* Add React Grab inject button to browser toolbar
Adds a toolbar button (cursor click icon) that injects the react-grab
script (unpkg.com/react-grab/dist/index.global.js) into the current
page. Hover over React elements and Cmd+C to copy component context
(file, component name, line number) for AI agents.
Button highlights when active, resets on navigation.
* Auto-activate selection mode on React Grab inject
First click: injects the script and auto-activates selection mode via
the react-grab:init event. Subsequent clicks toggle selection mode
on/off via window.__REACT_GRAB__.toggle().
* Bridge React Grab state back to Swift via WKScriptMessageHandler
Register a cmux-bridge plugin after injecting react-grab that posts
state changes back to Swift via webkit.messageHandlers. The button
now highlights accent color only when selection mode is actually
active (not just when the script is loaded), and deactivates when
the user exits selection mode via Escape or the react-grab toolbar.
* Fetch react-grab script via URLSession to bypass CSP
Sites like vercel.com block loading external scripts via CSP headers.
Fetch the script with URLSession (not subject to page CSP), cache it,
and inject inline via evaluateJavaScript. Also guard against duplicate
injection on repeated clicks.
* Prefetch react-grab script on first browser panel init
Kick off a low-priority background fetch of the react-grab script
when the first BrowserPanel is created. The script is cached
statically so clicking the button is instant.
* Eliminate react-grab button and callback lag
Three changes:
1. Fire-and-forget: use evaluateJavaScript with completionHandler
instead of await, so button taps return immediately.
2. Single JS payload: combine bootstrap listener + script source
into one evaluateJavaScript call (one IPC round-trip, not two).
3. Dedupe state callbacks: only post webkit message when isActive
actually changes, not on every hover/drag state update.
* Fix duplicate state callback on react-grab toggle
toggleReactGrab was sending an explicit postMessage AND the plugin's
onStateChange hook was firing too, causing two @Published updates per
toggle. Remove the explicit postMessage since the plugin hook handles
it. Also add dlog instrumentation for debugging.
* Add Cmd+Shift+G shortcut for React Grab (configurable)
- Add toggleReactGrab to KeyboardShortcutSettings with Cmd+Shift+G default
- Add View menu item with customizable shortcut
- Add command palette entry (searchable as "react grab" or "inspect element")
- Simplify button to use toggleOrInjectReactGrab, remove local state tracking
* Fix Codex review findings: pin version, verify hash, fix retry and state
1. Pin react-grab to exact version (0.1.29) with SHA-256 integrity
check. Script is verified before evaluation to prevent supply-chain
attacks via compromised CDN responses.
2. Clear prefetchTask on failure so subsequent attempts retry the
download instead of reusing a permanently failed task.
3. Remove premature isReactGrabActive=true. State is now only set
by the onStateChange message handler callback after confirmed
initialization, or explicitly reset on evaluation error.
* Extract React Grab into own file, make version configurable
Move all react-grab logic (settings, script loader, message handler,
BrowserPanel extension) into Sources/Panels/ReactGrab.swift.
Add a "React Grab Version" text field in Settings > Browser that lets
the user pin which npm version is fetched. Only versions with a known
SHA-256 integrity hash in ReactGrabSettings.knownHashes are accepted.
The cache invalidates when the configured version changes.
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* Add hover background to split action buttons
Split buttons (terminal, browser, split right/down) now show a subtle
rounded-rect background highlight on hover. Matches standard macOS
toolbar button behavior.
* Prevent fade overlays from bleeding into bottom separator
* Render bottom separator above fade overlays to prevent bleed
* Exclude drop zone from scroll fade threshold
* Revert button hover, fix fade threshold with 32pt buffer
* Fix right fade threshold: subtract drop zone, 4pt tolerance
* Add leading padding to split buttons, fix fade threshold
* Rework tab bar: full-width scroll with floating split buttons
* Use ultraThinMaterial blur for floating split buttons
* Make split buttons group full height
* Full-height split button blur background
* Inset split buttons from bottom separator
* Fix blur overlapping separator
* Use matching tab bar bg for split buttons, clear separator
* Use regularMaterial blur for split buttons
* Try thickMaterial for split buttons
* Fade gradient + solid barFill for floating split buttons
* Use extended right fade as split buttons backdrop
* Add 5 debug styles for split button background
* Add Split Button Style debug window to Debug menu
* Clean up: no-bg floating buttons, alphabetical debug menu
- Split buttons float with no background (tabBarBackground covers the
area, scroll padding prevents tabs from appearing behind buttons)
- Default splitButtonsWidth to 120 so first render has correct padding
- Remove split button style debug window and debug styles
- Alphabetize Debug Windows menu entries, remove dividers
* Revert to HStack sibling layout, add debug menu docs to CLAUDE.md
- Split buttons are HStack siblings of the ScrollView, not overlays.
Single .background() on parent, no compositing mismatch.
- Alphabetize Debug Windows menu, remove dividers.
- Document Debug menu in CLAUDE.md.
* Add Split Button Layout debug window with 5 switchable approaches
* Fix fade gradient color to match tab bar background
* Add fade color debug window with 6 color options
* Use mask for scroll fades, fixes color mismatch
* Hide scroll fades in minimal mode unless hovering
* Remove split buttons from layout when hidden in minimal mode
* Always show fades, overlay buttons to prevent scroll jump
* Add hover-only mask fade behind split buttons
* Reduce button mask area to 90pt
* Animate button mask smoothly
* Fade entire button group together via opacity
* Add blur behind buttons with fade mask
* Use theme barBackground for button backdrop
* Blur + theme tint for button backdrop
* More tint (0.85), less blur
* Tint 0.2, clear bottom border
* Use terminal bg color, add scroll trailing padding for buttons
* Less blur, paneBackground at 0.75 opacity
* paneBackground at 0.9 opacity
* 0.97 opacity for button backdrop
* Test: fully opaque paneBackground
* Test: solid red backdrop
* Gradient + solid paneBackground backdrop, no mask
* Force opaque paneBackground for button backdrop
* Use barBackground for button backdrop
* Use terminal bg (paneBackground forced opaque)
* Pre-composite backdrop color for exact match
* Add 6 switchable backdrop styles in debug window
* Mask-based button area hiding, no backdrop color needed
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
Three issues caused the Bonsplit horizontal tab bar to be hidden
when entering fullscreen with minimal mode enabled:
1. ignoresSafeArea(.container, edges: .top) was applied unconditionally
in minimal mode, pushing content behind the fullscreen menu bar area.
Now gated on !isFullScreen.
2. effectiveTitlebarPadding returned -titlebarPadding in minimal mode
regardless of fullscreen state. In fullscreen there is no native
titlebar to compensate for, so the negative offset pushed content
off the top of the screen. Now returns 0 in fullscreen.
3. Traffic light leading inset (80px) was applied in fullscreen minimal
mode even though there are no traffic light buttons. Now gated on
!isFullScreen, and syncTrafficLightInset is called on fullscreen
enter/exit.
Closes https://github.com/manaflow-ai/cmux/issues/2317
Based on https://github.com/manaflow-ai/cmux/pull/2341
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* Add cmd-click fallback for bare filenames in terminal output
When cmd-clicking text that ghostty's built-in URL/path regex doesn't
match (e.g. bare filenames from `ls` like README.md, src, config.json),
fall back to checking if the word under cursor is a valid file or
directory in the terminal panel's CWD. Uses the existing
ghostty_surface_quicklook_word API to extract the word, then resolves
it against the panel's working directory and opens it if it exists.
* Add pointing-hand cursor on Cmd-hover over bare filenames
When holding Cmd and hovering over a word that resolves to an existing
file/directory in the terminal's CWD, show the pointing-hand cursor.
Hooks into mouseMoved and flagsChanged so the cursor updates both when
moving the mouse with Cmd held and when pressing/releasing Cmd while
the mouse is stationary.
* Address PR review comments
- Refresh ghostty mouse position before quicklook_word in mouseUp and
flagsChanged so stale coordinates don't resolve the wrong word
- Use failable String(bytes:encoding:.utf8) instead of lossy decoding
- Skip absolute-path words (already handled by ghostty's regex)
- Guard against remote terminal sessions (local fileExists would be wrong)
- Use invalidateCursorRects instead of forcing iBeam on hover deactivation
to avoid overwriting ghostty/AppKit's cursor state
* Add preferred editor setting for cmd-click file opens
New "Open Files With" picker in Settings > App lets users choose which
editor opens when cmd-clicking bare filenames. Options: System Default,
Cursor, VS Code, Windsurf, Zed, Sublime Text, Xcode. Reuses the
existing TerminalDirectoryOpenTarget app detection infrastructure.
Defaults to system default (NSWorkspace default handler).
* Replace editor picker with free-form command field, respect $VISUAL/$EDITOR
The "Open Files With" setting is now a text field where users can type
any command (code, zed, subl, open -a Xcode, etc.). Resolution order:
1. User-configured command from settings
2. $VISUAL environment variable
3. $EDITOR environment variable
4. System default (NSWorkspace)
Removes the fixed PreferredEditor enum in favor of flexibility.
* Fix stuck pointing-hand cursor using NSCursor push/pop
invalidateCursorRects did nothing since the view has no cursor rects.
Use NSCursor push/pop stack instead so the previous cursor is properly
restored when the hover deactivates.
* Remove $VISUAL/$EDITOR fallback, use system default when empty
$EDITOR/$VISUAL are typically terminal editors (vim, nano) that can't
launch as GUI subprocesses. Empty field now falls back to system default
(opens in Finder/default app) which is the expected behavior.
* Address PR review comments (round 2)
- Use broader CWD fallback chain (panelDirectories → requestedWorkingDirectory
→ workspace currentDirectory) matching Workspace split creation logic
- Pop cursor stack in viewDidMoveToWindow to balance push if view is removed
while hover is active
- Reset preferredEditorCommand in resetAllSettings()
- Fall back to NSWorkspace.open when the custom editor command exits non-zero
(e.g. command not found exits 127 but /bin/sh itself succeeds)
* Clear cursor on mouse exit to prevent stuck pointing-hand
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
- Add 5-minute per-host cooldown for remote error notifications
- Add exponential backoff (capped at 60s) to proxy broker and session controller retries
- Add default SSH ConnectTimeout/ServerAliveInterval/ServerAliveCountMax to detect dead connections faster
- Fix error status clearing to only reset on actual .connected state
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(workspace): defer layout follow-up flush to avoid re-entrant displayIfNeeded crash
beginEventDrivenLayoutFollowUp() ended with a synchronous call to
attemptEventDrivenLayoutFollowUp(), which calls flushWorkspaceWindowLayouts()
→ window.contentView?.displayIfNeeded(). This is fine when invoked from
user-event handlers, but splitTabBar(_:didChangeGeometry:) fires from inside
SwiftUI's .onChange(of: geometry) during an active AppKit display/layout pass.
Calling displayIfNeeded() re-entrantly during that pass caused AppKit to
increment the per-window Update Constraints pass counter on every display
cycle. Once the counter exceeded the view-count limit AppKit threw an
NSGenericException and crashed:
'The window has been marked as needing another Update Constraints in Window
pass, but it has already had more Update Constraints in Window passes than
there are views in the window.'
Fix: replace the direct attemptEventDrivenLayoutFollowUp() call with
scheduleLayoutFollowUpAttempt(), which defers via asyncAfter(.now() + 0).
When layoutFollowUpStalledAttemptCount == 0 the backoff delay is zero, so
there is no meaningful latency increase — the flush simply runs at the start
of the next run loop iteration, after the current layout pass has fully
unwound. The NSWindow.didUpdateNotification observer and the existing timeout
still drive retries, so convergence is unaffected.
Made-with: Bunny
* fix(workspace): supersede stale layout follow-up retry on reset
scheduleLayoutFollowUpAttempt() is a no-op when
layoutFollowUpAttemptScheduled is true, so a pending retry with a
long backoff delay would survive a beginEventDrivenLayoutFollowUp()
call even though that call resets layoutFollowUpStalledAttemptCount
to 0. The stale closure would then fire after its original delay
rather than immediately.
Adds a layoutFollowUpAttemptVersion counter. beginEventDrivenLayoutFollowUp()
increments the version and clears layoutFollowUpAttemptScheduled,
allowing a fresh asyncAfter(0) attempt to be enqueued. Pending
closures capture the version at scheduling time and exit early if it
no longer matches. clearLayoutFollowUp() also increments the version
to cancel any in-flight closure during teardown.
Made-with: Bunny
* Add "Match Terminal Background" sidebar setting
Adds a toggle in Settings > Sidebar Appearance that makes the sidebar
use the same background color and transparency as the terminal area.
Uses layer-level opacity on a fully opaque background color (the same
technique as TitlebarLayerBackground) with effective opacity formula
`1 - (1-alpha)^2` to account for the terminal's two stacked
semi-transparent layers (Bonsplit chrome + Ghostty Metal surface).
Also adds a 1px trailing border derived from the terminal chrome color,
matching the bonsplit tab bar separator logic.
* Fix sidebar border color not updating on theme change
Add @State + .onReceive(.ghosttyDefaultBackgroundDidChange) to
SidebarTrailingBorder so the separator color recomputes when the
Ghostty theme changes, matching the pattern used in SidebarBackdrop.
* Address review comments: localize debug toggle, fix separator refresh
- Localize the debug panel toggle label (Codex P1)
- Add .onAppear to SidebarTrailingBorder for initial color (Cubic P2)
- Fix stale doc comment on SidebarTerminalBackgroundView (Cubic P3)
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* Skip quit confirmation for tagged DEV builds
Tagged DEV builds are ephemeral dev iterations, so the "Quit cmux?"
dialog just adds friction. Check SocketControlSettings.launchTag() in
both applicationShouldTerminate and handleQuitShortcutWarning to bypass
the confirmation when a tag is present.
Closes https://github.com/manaflow-ai/cmux/issues/2286
* Use bundle ID instead of env var for tagged DEV detection
CMUX_TAG env var is only set when reload.sh --launch opens the app.
When the user cmd-clicks the app path, it launches via Finder without
the env var, so launchTag() returns nil and the quit dialog still shows.
Switch to checking the bundle identifier (com.cmuxterm.app.debug.<tag>)
which is baked into the built app and available regardless of how it was
launched. Add SocketControlSettings.isTaggedDevBuild() helper.
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* perf: coalesce high-frequency scrollbar updates to reduce main-thread pressure
During bulk terminal output (e.g. `seq 1 100000`), GHOSTTY_ACTION_SCROLLBAR
fires thousands of times per second. Previously each callback enqueued a
separate DispatchQueue.main.async block that updated the scrollbar property
and posted a NotificationCenter notification, causing the main thread to
process thousands of redundant scroll-geometry recalculations.
This change adds a lightweight coalescing layer: the action callback stores
the latest scrollbar value behind an NSLock and schedules at most one async
flush. The flush picks up whichever value is current at execution time,
collapsing N callbacks into a single synchronizeScrollView() pass.
Measured improvement on `time seq 1 10000`:
- Before: ~0.052s (26% CPU — seq blocked on PTY backpressure)
- After: expected ~0.025-0.030s (reduced main-thread contention)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* refactor: use defer for NSLock release in scrollbar coalescing
Address review feedback: wrap unlock() in defer blocks in both
enqueueScrollbarUpdate and flushPendingScrollbar to guarantee
lock release on any future early-return or exception path.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* perf: coalesce wakeup→tick dispatches to eliminate main-thread queue flooding
During bulk terminal output, Ghostty's I/O thread fires wakeup_cb thousands
of times per second. Previously each wakeup enqueued a separate
DispatchQueue.main.async { tick() } block, flooding the main queue and
starving the run loop. The main thread spent all its time draining tick
blocks, creating PTY backpressure that blocked the writing process.
Add a lightweight coalescing gate: scheduleTick() only enqueues a single
async block; subsequent wakeups while the block is pending are no-ops.
The pending tick picks up all accumulated state in one ghostty_app_tick()
call, collapsing N wakeups into 1 main-thread dispatch.
Combined with the earlier scrollbar coalescing, measured improvement:
time seq 1 10000:
- Ghostty standalone: 0.019s (82% CPU)
- cmux before: 0.052s (26% CPU) ← main-thread saturated
- cmux after: 0.016s (62% CPU) ← faster than standalone
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: release scrollbar lock before posting notification
Move NotificationCenter.post outside the _scrollbarLock critical section
in flushPendingScrollbar(). Holding the lock through observer dispatch
would block the I/O thread's enqueueScrollbarUpdate() calls behind
main-thread observer work, recreating the backpressure this change
aims to eliminate.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix minimal mode toggle not properly updating titlebar state
UpdateTitlebarAccessoryController only re-evaluated titlebar accessories
on window focus events (didBecomeKey/didBecomeMain), not when the
presentation mode actually changed. This caused:
1. Switching to minimal: accessories weren't immediately removed
2. Switching back to standard: accessories were never re-attached
(removeAccessoryIfPresent cleared attachedWindows, but no event
triggered re-attachment)
3. Repeated toggling left the window in inconsistent states
Add a UserDefaults observer that detects presentation mode changes and
re-evaluates all windows. When switching to minimal, accessories are
removed; when switching to standard, fresh accessories are created and
attached. Also handle the fullscreen edge case where re-attached
accessories must be hidden to avoid doubling with SwiftUI overlay
controls.
* Hide window toolbar in minimal mode to eliminate titlebar gap
The NSToolbar (attached by WindowToolbarController) creates a non-zero
titlebar area even in minimal mode, leaving an empty gap above the
Bonsplit tab bar. In minimal mode there's no need for the toolbar (it
shows a "Cmd:" text that's hidden anyway with titleVisibility=.hidden).
Hide the toolbar when switching to minimal mode and restore it when
switching back to standard. Also set initial visibility on attachment
based on the current mode.
* Apply ignoresSafeArea to contentAndSidebarLayout in minimal mode
The titlebar gap persisted because intermediate SwiftUI views still
respected the window's safe area even though MainWindowHostingView
zeroes safeAreaInsets. Apply .ignoresSafeArea(.container, edges: .top)
directly to the contentAndSidebarLayout when in minimal mode so the
entire content (sidebar + terminal) extends into the titlebar area.
* Remove toolbar entirely in minimal mode instead of just hiding
toolbar.isVisible=false still reserves titlebar space. Remove the
toolbar entirely (window.toolbar=nil) when switching to minimal mode
and re-attach it when switching back to standard. Skip toolbar
attachment entirely when launching in minimal mode.
* Add BonsplitTabDragUITests.swift to cmuxUITests target
The test file existed on disk but was missing from the Xcode project,
causing all BonsplitTabDrag UI tests (including minimal mode tests) to
silently report 0 tests on CI.
* Use negative titlebar padding in minimal mode to extend content into titlebar
The native titlebar area (28.5pt for traffic lights) persists even
without a toolbar and despite .ignoresSafeArea() modifiers. Use
negative padding (-titlebarPadding) in minimal mode to pull the
terminal content up into the titlebar area. The sidebar's internal
trafficLightPadding spacer keeps sidebar content properly offset
below the traffic lights.
* Add window drag handle to Bonsplit top strip in minimal mode
In minimal mode, the custom titlebar (which provides the window drag
handle) is hidden. Add a WindowDragHandleView to the top strip overlay
so users can drag-to-move the window from the Bonsplit tab bar area.
The TitlebarDoubleClickMonitorView is kept as a background for
double-click-to-zoom.
* Use native titlebar drag in minimal mode instead of WindowDragHandleView
WindowDragHandleView defers to interactive siblings (Bonsplit tab bar),
so it never captures hits. Instead, set window.isMovable=true in
minimal mode so the native titlebar area handles drag-to-move and
double-click-to-zoom. Remove the non-functional overlay from
WorkspaceContentView.
* Enable isMovableByWindowBackground in minimal mode for window dragging
window.isMovable alone doesn't work because the Bonsplit tab bar
captures all hits before the native titlebar drag engages. Use
isMovableByWindowBackground=true so any area that doesn't handle
mouse events becomes a drag handle. Also capture
workspacePresentationMode in the WindowAccessor closure so the
window properties update when toggling modes.
* Add debug logging for minimal mode window drag diagnosis
* Intercept double-click in minimal mode tab bar to zoom instead of new tab
Bonsplit's EmptyTabBarDoubleClickMonitorView creates a new tab on
double-click in the tab bar empty space. In minimal mode, intercept
these double-clicks with a higher-priority local event monitor and
perform the standard macOS titlebar action (zoom/minimize based on
System Settings) instead. Only intercepts in the top 30pt strip and
only when minimal mode is active.
* Fix double-click monitor ordering and coordinate calculation
NSEvent local monitors are called LIFO (last installed first). Install
the minimal-mode double-click interceptor with a 0.5s delay so it's
added after Bonsplit's EmptyTabBarDoubleClickMonitorView monitors,
ensuring it runs first and can consume the event. Also fix the
distance-from-top calculation to use window frame height instead of
contentLayoutRect height, since the tab bar is in the titlebar area.
* Remove unnecessary delay from double-click monitor installation
* Show split buttons on hover only in minimal mode, fix sidebar controls re-attachment
Two fixes:
1. Add splitButtonsOnHover to BonsplitConfiguration.Appearance. In
minimal mode, the Bonsplit split buttons (terminal, browser, split
right/down) fade in only when hovering the tab bar. Revert to
always-visible when switching back to standard mode.
2. Delay titlebar accessory re-attachment when switching to standard
mode so the toolbar is re-added first. Without this, the accessory
attaches before the toolbar exists, causing the sidebar controls
to not appear in the titlebar.
* Fix splitButtonsOnHover via onChange instead of body eval, add debug logs
* Update bonsplit submodule for splitButtonsOnHover
* Remove debug logs, verified splitButtonsOnHover and accessory re-attachment on macmini
* Read presentationMode directly in TabBarView via @AppStorage
The @Observable configuration propagation wasn't reliably triggering
re-renders in TabBarView. Read the workspacePresentationMode directly
via @AppStorage in TabBarView instead, which SwiftUI reactively
updates when UserDefaults changes. Remove the syncSplitButtonsOnHover
workaround from WorkspaceContentView.
* Fix tab drag, double-click zone, and sidebar controls re-attachment
- Revert isMovableByWindowBackground to false; it breaks Bonsplit tab
reordering. Keep isMovable=true in minimal mode so the sidebar area
(which has WindowDragHandleView) is draggable.
- Increase double-click intercept zone from 30pt to 40pt to cover the
full tab bar height (33pt).
- Use asyncAfter(0.1s) for titlebar accessory re-attachment when
switching to standard mode, giving the toolbar time to re-attach.
* Add debug logging for titlebar accessory re-attachment diagnosis
* Fix crash and sidebar controls re-attachment
Remove debug logging that crashed when accessing window properties
during iteration. Increase deferred re-attachment delay to 0.3s to
give the WindowAccessor callback time to set the window identifier
and toolbar before attachIfNeeded checks isMainTerminalWindow.
* Keep titlebar accessories attached in minimal mode instead of removing
The remove/re-add cycle was fragile: re-attachment depended on window
identifiers being set, toolbar being re-added, and timing delays.
Instead, keep TitlebarControlsAccessoryViewController always attached
and let its own UserDefaults observer handle visibility. It already
hides itself (view.isHidden=true, preferredContentSize=.zero) in
minimal mode and shows itself in standard mode. No timing hacks needed.
* Force titlebar accessory layout after toolbar re-addition
* Use both self.isHidden and view.alphaValue/isHidden for accessory visibility
self.isHidden alone doesn't reliably hide the accessory when the
toolbar is nil on macOS 26. Add view.alphaValue=0 and view.isHidden
as visual fallbacks. Crucially, don't zero preferredContentSize or
frames so fittingSize returns valid values when switching back.
* Set window.isMovable=false always to fix sidebar button clicks
window.isMovable=true in minimal mode blocks clicks on the sidebar
controls because the native titlebar drag intercepts mouse events in
the overlapping area. The sidebar's WindowDragHandleView already
handles drag-to-move via performDrag with withTemporaryWindowMovableEnabled,
so native isMovable isn't needed.
* Add drag-to-move from empty bonsplit tab bar space in minimal mode
* Use overlay for tab bar drag, smart hitTest passes through tabs/buttons
* Add double-click zoom/minimize to tab bar drag view
* Add leading padding for traffic lights when sidebar collapsed in minimal mode
* Add traffic light inset to tab bar when sidebar collapsed in minimal mode
* Fix accessory space and double-click in minimal mode
- Zero preferredContentSize in minimal mode (so accessory takes no
space) but seed hostingView with cached size before querying
fittingSize when switching back (so size can be restored).
- Skip EmptyTabBarDoubleClickMonitorView in minimal mode so
DraggableTabBarView handles double-click for zoom instead.
- Remove redundant ContentView double-click monitor.
* Auto-detect traffic light inset in TabBarView via GeometryReader
Instead of propagating sidebar state through config, the tab bar
detects its own position relative to the window. If in minimal mode
and the tab bar's leading edge is near the window edge (< 20pt, no
sidebar), add 72pt spacer for traffic light clearance.
* Increase traffic light spacer to 80pt
* Fix tab click passthrough in minimal mode drag overlay
* Check full window for interactive hits in drag overlay
* Fix drag overlay capturing all clicks via reentrancy guard in hitTest
* Distinguish interactive controls from hosting views in drag hitTest
* Walk ancestor chain for button detection in drag overlay hitTest
* Replace overlay with background drag view per ensemble recommendation
* Only add traffic light inset for top-left pane
* Use GeometryReader for traffic light inset, check screen position
* Fix operator precedence in traffic light inset check
* Use window frame for traffic light inset detection
* Set tabBarLeadingInset from ContentView via onChange handlers
Replace unreliable coordinate-based detection with direct state from
ContentView, which knows both sidebar visibility and minimal mode.
Syncs on appear, sidebar toggle, and mode toggle.
* Use allPaneIds.first for top-left pane detection, no hierarchy threading needed
* Update bonsplit submodule to merged main
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* Add customizable sidebar selection highlight color
Expose a `sidebarSelectionColorHex` user default that overrides the
hardcoded blue (#0091FF) selection highlight in the sidebar. Add a
"Selection Highlight" color picker in Settings > Workspace Colors,
following the same pattern as existing tint color pickers. Falls back
to the default accent color when no custom color is set.
Closes#1753
* Fix review feedback: reactivity, reset button, localization
- Add @AppStorage subscription in TabItemView so sidebar selection
color updates reactively when changed in Settings
- Add Reset button in Settings > Workspace Colors > Selection Highlight
- Localize debug panel strings for Selection Color picker
- Clear sidebarSelectionColorHex in resetAllSettings()
* Add customizable notification badge color in sidebar
Add `sidebarNotificationBadgeColorHex` user default to override the
unread notification badge color on workspace tabs. Add a "Notification
Badge" color picker in Settings > Workspace Colors, following the same
pattern as the selection highlight picker. Falls back to the default
accent color when no custom color is set.
Previously, v2SurfaceSendText, v2SurfaceSendKey, v2SurfaceClearHistory, and
v2SurfaceReadText would silently fall back to ws.focusedPanelId when a caller
supplied a surface_id that could not be resolved (e.g. a stale ref or an ordinal
whose mapping had not yet been registered). This caused two distinct bugs:
- #2042: Commands like `cmux send --surface surface:9999` would succeed (exit 0)
and deliver input to the focused pane instead of returning an error, making
automation that targets specific surfaces unreliable.
- #2045: When the fallback landed on a browser panel, the subsequent
ws.terminalPanel(for:) check failed and returned "Surface is not a terminal",
making valid terminal surfaces appear broken when addressed by ref.
The fix adds an explicit check: if params["surface_id"] is present but
v2UUID() returns nil (resolution failure), we immediately return a not_found
error instead of falling back to the focused pane. When surface_id is absent,
the existing focused-pane fallback is preserved for backward compatibility.
Fixes#2042, Fixes#2045
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Support modifier+key combinations in send-key (ctrl+enter, shift+tab, etc.)
The send-key command only supported a few hardcoded ctrl+letter combos
and bare special keys. Generic modifier combinations like ctrl+enter
(needed for GitHub Copilot CLI submission) returned "Unknown key".
Now the parser splits on + and - separators, accumulates modifier flags
(ctrl, shift, alt/opt, cmd/super), and resolves the base key via a new
keycodeForNamedKey helper that maps named keys (enter, tab, escape,
backspace, space, arrow keys) to virtual keycodes.
Closes#1990
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix delete key mapping and filter empty parts in modifier parser
- Separate "delete" (forward delete, kVK_ForwardDelete) from "backspace"
(kVK_Delete) in keycodeForNamedKey. The original mapping had both
pointing to kVK_Delete (which is actually Backspace on macOS).
- Filter empty strings from split results to reject malformed inputs
like "+shift+a" that would produce an empty modifier part.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix single named keys silently failing and remove force unwrap in send-key
Single named keys like "space", "up", "delete" were rejected by
guard parts.count >= 2 before reaching keycodeForNamedKey(). Now
standalone named keys are handled before the modifier-combo path.
Also replaced parts.last! with guard let for safe unwrapping.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Fix nightly SSH remote daemon checksum mismatch
Each nightly build overwrites the shared cmuxd-remote-* assets on the
nightly release, but older nightly DMGs have manifests with checksums
from their build time. When a user's nightly is even one build behind,
the downloaded binary doesn't match their embedded manifest.
Two-layer fix:
1. CI: version nightly remote daemon asset names with the build number
(e.g. cmuxd-remote-darwin-arm64-2362248028801) so each nightly's
manifest points to immutable files. Unsuffixed "latest" copies are
still uploaded for tooling compatibility.
2. Client: on checksum mismatch, fetch the live manifest from the
release and verify against that. This handles users on older
nightlies that predate the CI fix.
Fixes https://github.com/manaflow-ai/cmux/issues/1745
* Fix unsuffixed checksums file to use generic filenames
Regenerate cmuxd-remote-checksums.txt from the unsuffixed alias
binaries so `shasum -c` works against the generic asset names.
Also document that unsuffixed manifest intentionally keeps versioned
downloadURLs and that aliases don't carry attestation.
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
Port links were reusing the PR-link preference
(openSidebarPullRequestLinksInCmuxBrowser), causing inconsistent
behavior when users toggled that setting. Adds a dedicated
openSidebarPortLinksInCmuxBrowser setting with its own toggle in
Settings so port and PR link behavior can be controlled independently.
Addresses review feedback from https://github.com/manaflow-ai/cmux/pull/1844
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* Add `cmux omo` command for OpenCode + oh-my-openagent integration
Same pattern as `cmux claude-teams`: creates a tmux shim so
oh-my-openagent's TmuxSessionManager spawns agents as native cmux
splits instead of tmux panes. Sets TMUX/TMUX_PANE env vars, prepends
shim to PATH, and execs into opencode.
Closes https://github.com/manaflow-ai/cmux/issues/2085
* Auto-install oh-my-opencode plugin when running cmux omo
Before launching opencode, cmux omo now:
- Checks if oh-my-opencode is registered in ~/.config/opencode/opencode.json
- If not, creates/updates the config with the plugin entry
- Checks if the npm package is installed in node_modules
- If not, runs bun add (or npm install) to install it
- Then proceeds with tmux shim setup and exec
* Use shadow config dir to avoid modifying user's opencode setup
Instead of writing directly to ~/.config/opencode/opencode.json,
cmux omo now creates a shadow config at ~/.cmuxterm/omo-config/ that
layers oh-my-opencode on top of the user's existing config. Symlinks
node_modules, package.json, bun.lock, and plugin config from the
original dir. Sets OPENCODE_CONFIG_DIR to the shadow directory.
Running plain `opencode` remains unaffected.
* Add Agent Integrations docs section with Claude Code Teams and oh-my-opencode pages
Adds sectioned sidebar navigation to the docs site. The new Agent
Integrations section contains separate pages for cmux claude-teams and
cmux omo, documenting usage, tmux shim mechanics, directory layout,
environment variables, and the shadow config approach. Both pages
include a nightly-only warning. Full English and Japanese translations,
nav item keys added to all 19 locales.
* Remove uppercase from sidebar section headers
* Add more spacing above and below sidebar section headers
* Enable tmux mode in oh-my-opencode config, improve docs
- cmux omo now writes tmux.enabled=true to the shadow oh-my-opencode.json
config. Without this, oh-my-openagent's TmuxSessionManager won't spawn
visual panes even though $TMUX is set (the config defaults to false).
- Nightly warnings now link to /nightly instead of generic text.
- Added "What you get" section to oh-my-opencode docs explaining the
visual pane behavior (auto-layout, idle cleanup, queueing).
- Added tmux.enabled step to first-run and how-it-works sections.
* Add terminal-notifier shim to route oh-my-openagent notifications to cmux
oh-my-openagent sends macOS notifications via terminal-notifier
(args: -title <t> -message <m> [-activate <id>]). The shim in
~/.cmuxterm/omo-bin/terminal-notifier intercepts these calls and
routes them through cmux notify, so notifications appear in cmux's
sidebar panel instead of as raw macOS notifications.
* Add pane geometry to tmux-compat for oh-my-openagent grid planning
oh-my-openagent's TmuxSessionManager needs pane geometry (columns,
rows, position, window dimensions) to decide where to spawn agent
panes. Without this data, agents run headlessly.
Server side:
- pane.list v2 response now includes pixel_frame, cell_size, columns,
rows per pane, plus container_frame at the top level
- Uses BonsplitController.layoutSnapshot() for pixel geometry and
ghostty_surface_size() for terminal grid dimensions
CLI side:
- tmuxEnrichContextWithGeometry() computes character-cell positions
from pixel frames and cell dimensions for tmux format variables
(pane_width, pane_height, pane_left, pane_top, pane_active,
window_width, window_height)
- list-panes now resolves pane targets (%uuid) via tmuxResolvePaneTarget
instead of failing with "Workspace not found"
- display-message enriched with geometry for format strings like
#{pane_width},#{window_width}
- tmux -V now returns "tmux 3.4" (needed by oh-my-openagent's
tmux-path-resolver verification)
* Add socket tests for tmux-compat pane geometry
6 tests verifying the geometry enrichment works end-to-end:
- pane.list returns pixel_frame, columns, rows, cell_size, container_frame
- tmux -V returns version string
- list-panes -F renders geometry format variables as integers
- list-panes -t %<uuid> resolves pane targets
- display -p renders pane_width and window_width
- After split, two panes have different positions and halved widths
All 6 pass on macmini (cmux-macmini).
* Handle tmux -V in shim script directly (no socket needed)
oh-my-openagent's tmux-path-resolver runs tmux -V to verify the binary
works. The __tmux-compat handler requires a socket connection, which
may not be established at verification time. Handle -V in the bash
shim directly to avoid the socket dependency.
* Lower default tmux pane min widths for cmux omo
oh-my-openagent defaults: main_pane_min_width=120, agent_pane_min_width=40,
requiring 161+ columns. Most terminal windows are narrower, causing
decideSpawnActions to return canSpawn=false and defer agents forever.
cmux omo now sets: main_pane_min_width=60, agent_pane_min_width=30,
main_pane_size=50, requiring only 91 columns. Also moved tmux -V
handling into the bash shim to avoid needing a socket connection for
the version check.
* Resolve merge conflicts with main (main-vertical layout, focus param)
- Keep upstream main-vertical layout anchoring from #2119
- Keep upstream focus param (v2Bool) instead of no_focus
- Combine with our -d flag handling: -d sets focus=false
- Include customCommands nav item from main
* Implement select-layout equalize and resize-pane absolute width
When oh-my-openagent spawns agent panes, it calls select-layout
main-vertical after each split to redistribute panes evenly, then
resize-pane -x <columns> to set the main pane width. Both were
previously no-ops, causing cascading uneven splits.
Server side:
- Add workspace.equalize_splits v2 API that calls the existing
TabManager.equalizeSplits (sets all dividers to 0.5)
CLI side:
- select-layout now calls workspace.equalize_splits before tracking
main-vertical state
- resize-pane -x <columns> without directional flags now computes
the pixel delta from current to desired width and resizes accordingly
* Fix equalize to use proportional divider positions
The previous equalize set all dividers to 0.5, which in a right-
recursive binary tree (from successive splits) gives 50/25/12.5/6.25%
instead of equal sizes.
New algorithm counts leaf panes on each side of each split and sets
the divider to N_left / (N_left + N_right). For 5 panes in a chain:
1/5, 1/4, 1/3, 1/2, giving each pane exactly 20%.
* Fix select-layout main-vertical to only equalize vertical splits
The proportional equalize was treating the top-level horizontal split
(main vs agent column) the same as vertical splits, setting the main
pane to 1/6 of the window with 5 agents.
For main-vertical layout, only equalize vertical splits (the agent
column), leaving the horizontal main/agent divider untouched. The
subsequent resize-pane -x handles the main pane width.
workspace.equalize_splits now accepts an optional orientation filter
("vertical" or "horizontal") to scope which splits get equalized.
* Re-equalize agent column after kill-pane
* Address PR review comments
- Fix cmux omo --help: remove omo from the help-bypass guard so
--help shows usage text instead of trying to launch opencode
- Don't overwrite unreadable opencode.json: fail with an error
instead of silently resetting to empty config
- Drain installer pipes concurrently before waitUntilExit to
prevent deadlock from full pipe buffers during bun/npm install
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* feat(sidebar): make listening ports clickable to open in browser
Wrap each sidebar port in a Button that opens http://localhost:{port}
in the cmux built-in browser (or system browser as fallback), matching
the existing PR link click behavior.
Fixes#1602
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: address bot review feedback on port clickability
- Localize port label text with String(localized:) instead of bare literal
- Add sidebar.port.label and sidebar.port.openTooltip keys to
Localizable.xcstrings with English and Japanese translations
- Respect openSidebarPullRequestLinksInCmuxBrowser user preference in
openPortLink, matching the openPullRequestLink pattern exactly
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: keyboard shortcuts not working with Russian and other non-Latin layouts
When a non-Latin input source (Russian, etc.) is active, event characters
are non-ASCII. The ANSI keyCode fallback was blocked when the layout-based
translation resolved a character, leaving no safety net. Now the keyCode
fallback is always available for non-Latin layouts, matching the physical
key position — similar to Ghostty's `physical:` keybinding behavior.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* test: add unit tests for Russian keyboard layout shortcut matching
Two tests for Cmd+T with non-Latin (Russian) keyboard layout:
1. Layout provider returns "t" (normal ASCII fallback) — verifies
the layout-based matching path works with Cyrillic event chars.
2. Layout provider returns nil (translation failure) — verifies the
ANSI keyCode fallback catches the shortcut by physical key position.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: scope ANSI keyCode fallback to non-ASCII events, add Russian to comments
Address review feedback:
- Split !hasUsableEventChars into two precise conditions:
(hasEventChars && !eventCharsAreASCII) for non-Latin layouts, and
(!hasEventChars && layoutCharacter empty) for synthetic/empty-char events.
This prevents unintended keyCode fallback on Dvorak/Colemak with empty
synthetic events.
- Add "Russian" to the non-Latin layout list in the guard comment at line 10626.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Allow naming a workspace at creation time instead of requiring a
separate rename-workspace call afterward.
Threads a title parameter through:
- CLI: --name <title> flag parsed and sent as "title" in v2 params
- V2 handler: extracts title, passes to TabManager.addWorkspace()
- TabManager: uses provided title instead of auto-generated
"Terminal N" and calls setCustomTitle() to persist it
- V1 handler: accepts optional name argument
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Previously, GHOSTTY_ACTION_PWD used AppDelegate.shared?.tabManager to
update the current working directory, which only returns the key window's
TabManager. In multi-window scenarios, cwd updates from non-key windows
were written to the wrong TabManager, causing panelDirectories to be
empty at save time and falling back to $HOME on restore.
Fix by using tabManagerFor(tabId:) to precisely route the update to
whichever window owns the tab, ensuring all windows persist their
working directories correctly.
Fixes#2125
The Korean CJK font mapping added in PR #1017 was removed in PR #1700,
but the koreanRanges static property was left behind as dead code.
Related: #1700, #1693
applicationShouldTerminate was returning .terminateNow unconditionally,
bypassing QuitWarningSettings. Now it shows the same confirmation alert
used by handleQuitShortcutWarning, with an isQuitWarningConfirmed flag
to prevent a double dialog when the Cmd+Q shortcut path already confirmed.
Fixes#2139
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix claude-teams pane anchoring: main-vertical layout + focus
Claude's agent teams sends `split-window -h` for each teammate then
`select-layout main-vertical` to stack them vertically. Three fixes:
1. Implement select-layout main-vertical: track layout state in the
tmux compat store so subsequent horizontal splits of the leader
pane get redirected to vertical splits of the right-side column.
2. Pass focus:false to surface.split from the split-window handler
so internal bonsplit focus stays on the leader pane.
3. Fix tmuxCompatStoreURL to respect $HOME env var (was using
NSString.expandingTildeInPath which ignores $HOME).
Closes https://github.com/manaflow-ai/cmux/issues/2118
* Fix claude-teams split routing: auto-seed main-vertical on first right split
The split-window routing was broken in two ways:
1. After the first teammate split (right), the main-vertical state wasn't
being created, so all subsequent splits also went right instead of
stacking down in the right column.
2. The old code only redirected splits when main-vertical was already active
AND the target was the leader surface. Claude's teams protocol targets
arbitrary panes from list-panes, not necessarily the leader.
Now the first teammate split goes right (creating the column), auto-seeds
the main-vertical state, and all subsequent splits stack downward. The
caller's surface (CMUX_SURFACE_ID) is used as the anchor regardless of
which pane Claude targets.
Also adds caller surface preference in pane target resolution so the
caller's exact surface is used when the target pane matches, preventing
stale selected-surface references after tab switches.
---------
Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com>
* test: reproduce Cmd+N snapshot workspace lifetime race
* fix: retain snapshot workspaces through Cmd+N creation
* fix: repair workspace lifetime regression test
* fix: extract workspace config through self to avoid Xcode 16.x ARC crash
The snapshot approach (c1998e34) navigated workspace → panel → surface
through local variables. Xcode 16.4's -O ARC optimizer aggressively
elides retains on these locals through inlined call chains, causing
use-after-free on every Cmd+N in CI-built nightlies.
Fix: extract preferredWorkingDirectory and inheritedTerminalFontPoints
through self (always retained) BEFORE capturing locals. The snapshot
is now purely value-typed with no Workspace references held in locals.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
