## What happened

Opening the audio panel crashed the Omarchy shell (SIGSEGV). Quickshell's crash handler restarted it in place.

## Mechanism

`shell/plugins/panels/audio/Panel.qml` stores live `PwNode` objects in `cachedAudioSinks` / `cachedAudioSources` (lines 76-77, updated at 151-152), and `audioSinks` / `audioSources` fall back to those caches when `rawAudioSinks` / `rawAudioSources` is empty (lines 93-94).

When WirePlumber restarts (or a device goes away), those `PwNode`s are destroyed but stay in the cache as dead wrappers. If no replacement sink/source shows up, the live list is empty and the panel uses the stale cache. Opening the panel then runs:

`onOpenedChanged` → `refreshDisplayAudioModels()` → `displayAudio* = listSnapshot(...)` → `Repeater { model: root.displayAudio* }` rebuilds with a null QObject.

While the delegate incubates, the non-library `Model.js` is re-evaluated in the delegate context, and its top-level `typeof module` lookup goes through the delegate's model object. Qt's JS-array model adaptor (`VDMListDelegateDataType`, libQt6QmlModels) then calls a virtual method on the null QObject* (`mov (%rax),%rdx` with `rax = 0`).

This is the same class of bug that the bluetooth and network panels already work around by building primitives-only rows (see the comments on `deviceRow()` and `wifiRow()` in their `Model.js`).

In this session WirePlumber had been restarted 5 times, and afterwards the machine had no hardware sinks or sources at all (only the `omarchy_speaker_tuning` filter sink), so the fallback to the stale cache was taken.

## Stack (abridged; Qt frames unsymbolized because Arch debuginfod has no debuginfo for these qt6-declarative build IDs)

```
#4  libQt6QmlModels.so.6 + 0xb3530        <- SIGSEGV, null QObject* virtual call
#5  libQt6QmlModels.so.6 + 0xb3f01
#6  libQt6Qml.so.6 + 0x3880a7
#7  QV4::QObjectWrapper::getQmlProperty
#8  QV4::QQmlContextWrapper::getPropertyAndBase
#10 QV4::Runtime::TypeofName::call
#13 QV4::Module::evaluate
#15 QQmlEnginePrivate::createInternalContext
#19 QQmlIncubatorPrivate::incubate
#22 QQuickRepeater::regenerate
#23 QQuickRepeater::setModel
#29 QQmlBinding::update
#36 QV4::Runtime::StoreNameSloppy::call   <- displayAudio* = ...
#40 QV4::Runtime::CallQmlContextPropertyLookup::call   <- refreshDisplayAudioModels()
#46 QQmlBoundSignalExpression::evaluate   <- onOpenedChanged
...
#106 QQuickMouseArea::mouseReleaseEvent
```

Note: I matched the QML side of the stack to the audio panel from the frame shape, the code, and the system state, not from a symbolized QML filename. Every step lines up with this path.

## Expected

Opening the audio panel never crashes, even after audio devices disappear.

## Steps to reproduce (expected)

1. Have at least one hardware audio source/sink so the cache gets filled.
2. Restart WirePlumber, or remove the device, so that no hardware source/sink comes back.
3. Open the audio panel.

## Suggested fix

Do what bluetooth and network already do: give the Repeaters primitive rows and resolve the `PwNode` on demand. Alternatively, filter `cachedAudioSinks` / `cachedAudioSources` against `Pipewire.nodes.values` before using them, or don't cache node objects at all.

## System

- Omarchy 4.0.4-1, kernel 7.2.5-3-omarchy
- Dell XPS 14 DA14260 (Intel Xe2 graphics)
- quickshell 0.3.1-1, qt6-declarative 6.11.2-1, qt6-base 6.11.2-3
- pipewire 1.6.8-1, wireplumber 0.5.17-1, hyprland 0.56.2-2

(`omarchy debug` isn't available in 4.0.4, so I collected these details by hand.)

---
Filed by Claude Opus 5.5 via Claude Code.