OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
omarchy-core low occasional DISPUTE

Dismiss the "Process crashed" notification, or turn crash capture off

omarchyomarchy-shellquickshell
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Checked every claim against the shipped source on this workstation (omarchy 4.0.2-1, Hyprland 0.56.2) and against the upstream tags, and re-read issue #7711 in full today. Confirmed on this machine, all read from the installed files. /usr/share/omarchy/bin/omarchy-crash-watch sends --urgency critical with --exec omarchy-agent-crash. /usr/lib/systemd/user/omarchy-crash-watch.service carries ConditionPathExists=!%h/.local/state/omarchy/toggles/crash-capture-off. omarchy-toggle-crash-capture calls omarchy-toggle crash-capture-off, stops the unit and sends "Crash capture disabled". omarchy-toggle-enabled tests $HOME/.local/state/omarchy/toggles/<flag>. The menu entry is trigger.toggle.crash-capture with label "Crash Capture" in /usr/share/omarchy/default/omarchy/omarchy-menu.jsonc:87, under trigger ("Trigger") and trigger.toggle ("Toggle"). SUPER + SPACE is the Omarchy menu in /usr/share/omarchy/default/hypr/bindings/utilities.lua:1. core_pattern pipes to /usr/lib/systemd/systemd-coredump from systemd 261.2-1, so dumps survive the watcher being off. NotificationCard.qml:77-87 maps the right button to closeRequested(), which Service.qml:1055 routes to service.dismissPopup(), so right-click dismiss is real. Issue #7711 does support the symptom: the reporter asks for a visible dismiss affordance on Omarchy 4.0.0-1, one commenter answers "You can still right click to dismiss it", and two more name the crash-capture toggle and place it in the menu under triggers.

One claim was wrong. The cause said the close affordance "is only revealed on hover", which the record took from a commenter summarising the two pull requests. No release has any close button at all: I fetched NotificationCard.qml at v4.0.1, v4.0.2, v4.0.3 and quattro, and the local file is byte-identical to both v4.0.2 and v4.0.3 while only quattro HEAD (unreleased, after the v4.0.3 tag of 2026-09-08) adds the hover-revealed ✕ that #8980 and #9010 then fight over. The cause is rewritten to say that, and the "neither merged as of 2026-09-07" date is refreshed: gh api repos/omacom/omarchy/issues/8980 and /9010 both report state: open on 2026-09-11. I also added the OMARCHY_CRASH_IGNORE drop-in to the fix, because the symptom's own audience is people who crash programs deliberately and the shipped watcher already supports skipping named binaries without giving up crash capture.

Not exercised: I did not crash a process, did not toggle crash capture, and did not click a toast, because this audit is read-only on the workstation. The right-click gesture, the toast's non-expiry and the drop-in are therefore confirmed from source and from systemd's documented drop-in behaviour rather than from a live trial. ulimit -c is unlimited and DefaultLimitCORE=infinity, so the record's sleep 100 & kill -SEGV $! check should produce a dump here, but I did not run it.

The cause above was rewritten on 2026-09-11 to match this note; the fix was corrected by the audit itself.

Symptom

Omarchy 4 shows a critical "Process crashed" notification whenever a process dumps core. Clicking it launches omarchy-agent-crash, the AI crash diagnosis. There is no visible close control on the toast, and because the notification is critical it does not expire, so the only obvious way to make it go away is to start a diagnosis you did not want. People who crash programs on purpose (compiler and debugger work) get one after every crash. Reported on Omarchy 4.0.0-1.

Cause

omarchy-crash-watch, run by the user unit omarchy-crash-watch.service, sends the toast with --urgency critical and --exec omarchy-agent-crash, so a left click on the card runs the diagnosis. The toast never expires on its own: durationFor() in shell/plugins/notifications/Service.qml:102-105 returns 0 for NotificationUrgency.Critical, and the popup countdown only runs while that duration is above zero.

No shipped Omarchy 4 release puts a close button on the notification card. shell/plugins/notifications/components/NotificationCard.qml is byte-identical on this workstation (omarchy 4.0.2-1) and at the v4.0.1, v4.0.2 and v4.0.3 tags, and the only pointer handling in it is one full-card MouseArea at line 77 with acceptedButtons: Qt.LeftButton | Qt.RightButton, which maps the right button to closeRequested() and every other button to cardClicked(). A hover-revealed ✕ was added to that file on the unreleased quattro branch after v4.0.3, and the two open pull requests argue over it rather than introduce it: #8980 keeps it on without hover, #9010 replaces it with a themed PanelActionButton. Both were still open on 2026-09-11. So on every release through v4.0.3 the right click is the only dismiss gesture the card offers, and there is nothing on screen to say so.

Fix

Dismiss one toast: right-click it. The notification card handles a right button click as a close request and a left click as the card's action (shell/plugins/notifications/components/NotificationCard.qml:77-87 on Omarchy 4.0.2-1, wired to service.dismissPopup() at Service.qml:1055). One commenter reported this and it holds in the shipped source.

Stop the notifications altogether: toggle crash capture off. Two reporters confirmed the toggle exists. Either from the Omarchy menu, Super+Space, then Trigger, Toggle, Crash Capture, or from a terminal:


omarchy-toggle-crash-capture

That writes the flag ~/.local/state/omarchy/toggles/crash-capture-off, stops omarchy-crash-watch.service for this session, and sends a "Crash capture disabled" notification. The unit carries ConditionPathExists=!%h/.local/state/omarchy/toggles/crash-capture-off, so it stays off on the next login until you run the same command again, which removes the flag and starts the watcher.

Silence only the programs you crash on purpose, and keep the rest: omarchy-crash-watch reads an extended regex of process names to skip from OMARCHY_CRASH_IGNORE, matched against the executable's basename. Set it on the unit rather than in your shell, because the watcher is a systemd user service and does not inherit your interactive environment:


mkdir -p ~/.config/systemd/user/omarchy-crash-watch.service.d
cat > ~/.config/systemd/user/omarchy-crash-watch.service.d/ignore.conf <<'EOF'
[Service]
Environment=OMARCHY_CRASH_IGNORE=^(cc1|cc1plus|lldb|my-test-binary)$
EOF
systemctl --user daemon-reload
systemctl --user restart omarchy-crash-watch.service

The same mechanism exposes OMARCHY_CRASH_DEDUPE_SECONDS, which defaults to 60 and is the window in which one program is announced at most once.

Core dumps are still captured by systemd-coredump with the watcher off or the pattern set. /proc/sys/kernel/core_pattern pipes to /usr/lib/systemd/systemd-coredump regardless, only the notification and the one-click diagnosis go away, and coredumpctl list still works.

Verify


omarchy-toggle-enabled crash-capture-off && echo off
systemctl --user is-active omarchy-crash-watch.service

off and inactive while disabled. Crash something (sleep 100 & kill -SEGV $!) and no toast appears.

Sources

crash-capture-notification-dismiss-or-disable