Skip to content

Sync upstream quattro through #19168 - #135

Merged
mark-groves merged 61 commits into
quattrofrom
cursor/upstream-changes-pr-3b36
Nov 17, 2026
Merged

mark-groves merged 61 commits into
quattrofrom
cursor/upstream-changes-pr-3b36

Conversation

@cursor

@cursor cursor Bot commented Nov 16, 2026

Copy link
Copy Markdown

What changes

Upstream quattro moved past omacom#19017 (29fbc4e2d), which is what #134 merged. Sixty-one commits replayed onto current quattro, through 3b36b9e19 (T2 packages from the Omarchy repository, omacom#19168). No merge commits. Nothing skipped. No conflicts. Catalogs, the menu, the mise tool list, the agents panel, Menu.qml, shell.qml, test/cli, and the agents and default-agent tests auto-merged. All 59 paths both sides changed three-way merge to this branch. Fork delta vs 3b36b9e19 is 253 paths, the same set as quattro vs 29fbc4e2d. On the 194 paths upstream did not touch, nothing differs. Face, Grok Bot, Cursor, linux / linux-headers, and the dropped translation-sync workflow stay.

Verification

  • Cherry-pick of the sixty-one commits from 4bf10ed28 through 3b36b9e19 applied onto quattro 07f7562f8. No conflicts. git diff --name-only of HEAD vs upstream 3b36b9e19 is 253 paths, the same set as quattro vs 29fbc4e2d. On the 194 paths upstream did not touch, nothing differs. All 59 paths both sides changed three-way merge to this branch. No conflict markers remain. linux and linux-headers are still in the other-packages list. Face, Grok Bot, and Cursor are still in the menu. The translation-sync workflow is still absent. cursor-agent is still out of the mise tool list (20 auto_update, 30 lazy). Power profiles still write through busctl. FaceChromeCanvas is still immediately before FilterInput. shownProviders stays.
  • ./test/cli passed (138).
  • ./test/shell was not run as a whole. The files this sync covers passed: i18n-test.sh (401, one structural skip), agents-panel-test.sh (119), default-agent-test.sh (48), network-test.sh (104), weather-test.sh (58), screenrecording-test.sh (40), image-deferred-hardware-test.sh (38), keybindings-menu-test.sh (29), plugin-clone-test.sh (21), lid-inhibit-test.sh (20), preinstalls-test.sh (15), menu-images-test.sh (14), monitor-modeless-test.sh (15), theme-install-guards-test.sh (13), herdr-functions-test.sh (12), audio-pactl-test.sh (11), hw-nvidia-test.sh (11), theme-staging-test.sh (9), agent-usage-update-test.sh (8), launcher-remove-test.sh (7), hardware-quirks-test.sh (7), screenrecording-concurrency-test.sh (7), screenrecording-wf-audio-test.sh (7), notifications-test.sh (4), snapdragon-hardware-test.sh (4), agent-instructions-test.sh (4), capture-picker-binds-test.sh (2).
  • pointer-move-gate-test.sh skipped: no Wayland compositor, so the new submenu-click fixture did not run.
  • pacman-templates-test.sh stopped on a missing pacman-conf, before any assertion. This sandbox does not have it. The T2 change only removes the old-repository lines from that file.
  • This checkout has no Omarchy session, so the new icons, the Wi-Fi prompt, the network panel, the agents panel, and the launcher were not drawn. No other sync was open.

Screenshots

No visible change captured here: this checkout has no Omarchy session, so nothing was drawn. Checkmate and Cards, the Media Viewer and Chromium icons, the Wi-Fi passphrase prompt, the network panel during a switch, and the agent starter tiles are visible on a machine. The tests above cover the behavior that can run here.

Open in Web View Automation 

birkskyum and others added 30 commits November 16, 2026 23:02
* Draw Media Viewer's icon as the set's other apps are drawn

Omaview's own icon was a dark tile with a yellow button, unlike every
other icon in the launcher. Both sets now draw Image Viewer's photo print
with Media Player's play button on its corner, in the theme's colors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Draw Chromium's icon in the icon sets

Chromium's logo is Chrome's pinwheel with its segments joined, each a
shade of its own, around a ring. Both sets draw it with the top segment
darkened and the right one lightened from the main color, so the three
read apart in light and dark themes alike.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Markdown had no default of its own, so .md files opened in whichever app
the desktop database listed first, Obsidian on an install that has it.
The packaged list now names Omawrite for text/markdown and its alias.
An exact type wins over the text/plain editor lines in a user's own list,
and a Markdown choice the user made there still wins over this one.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…run (omacom#18550)

* Install Node with opencode, qwen, ghui and playwright on their first run

mise's shim installs a registry tool whose first backend is npm on its own, before the lazy install that would bring the lazy Node it needs, so the first run of these four fails with "failed to resolve configured install dependency 'node@latest' ... while offline". Keeping them out of the shim's automatic install sends their first command through the lazy install instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Say why the auto-install list lives with the tool declarations

In omarchy.toml it would outlive an opt-out of the default tools and keep a user's own declaration of one of these tools from installing on its first command.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: muse-wonder via pi, thinking high <noreply@meta.com>

---------

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: muse-wonder via pi, thinking high <noreply@meta.com>
…it (omacom#18551)

wf-recorder acts on SIGINT only when its next frame arrives, and by default it asks only for frames that changed, so on a still screen the stop waited 5 s, killed the recorder and warned the video may be corrupted. -D makes it capture every frame.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
--set checked only the shape of the pair, so 91,181 was stored and the panel asked Open-Meteo for a place that cannot exist, which never answers and leaves the bar without its weather icon. The range check runs inside the jq call that builds the file, since bash has no floats, and the file is written only when jq produced something. The panel's reader drops an off-globe pair and keeps the name, which also repairs a file saved before this change or written by hand.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Muse Wonder High via Pi <noreply@meta.com>
…m#18554)

omarchy agent prompt '' counted one word and forwarded --prompt "", where omarchy-agent's ${2:?} check aborted with bash's own line-number error.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Stop agent usage-update on an empty --except

An empty or missing --except value failed the array assignment with a bad subscript, which does not stop the script, so it went on to refresh every agent and exit 0.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep the --except refusal and its assignment as two branches of one if

AGENTS.md asks for a full if/else rather than an exit that makes the lines after it unreachable.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ord (omacom#18556)

Codes were matched inside a slice cut out around each keyword, and the slice's edges counted as word boundaries, so a number crossing a window's start was offered as its tail (23456) and one crossing its end was cut short (48291). Matching over the whole text reads each code to its real edges; keeping only matches whose code starts inside a window stops a number that merely grazes a window's start from being pulled in and outranking the real code.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
omarchy-theme-set lowercased the name and turned spaces into dashes before looking for the folder, so a hand-made "Walk Handmade Theme" could never be set, and a hand-made "Gruvbox" set the bundled gruvbox instead, from the CLI and from the picker, which passes folder names. omarchy-theme-remove did no conversion at all, so the listed name of a dashed folder, and its own example, were never found. set, remove and dev theme-preview now try a folder of exactly the given name first and fall back to the existing conversion, and remove says when the name it found is a bundled theme rather than calling it missing.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…they bind (omacom#18558)

Hyprland 0.57 reports a Lua bind as HL.Dispatcher(<name>) or function: 0x... with a display-only number, where 0.56 said __lua. The menu recognised only __lua, so it dispatched the reported name and number, which Hyprland rejects, showed description-less binds like the lid switches, and kept the two Close window chords on separate rows. All three names now take their command from the Lua source. The records cache key moves to v20 so records written by the old script are rebuilt.

Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Muse Wonder High <noreply@meta.com>
…18559)

With no default source, wpctl prints an error and exits 0 with nothing on stdout, so input mute read the empty answer as unmuted and showed "Microphone on". It now asks for the source's volume first and, without a Volume: answer, shows "No microphone found" and exits 1. The screen recorder's wf-recorder path likewise took pactl's @DEFAULT_SOURCE@ placeholder for a device.

Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex Medium <noreply@openai.com>
…ckground picker (omacom#18561)

* Say there are no plugins instead of opening an empty removal picker

plugin remove only checked that the plugins folder exists, and removing the last plugin leaves the folder, so gum choose got no options and the command ended on "cancelled". It now collects the list first and says there are no plugins installed, and a folder it cannot read says so rather than ending silently under pipefail.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Tell the background switcher's user there are no backgrounds

A theme with only colors.toml has no backgrounds, and menu-images opened the image picker on no rows. The picker reveals itself on its first preview, so it never appeared but held the keyboard until Escape. menu-images now exits 3 when it has nothing to offer, after the preload branch so the shell's cached copy is still refreshed, and the switcher sends the "No background was found for theme" notification that theme set and bg next already send.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…led (omacom#18562)

Escape makes gum exit 1, which set -e passed straight out, so the floating terminal ended a cancel on a red Failed. 130 is what the wrapper reads as the user backing out, as the Docker DB picker and passwordless sudo already use.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…acom#18565)

A click into a submenu rebuilds the rows under a still pointer and relied on Qt hovering the new row so its onEntered could select it. An absolute pointer (a VM tablet, a remote desktop, a pen tablet) that jumps onto a row and clicks in one event gets no such hover, so the cursor stayed on the first row and Enter applied it. The click now passes its own point, and setActiveMenu selects the row under it through the same pointer gate once the rows are rebuilt.

The fixture loads the real Menu.qml and clicks into Bar Background with no hover at all, which is what the absolute pointer leaves it.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex Medium <noreply@openai.com>
Co-authored-by: Muse Wonder High <noreply@meta.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Validate pane counts before creating layouts

* Fold the pane count check into the existing usage guard

One line per function instead of a four-line if that repeats the usage message. The tests now also cover -1 and a bare word that bash arithmetic would read as a variable name.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…acom#13237)

* Cap the command list column width to keep summaries from wrapping

* Correct the command column comment's example

The comment named agent usage update as the usage pushing the column past 90 characters, but that usage is 82; the ones the cap exists for are over 200.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Say that any usage over the cap pushes only its own summary

The comment named usages past 200 characters, but every usage longer than 48 characters is the case the cap handles.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Test that the commands table pads its usage column to at most 48 characters

Adapted from the test on omacom#10847, which fixes the same bug with a different cap: a short command's summary must start within the capped column, not after the longest usage in the table. It fails when the cap is removed.

Co-Authored-By: Harlan Jones <harlanljones@gmail.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Harlan Jones <harlanljones@gmail.com>
Pressing Esc in Install > Development > Docker DBs fell through to
'|| main_menu', which nothing defines anymore. Bash printed
'main_menu: command not found' and then the no-selection message.

Cancelling now exits 130, the status the floating-terminal wrapper
treats as the user bailing, so the terminal closes without the Done
prompt. Same convention as omarchy-sudo-passwordless and
omarchy-agent-account-add.

Co-authored-by: steelcityappliance <steelcityappliance@users.noreply.github.com>
…macom#19088)

* Open config files without a notification of their path

Plain shows the path of the file it opens, so the toast naming it says
nothing the editor doesn't. The command stays, as the Setup menu and
the agent instructions open config files through it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Open config files with omarchy-launch-editor

Without its notification, omarchy-launch-config-editor only handed its
path to omarchy-launch-editor, so the Setup menu and the agent
instructions call that directly and the command goes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Rearm idle monitor after timeout changes

* Correct why the idle monitor needs re-arming

Quickshell does replace the notification when the timeout changes: updateNotification() deletes it and creates a new one. The failure is that the replacement can land at the deleted one's address, so Qt sees bNotification unchanged and isIdle never subscribes to the replacement. Disabling first sends the pointer through null, which is why the pulse works.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>

* Retime the idle monitor only once the user is back

Since quattro's idle service closes the screensaver and cancels the cycle on any active signal past the launch grace, the re-arm pulse fired mid-cycle read as the user returning: a shell.json timeout change while the screensaver was up closed it, ran omarchy-system-wake and pushed the lock back by the time already spent idle. Ignoring the pulse's active signal instead would leave a user who then comes back unseen, since the fresh notification starts out active and can only report a return after it first idles.

So the monitor's timeout is no longer bound. It is set, and pulsed, only while the monitor reports active: straight away when the user is at the machine, as when they edit shell.json, and otherwise on the return that ends the cycle. Until then the running cycle keeps its notification, so it still sees the return and keeps its deadlines.

Co-Authored-By: Codex Medium <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex Medium <noreply@openai.com>
Co-authored-by: David Heinemeier Hansson <david@hey.com>
…macom#19096)

omarchy-kernel compared menu entries through printf | grep -q under pipefail, so an early match could SIGPIPE printf and read as a miss. The privileged environment sanitizer waited on a process substitution Bash could already have reaped; it now reads a lastpipe pipeline and takes env's status from PIPESTATUS, failing closed with job control on.

The screen recording test let a pending start's shortened ready timeout run out before the toggle arrived, the thumbnail test checked for files before generate_thumbnail moved them into place, and the router timing read a wall clock that time sync steps right after the CI VM boots.

Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…macom#19097)

On a packaged install omarchy-debug and omarchy-debug-idle ship to /usr/bin from omarchy-settings, outside the directory the router scans, so `omarchy debug` prints "Unknown Omarchy command" while `omarchy-debug` works everywhere. The main help, the command examples and the CLI manual chapter now name the command that works, and the debug group is no longer listed among the groups, since nothing in it is routable on a packaged install.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Hide agent starter tiles until an agent is ready

Show "Make something cool" only when a local model is running or a
signed-in subscription has a coding agent to launch, so it no longer
appears while the first local model is still downloading.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Count expired agents as not ready and keep arrow keys on a live row

Agents the panel can't sign in, like Copilot, no longer count as ready
for the starters once their sign-in lapses. Left and Right clamp the
cursor's row first, so the starters disappearing under it can't throw.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Omaview reshapes its window as it steps to a picture of another shape,
and draws the new picture at once, so the resize animation showed it
clipped to the old shape until the window caught up.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…om#19103)

* Space out password dots and add a show-password toggle to Wi-Fi prompt

The password dot is wider than a monospace cell, so masked text in the
shared TextField ran together. Space the dots while there are any, and
give TextField a `revealed` property. The Wi-Fi passphrase prompt uses it
for an eye button to the right of the connect check.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Scroll the Wi-Fi list to show a passphrase prompt opened on a bottom row

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Smooth the Wi-Fi passphrase prompt as it moves between networks

The prompt now slides open and shut and fades, instead of jumping, and
while one is open a scan no longer reshuffles the list around it: rows on
screen keep their place and signal until the prompt closes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Fix held Wi-Fi rows on odd SSIDs and re-mask a quickly reopened prompt

Key held rows on a prototype-free object, so a network named like an
Object method can't break the list refresh, and reset the show-password
toggle when the prompt closes rather than when its fade ends.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
omarchybot and others added 28 commits November 16, 2026 23:03
* Extract embedded English into catalogs

* Update translations

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Avoid Baffin GPU hangs when screen recording

* Keep the encoder test away from a real recording

The test left pgrep and XDG_RUNTIME_DIR pointing at the machine running it, so a recording in progress sent the script down its stop path and SIGINTed the user's recorder, and a recorder state file could land in the real runtime directory. Its omarchy-shell stub also exited 1, which, whenever the stub recorder was still alive at the start check, failed the start and ran cleanup_webcam's real pkill of a running webcam overlay. Stub pgrep, use the temp dir as the runtime directory as screenrecording-test.sh already does, and make the stubs succeed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>

* Record on the CPU on every AMD Polaris GPU, not only Baffin

omacom#9555 has the same ring vce0 timeout and failed GPU reset on an RX 580 (1002:67df) and an RX 590 as on the Baffin MacBook, so match every device ID amdgpu assigns to CHIP_POLARIS10, 11, 12 and CHIP_VEGAM, all of which run the same VCE 3.4 block. CPU encoding stays the workaround: it is what the RX 580 reporters and the Baffin author confirmed, while HEVC (which Mesa encodes through UVD rather than VCE on these cards) is only suggested, not yet tried on affected hardware.

Read the IDs from sysfs, as the omarchy-hw-* GPU checks do, rather than lspci, which reads config space and wakes a runtime-suspended dGPU on every recording start. That also lets the test fake the hardware with OMARCHY_PCI_DEVICES_PATH, so the encoder check becomes one case in screenrecording-test.sh, through the real entry point, instead of a separate file.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex Medium <noreply@openai.com>
…om#7188)

The portal backend is the opt-in path for HDR, external-GPU monitors and
window capture, but --fullscreen was matched first and passed the focused
monitor straight to the kms backend, so the variable never applied there.

On a hybrid-GPU laptop the kms backend only enumerates the connectors of
the single DRM card it attaches to, so a monitor driven by the other GPU
is not visible to it and recording fails. That left no configuration in
which ALT + PRINT could record such a monitor, even though the portal
backend handles it fine.

Check the variable first so it applies to every capture path.

Co-authored-by: juanitoclavel <juanitoclavel@gmail.com>
omacom#13817)

* Document omarchy-debug directly in bundled agent skills (omacom#13702)

* Say why diagnostics are the exception to the omarchy <group> <action> form

The skill tells agents to always prefer the routed form, so without a reason an agent can "correct" omarchy-debug back to omarchy debug, which fails on a package install: the router only searches its own directory, and omarchy-debug ships to /usr/bin from omarchy-settings. The manual gets the same pointer for people. Both say the routed form can fail rather than that it does not exist, since it works on a dev-linked checkout and will work everywhere once the router looks further. Folded in from omacom#14129.

Co-Authored-By: plusultra <243763147+HerrStolzier@users.noreply.github.com>
Co-Authored-By: Codex Medium <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: szaidi <szaidi@users.noreply.github.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: plusultra <243763147+HerrStolzier@users.noreply.github.com>
Co-authored-by: Codex Medium <noreply@openai.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* Explain when GTK4 breaks on NVIDIA and how to choose another renderer

The comment blamed Vulkan on NVIDIA in general; the failure shows most
reliably when GTK's Vulkan device isn't the compositor's GPU. Point users
who want another renderer at hl.env in their own config, which loads after
the defaults: Hyprland's environment can't tell a user's session value
from one set on an earlier load, since each reload starts a new Lua state.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Date GTK's Vulkan default to 4.16, the release that made it

GTK 4.14 made the new GL renderer the default; Vulkan became the default on Wayland in 4.16.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Force the GL renderer for GTK4 on NVIDIA

GTK 4.14+ defaults to Vulkan, which leaves Nautilus and other GTK4 apps
as black windows or crashing on NVIDIA. The cursor change is dropped:
Hyprland's auto defaults already pick a CPU buffer on NVIDIA.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Cover GSK_RENDERER in the NVIDIA environment test

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
* Install Checkmate by default

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

* Draw Checkmate in the Papercraft and Arcade icon sets

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…m#19106)

* Hold the network panel's layout while switching Wi-Fi networks

Connecting to one network while another is up dropped the panel to its
disconnected layout in between: the stats grid, band row and hero all
vanished and came back. Through the switch the hero now names the network
being joined, and the sections hold their place reading "--" until the
new link reports.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep the network hero's buttons in place while switching Wi-Fi

The QR and speed-test buttons vanished as the old network dropped and
came back with the new one, reflowing the name beside them. The ones that
were showing now stay, disabled, until the switch settles.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Only an uninstall that needs a terminal closes it; other removals drop the entry in place.

Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
…holds every device tree (omacom#19139)

* Build Snapdragon boot images with systemd-stub-omarchy, whose header holds every device tree

* Run the Snapdragon test without the installed stub unless a case names one
The UX3407QA and UX3407RA boot to a black screen at the unlock prompt: the display driver waits for the panel, USB-C and HDMI chains, which load only after unlock. Load them early, as on the Acer Aspire 16 AI, which has the same retimers and HDMI bridge.
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…m#19152)

Their install migrations didn't redraw the icon set, so the launcher
showed each package's own icon until the next theme switch.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…acom#18629)

* Show the disk unlock prompt on the Lenovo IdeaCentre Mini 01Q8X10

* Name the IdeaCentre Mini's X Plus and Snapdragon X models
* Show the disk unlock prompt on the Lenovo IdeaPad Slim 5x

* Rebuild the Slim 5x's boot image unless this migration did already, and test the drop-in's board gate instead of its contents
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…acom#18776)

* Let the DSPs start when a Snapdragon root disk is on a USB-A port

* Record the DSP blacklist rebuild as owed before the setup can lift it
Omarchy now builds linux-t2, its headers, apple-t2-audio-config,
apple-bcm-firmware-fetcher and t2fanrd in its own repository. T2
installs no longer need the extra repository the installer added for
them, so the installer stops adding it.

The migration moves existing T2 installs over once per machine. It
reinstalls every installed T2 package from [omarchy] by name while the
old repository is still configured, then drops that repository's
section and checks the result with pacman-conf. If [omarchy] does not
serve a package yet, or the reinstall fails, it changes nothing and
stays pending.
@mark-groves
mark-groves merged commit 13de75f into quattro Nov 17, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.