## What happened

Running `omarchy restart shell` from a terminal whose `HYPRLAND_INSTANCE_SIGNATURE` names an
instance that no longer exists kills the running shell and never starts a new one. The session is
left with no bar and no shell until it is relaunched by hand; the command reports only:

```
Omarchy shell did not become ready after restart.
```

The two halves of the restart do not agree on how they reach the session:

- the kill — `quickshell kill -p "$CONFIG_DIR" --any-display` — matches on the Wayland display, so
  it works regardless of the signature, and it did kill the live shell;
- the relaunch — `hyprctl dispatch 'hl.dsp.exec_cmd("omarchy-launch-shell")'` — goes through the
  Hyprland socket named by `HYPRLAND_INSTANCE_SIGNATURE`, which was dead, so nothing was launched.

`hyprctl` reports that failure clearly, but the script discards it:

```bash
$ HYPRLAND_INSTANCE_SIGNATURE=dead_instance_0000 hyprctl dispatch 'hl.dsp.exec_cmd("true")'
Couldn't connect to /run/user/1000/hypr/dead_instance_0000/.socket.sock. (4)
$ echo $?
4
```

```bash
# omarchy-restart-shell
hyprctl dispatch 'hl.dsp.exec_cmd("omarchy-launch-shell")' >/dev/null   # exit status ignored
```

The script does derive a signature from the newest instance runtime dir, but only when the
variable is **empty**:

```bash
if [[ -z ${HYPRLAND_INSTANCE_SIGNATURE:-} ]]; then
  hypr_dir=$(find "${XDG_RUNTIME_DIR:-/run/user/$UID}/hypr" ... )
  [[ -n $hypr_dir ]] && export HYPRLAND_INSTANCE_SIGNATURE=${hypr_dir##*/}
fi
```

A value that is set but dead therefore goes unchecked, and it is the case that costs the user their
shell — an unset one is handled.

In the session where this happened, the terminal's environment named an instance ~11 days older
than the live one, while `systemctl --user show-environment` had the correct, live value. I did not
establish how the terminal's copy went stale, and the bug does not depend on that: any dead value
reproduces it.

The journal shows the whole sequence — the old shell exits on the IPC kill, and no new instance
ever logs anything:

```
omarchy-shell[3217226]: INFO: Exiting due to IPC request.
(no further omarchy-shell entries)
```

## Expected

Either the shell comes back, or the running shell is left alone. Being killed and not replaced is
the one outcome that should not be possible — a restart that cannot relaunch should fail before it
kills anything.

## Steps to reproduce

```bash
export HYPRLAND_INSTANCE_SIGNATURE=dead_instance_0000   # any instance that is gone
omarchy restart shell
```

The bar disappears and does not return; the command prints "Omarchy shell did not become ready
after restart". Recovery:

```bash
HYPRLAND_INSTANCE_SIGNATURE=$(systemctl --user show-environment | sed -n 's/^HYPRLAND_INSTANCE_SIGNATURE=//p') \
  hyprctl dispatch 'hl.dsp.exec_cmd("omarchy-launch-shell")'
```

## Suggested fix

Validate the signature **before** the kill, and treat a set-but-dead value like an unset one:

1. probe the socket (`hyprctl version >/dev/null 2>&1`, or test
   `$XDG_RUNTIME_DIR/hypr/$HYPRLAND_INSTANCE_SIGNATURE/.socket.sock`);
2. if it fails, fall back to `systemctl --user show-environment`, then to the newest instance dir —
   the existing `find` already does the last part;
3. if no signature works, exit before killing, saying so;
4. check the relaunch dispatch's exit status rather than sending it to `/dev/null`, so the failure
   is reported where it happens instead of 2 seconds later as a timeout.

## System

- omarchy 4.0.2-1
- Hyprland 0.56.2 (commit efb5099, v0.56.2 branch)
- Quickshell shell from `/usr/share/omarchy/shell`
- `omarchy debug --no-sudo --print` output available on request; nothing in it is specific to this
  failure, which is entirely in `omarchy-restart-shell`'s control flow.

---

Filed by Claude Opus 5 via Claude Code.