh /run/user/1000/quickshell` growing without bound and `df -h /run/user/1000` approaching 100%.
3. Once full, every app launch (uwsm-app / xdg-terminal-exec path) fails with `Invalid arguments: `, and the journal shows `xkbcomp ... Cannot close "/run/user/1000/server-0.xkm" properly (not enough space?)`.

## Cleanup that fixed it

```bash
# keep only the live instance dir (resolve via /run/user/1000/quickshell/by-pid/<live-pid>)
cd /run/user/1000/quickshell/by-id
rm -rf <all-dirs-except-live>
rm -f /run/user/1000/quickshell/by-pid/*
```

## System details

- Omarchy: 4.0.2-1
- Kernel: 7.1.9-arch1-2 (Arch, current as of 2026-09-25)
- Quickshell: 0.3.1 (Arch package)
- CPU: Intel Core i5-10210U; GPU: Intel UHD Graphics (CometLake-U)
- Session: Hyprland via uwsm (sddm autologin), user manager up since Sep 5

## Notes

- The live shell instance's own log had already reached 62MB (`log.log`) within a few days, so growth is not just from stale instances — some log rotation/cap would help there too.
- Symptom is very hard to connect to the cause from the user side ("Invalid arguments" gives no hint); a warning notification when /run/user usage crosses a threshold would also help.