Focus stealing: let activation tokens decide, instead of per-app rules

[dmkenney]

Status: every piece is now a PR or an upstream thread. Merge order and latest test results are in the status comment.

	What
Omarchy
#16079	Browser launches pass the click's token. Reviewed, marked ready
#16097	Apps launcher passes the click's token
omacom/omarchy-pkgs#1367	quickshell: tokens on notification clicks, plus Quickshell.activationToken()
omacom/omarchy-pkgs#1374	Ghostty 1.3.1 passes the token on link clicks, until 1.4.0 ships the upstream fix
omacom/omarchy-pkgs#1375	Alacritty package with the token on link clicks
omacom/omarchy-pkgs#1379	Hyprland for x86 with the stricter check (merge after #16097 and #1367)
Upstream
hyprwm/Hyprland#16608	The stricter serial check
rust-windowing/winit#4782	winit sets the click serial on tokens
rust-windowing/winit#4783	Same, for winit 0.30 (Alacritty)
quickshell-mirror/quickshell#1166	Tray clicks and a shared activation helper (another author). Merged
quickshell-mirror/quickshell#1351	Tokens on notification clicks. Reopened as a partial duplicate; now rebased on #1166's helper
quickshell-mirror/quickshell#1353	Quickshell.activationToken() (draft, on top of #1351)
ghostty-org/ghostty#14787	Token on link clicks (fixes ghostty-org/ghostty#14785). Merged, ships in Ghostty 1.4.0
Related
#14282	The original issue: any activation request jumps the workspace
#14283	Stopgap: focus off for browsers; its rule can go once this ships
#7738	Stopgap: notification clicks focus the urgent window. Closed; the new notification engine (#15932) replaced the code it changed
#15750	Agent-driven browsers opening new windows; not covered by tokens, still needed
#15661	Says the focus_on_activate window rule is a no-op; it does take effect on 0.56.2

The problem

When a script, an agent or another app opens a link, the browser asks Hyprland for focus and the view jumps to the browser's workspace (#14282). Omarchy sets misc:focus_on_activate = true, so every request is granted.

Turning that off is not a fix either. Clicking a notification raises the sender through the same request, so with it off a Teams or Slack toast click only closes the toast.

The open PRs work around this per app:

• Keep browsers from taking focus on activation requests #14283 turns focus_on_activate off for browser classes.
• Notifications: clicking a toast focuses the exact sending window when focus_on_activate is off #7738 catches the urgent event Hyprland emits when it denies a request, and focuses that window after a toast click.
• Keep agent-driven browsers from taking the keyboard #15750 parks agent-driven browsers on a special workspace.

I ran #14283 with a port of #7738 and it works. But Hyprland can't tell a click from an automated open, so the rules have to guess per app. Wayland already has the signal that tells them apart, and right now it gets dropped or ignored.

How Wayland is meant to handle it

xdg-activation-v1 tokens carry the serial of the input event that caused them. When you click a link, the app you clicked in asks for a token with that click's serial and passes it on (XDG_ACTIVATION_TOKEN). The app that receives it uses the token to request focus. The compositor can then grant focus when the serial is from a recent real input, and only mark the window urgent when it isn't. GNOME and KDE use this to stop windows from stealing focus.

What I tested

Omarchy quattro plus #14283, Hyprland 0.56.2, quickshell with omacom/omarchy-pkgs#1367, Brave 1.97, Chromium 154, Chrome 155, Firefox 158, foot 1.28, Alacritty 0.17, Ghostty 1.3.1. I logged the apps' Wayland traffic with WAYLAND_DEBUG=1 and Hyprland's urgent events.

Who sends a token with a real serial

Source	Token	Serial
Notification click (quickshell + omarchy-pkgs#1367)	yes	the click
Link clicked inside Brave (e.g. in a web app)	yes	the click
Link opened in foot (Ctrl+Shift+O)	yes	the keypress
Link clicked in Alacritty	no	none
Link clicked in Ghostty	no	none

What browsers do when handed a URL

Browser	Uses a token it's given	No token: makes its own with
Chromium, Chrome, Brave	yes, via XDG_ACTIVATION_TOKEN or --xdg-activation-token	the last input serial it received, even after losing focus
Firefox	yes	the last input serial it received

Omarchy drops the token

• omarchy-cmd-browser-handoff writes the URL straight to Chromium's singleton socket and doesn't pass the token. Adding --xdg-activation-token=$XDG_ACTIVATION_TOKEN to the argv works: Chromium, Chrome and Brave all activated with the token they were given.
• omarchy-launch-browser and omarchy-launch-webapp start a browser through systemd-run, which doesn't pass the caller's environment. XDG_ACTIVATION_TOKEN never arrives. --setenv=XDG_ACTIVATION_TOKEN fixes it.

So a link you open in foot reaches the browser without proof of the click, and the browser makes its own token with an old serial.

Hyprland is fixing half of it

Hyprland 0.56.2 stores the serial and never checks it. hyprwm/Hyprland#16427 (merged Sep 30, not released) now rejects a token unless its serial is one of the last 100 sent to that client, and each serial works only once.

I built Hyprland main (2860d07b) and ran it nested with focus_on_activate = true:

Step	Result
Real link open from foot, token passed through a fixed handoff	focus moves to Chromium
Key pressed in Chromium, focus moved to foot, then a URL handed over with no token	Chromium reuses that key's serial; focus moves to Chromium
Same again	serial already used, rejected, focus stays

So once #16427 ships, the first automated open after you last used the browser still takes focus. Repeats are blocked.

To close the gap I wrote a follow-up on a fork: dmkenney/Hyprland@activation-reject-stale-serials. It records the serial of the last wl_keyboard.leave sent to each client and rejects activation tokens with an older serial. Same nested test:

Step	Result
Key pressed in Chromium, focus moved to foot, URL handed over with no token	rejected, focus stays on foot
Same again	rejected
Key pressed in Chromium while it has focus, then URL handed over	accepted
Real link open from foot, token passed through a fixed handoff	focus moves to Chromium
Notification click in quickshell (with omarchy-pkgs#1367), token passed to Chromium the way it handles its own notifications	focus moves to Chromium

hyprwm only takes PRs from vouched contributors, so this needs someone who is vouched to pick it up, or a nod from the Hyprland side.

Proposed changes

Hyprland

1. Follow-up to Tell T2 Mac owners to save their Wi-Fi firmware before installing #16427: reject serials from before the client's last focus loss. Written and tested on the fork above; needs a vouched contributor to take it upstream.
2. Possibly: apply the same check when a new window first opens, which today always takes focus. This needs launchers to pass tokens too, so it's a later step.

Omarchy
3. omarchy-cmd-browser-handoff: pass --xdg-activation-token.
4. omarchy-launch-browser / omarchy-launch-webapp: systemd-run --setenv=XDG_ACTIVATION_TOKEN.
5. Once Hyprland ships 1, drop the per-app focus_on_activate rules (#14283's browser rule, apps/telegram.lua) and keep focus_on_activate = true. Valid tokens get focus; everything else gets marked urgent.

quickshell
6. omacom/omarchy-pkgs#1367 patches quickshell to send a token with notification actions. Upstream, this goes on top of quickshell-mirror/quickshell#1166 once it merges.

Terminals
7. Ghostty and Alacritty: request a token with the click serial when opening a link, like foot does.

Browsers
8. Chromium and Firefox: don't attach the last input serial to a focus request no input caused. With 1 in place this matters less, but it's the root of the stale-serial case.

3, 4 and 6 help now and fit the end state. 7 and 8 are independent. 5 waits for 1.

Existing PRs and issues

• Keep browsers from taking focus on activation requests #14283: works today. With 1 in place the browser rule isn't needed.
• Notifications: clicking a toast focuses the exact sending window when focus_on_activate is off #7738: closed. It can't be rebased onto the new notification engine (Replace the notification engine with Omapager #15932), which now focuses the sender's window itself when the app has no default action.
• Keep agent-driven browsers from taking the keyboard #15750: covers new windows and fullscreen from agent-driven browsers, which tokens don't cover until 2.
• apps/telegram.lua: focus_on_activate is not a window rule, so the Telegram focus-steal opt-out is a no-op (also affects #13688, #14283) #15661 says the focus_on_activate window rule is a no-op. In my tests on 0.56.2 it did take effect: with the global option on, Brave's requests were denied and its windows marked urgent.

Scope

Every result above comes from the apps and versions listed. Apps outside the tables weren't tested.

Feedback on the direction is welcome before I open the PRs.

[dmkenney]

Progress:

• Omarchy: Pass the activation token on to the browser #16079 passes the activation token through the browser handoff and the systemd-run launch (items 3 and 4).
• Hyprland: the stricter serial check is posted as Activation tokens with old serials still steal focus after #16427 hyprwm/Hyprland#16608 (item 1), and carried in Build Hyprland for x86_64 and grant activation only for fresh input omarchy-pkgs#1379 meanwhile.

[dmkenney]

Status: merge order and test results

All PRs and upstream threads are listed in the table at the top.

Omarchy (merge in this order)

1. Send an xdg-activation token with quickshell notification actions omarchy-pkgs#1367 – quickshell sends an activation token on notification clicks, and adds Quickshell.activationToken() for QML.
2. Raise a running app the launcher is asked to open #16097 – the Apps launcher passes a token for the click, so an app that's already running still comes forward. New omarchy-launch-desktop-entry: gio launch keeps the token, gtk-launch didn't, and D-Bus activatable apps (Nautilus) get it through org.freedesktop.Application.Activate.
3. Pass the activation token on to the browser #16079 – the browser handoff, the systemd-run launch and the link entry (StartupNotify=true) pass the token to the browser.
4. Build Hyprland for x86_64 and grant activation only for fresh input omarchy-pkgs#1379 – Hyprland built for x86_64 too, with protocols/activation: reject invalid serials hyprwm/Hyprland#16427 backported and the stricter check below.

Independent: omacom/omarchy-pkgs#1374 (Ghostty 1.3.1 passes a token to xdg-open) and omacom/omarchy-pkgs#1375 (Alacritty with the winit and hint patches).

Why the order matters: #16427 alone already breaks the Apps launcher for running apps. gtk-launch replaces the token with one that has no serial, and relaunching a running Nautilus stopped raising it in a nested test. #16097 with #1367 fixes it, also tested nested. So #1379 should land after both.

Tested since the write-up

• Patched Hyprland 0.56.2 (the Some applications UI are "zoomed in" #1379 build), nested: automated opens after leaving the browser are rejected; a link from foot raises the browser.
• Ghostty 1.3.1 with the Add update-icon right-click open release notes on github #1374 patch, and Alacritty with the Recommended approach for setting up multiple machines with custom setup #1375 patches: a real click's token reached the running Brave, which used it.
• Unpatched Ghostty main: no token is requested on a link click, and the OpenURI options have no activation_token.

After it ships

#14283's browser opt-out and apps/telegram.lua can go, and #7738 isn't needed for apps that send tokens. #15750 stays: Hyprland still focuses brand-new windows regardless of tokens.

[dmkenney]

Update on quickshell:

• service/notifications: send an activation token with invoked actions quickshell-mirror/quickshell#1351 was closed as a duplicate of service/tray: forward Wayland activation tokens on primary clicks quickshell-mirror/quickshell#1166. Omarchy v2.0.2 #1166 adds tokens for tray clicks and a shared activation helper. In its review the maintainer asked for Qt's own activation path (requestXdgActivationToken()) instead of a second xdg_activation_v1 client, which is what Switch Between 2 Workspace per monitor #1351 added.
• Omarchy v2.0.2 #1166 doesn't cover notification actions. Once it merges, the notification part goes on top of it as a small PR. core/qmlglobal: add activationToken() quickshell-mirror/quickshell#1353 waits for that.
• Send an xdg-activation token with quickshell notification actions omarchy-pkgs#1367 is unchanged and still works on its own, so the merge order above stays the same.

[dmkenney]

Update:

• service/tray: forward Wayland activation tokens on primary clicks quickshell-mirror/quickshell#1166 merged. The maintainer reopened service/notifications: send an activation token with invoked actions quickshell-mirror/quickshell#1351 as a partial duplicate and asked for it to use Omarchy v2.0.2 #1166's token helper. It's rebased on that now, with no second xdg_activation_v1 client. core/qmlglobal: add activationToken() quickshell-mirror/quickshell#1353 is rebased on top and stays a draft.
• gtk: pass an activation token to the OpenURI portal ghostty-org/ghostty#14787 merged and ships in Ghostty 1.4.0. Pass Ghostty's click token on to the link opener omarchy-pkgs#1374 covers 1.3.1 until then.
• Pass the activation token on to the browser #16079 was reviewed and marked ready.
• Send an xdg-activation token with quickshell notification actions omarchy-pkgs#1367 still carries the older quickshell patch. I'll refresh it to match Switch Between 2 Workspace per monitor #1351 once that approach is accepted upstream. It works as is, so the merge order stays the same.
• No change on Activation tokens with old serials still steal focus after #16427 hyprwm/Hyprland#16608, wayland: set the input serial on activation tokens rust-windowing/winit#4782 / Bash Prompt Erasing Output #4783, or the other omarchy-pkgs PRs.

[dmkenney]

Update:

• Pass the activation token on to the browser #16079: added the Edge handoff fix from Edge drops the first two arguments of a browser handoff that has more than one #16428 (Edge was opening a blank tab when handed a token), and the browser tests now skip when a system flags file is installed. omarchybot also brought in @vitally's Edge test from Send Edge the two fields its handoff expects beyond one argument #16433. Re-reviewed on 11/01 and 11/02 with nothing to change.
• Raise a running app the launcher is asked to open #16097: the launcher now waits for the token to arrive before launching, instead of blocking on it. Rebased on quattro on 11/02, dropping the test lines Remove tests that only read source text #16334 removed.
• service/notifications: send an activation token with invoked actions quickshell-mirror/quickshell#1351: the maintainer asked that the token request not block the UI thread. It no longer does: each request gets its own callback. Waiting for a re-review. core/qmlglobal: add activationToken() quickshell-mirror/quickshell#1353 is rebased on top and is still a draft.
• Send an xdg-activation token with quickshell notification actions omarchy-pkgs#1367: each token request now has its own callback, so two requests close together no longer get the same token.
• No change on Pass Ghostty's click token on to the link opener omarchy-pkgs#1374, Recommended approach for setting up multiple machines with custom setup #1375, Some applications UI are "zoomed in" #1379, Activation tokens with old serials still steal focus after #16427 hyprwm/Hyprland#16608 or wayland: set the input serial on activation tokens rust-windowing/winit#4782 / Bash Prompt Erasing Output #4783.