## What happened

The idle screensaver (`omarchy-launch-screensaver` / `omarchy-screensaver`) intermittently opens and then closes itself again about 1 second later, before it's ever perceptible on screen. Because the shell's idle service (`Service.qml`) treats the screensaver window closing as "the user dismissed it," it also cancels the *entire* pending idle cycle — including the separately-timed idle lock. So on affected cycles, neither the screensaver nor the idle lock actually happens.

Confirmed via `journalctl --user` (idle service debug logging), 3 real idle cycles logged in one session, `idle.screensaver=150`, `idle.lock=300`:

```
# Cycle 1 (worked correctly — screensaver stayed up ~2m10s until real dismissal)
idle-monitor: idle
idle-cycle-start: screensaver=150 lock=300
process-start: screensaver [...] || omarchy-launch-screensaver
idle-monitor: active
idle-monitor-active: screensaver cycle remains armed
process-exit: screensaver exitCode=0 status=0
... (~2m10s later, real user input) ...
idle-cycle-cancel: screensaver-dismissed
process-start: wake omarchy-system-wake

# Cycle 2 (bug — self-exits ~1.1s after launch, no real input)
idle-monitor: idle
idle-cycle-start: screensaver=150 lock=300
process-start: screensaver [...]
idle-monitor: active
idle-monitor-active: screensaver cycle remains armed
process-exit: screensaver exitCode=0 status=0
idle-cycle-cancel: screensaver-dismissed      # only ~1.1s after process-exit
process-start: wake omarchy-system-wake

# Cycle 3 (same bug, same ~1.1-1.2s timing)
```

## Root cause (traced)

`omarchy-screensaver`'s inner watchdog loop:

```bash
while pgrep -t "${tty#/dev/}" -x ttfx >/dev/null; do
  if read -n1 -t 1 || ! screensaver_in_focus; then
    exit_screensaver
  fi
done
```

`screensaver_in_focus()` checks `hyprctl activewindow -j | jq -e '.class == "org.omarchy.screensaver"'` once per second. On a race (window mapped but Hyprland hasn't settled focus onto it yet, or `follow_mouse=1` focus-follow timing), this check fails on its first poll and the script self-terminates via `exit_screensaver`, closing the window almost immediately after it opened.

The shell's idle service (`omarchy-shell/plugins/services/idle/Service.qml`, `handleScreensaverWindowClosed`) then correctly-by-design treats any screensaver window close during an active idle cycle as user dismissal and calls `cancelIdleCycle("screensaver-dismissed")`, which also stops the still-pending `lockTimer`. So the bug doesn't just hide the screensaver — it silently defeats idle-lock too.

## Expected

The screensaver should reliably stay up until real user input (keyboard/mouse) is received, and a spurious self-exit from the screensaver script should not be indistinguishable from a genuine user dismissal for lock-cancellation purposes.

## Environment

- Omarchy version: 4.0.4-1
- Kernel: 7.2.5-3-omarchy
- CPU: AMD Ryzen 5 Microsoft Surface (R) Edition
- GPU: AMD Picasso/Raven 2 (Radeon Vega, integrated)
- Machine: Microsoft Surface Laptop 3 (AMD 15" edition)
- Single monitor (eDP-1), `follow_mouse = 1` (Hyprland default)
- `ttfx` 0.3.2

## Workaround in use

Disabled the screensaver locally via `omarchy toggle screensaver off`, which removes the racy window-close path entirely and lets the idle lock fire reliably again.