Repository navigation
Fix agent resume when saved cwd is deleted - #4859
Merged
Merged
Conversation
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughAdd a quoted, existence-checked working-directory prefix (guarded ChangesSafe Working Directory Handling
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Poem
Caution Pre-merge checks failedPlease resolve all errors before merging. Addressing warnings is optional.
❌ Failed checks (1 error, 1 warning)
✅ Passed checks (16 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Greptile SummaryThis PR replaces hard
Confidence Score: 4/5Safe to merge for most restore paths; the tmux-compat startup script still hard-exits on a deleted saved directory, leaving that one path unrepaired despite the PR's stated goal. All major resume paths now correctly use the guarded cd form and are well-tested. However, CMUXCLI+TmuxCompatSupport.tmuxStartupScript (line 207) was only refactored to extract a local variable — the cd -- QUOTED || exit $? body is unchanged. A user in tmux-compat mode whose saved working directory was deleted will still get a hard shell exit before the agent command runs. This was flagged in the previous review and remains unresolved. CLI/CMUXCLI+TmuxCompatSupport.swift — tmuxStartupScript at line 207 still uses the non-guarded cd form. Important Files Changed
Reviews (9): Last reviewed commit: "chore: remove unused shell word helper" | Re-trigger Greptile |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@CLI/CMUXCLI`+TmuxCompatSupport.swift:
- Around line 49-52: Replace the two-step existence check plus cd with the same
race-safe fallback used elsewhere: instead of appending "{ [ ! -d \(quotedCwd) ]
|| cd -- \(quotedCwd); }" build a single attempt-first cd with a safe fallback
(the same snippet used in the other tmux path), e.g. try "cd -- \(quotedCwd)
2>/dev/null || { [ -d \(quotedCwd) ] && cd -- \(quotedCwd); }" so that the code
in CMUXCLI+TmuxCompatSupport.swift uses tmuxShellQuote/resolvePath and performs
a race-safe cwd change for both tmux paths.
In `@Sources/RestorableAgentSession.swift`:
- Around line 35-40: The existing guard in
TerminalStartupWorkingDirectoryPrefix.optionalChangeDirectoryPrefix uses "[ ! -d
\(quoted) ] || cd -- \(quoted);" which can race if the directory is removed
between the test and cd; change it to attempt cd directly and swallow failures
so the command group always succeeds: replace the "[ ! -d ... ] || cd -- ..."
pattern with "cd -- \(quoted) 2>/dev/null || true" (still inside the same group
and returned string) so a transient deletion doesn't make the group return
non-zero and skip the resume command.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 922ec202-7bda-455c-a8db-eff7826fe076
📒 Files selected for processing (13)
CLI/CMUXCLI+TmuxCompatSupport.swiftCLI/cmux.swiftSources/RestorableAgentSession.swiftSources/SessionPersistence.swiftSources/SessionRestoredTerminalCommandStore.swiftcmuxTests/AgentHibernationTests.swiftcmuxTests/CLINotifyProcessIntegrationRegressionTests.swiftcmuxTests/PiVaultAgentPersistenceTests.swiftcmuxTests/RestorableAgentHookProviderHermesTests.swiftcmuxTests/RestorableAgentHookProviderResumeTests.swiftcmuxTests/RovoDevSessionIndexTests.swiftcmuxTests/SessionPersistenceTests.swiftcmuxTests/WorkspaceUnitTests.swift
lawrencecchen
dismissed
coderabbitai[bot]’s stale review
May 27, 2026 07:00
Stale CodeRabbit review. Addressed by 05146ab; current CodeRabbit check passes.
There was a problem hiding this comment.
Actionable comments posted: 1
♻️ Duplicate comments (1)
CLI/CMUXCLI+TmuxCompatSupport.swift (1)
49-52:⚠️ Potential issue | 🟠 Major | ⚡ Quick winRestore the guarded-cwd fallback in both tmux startup paths.
Both paths now hard-fail on missing/deleted cwd (
cd -- ...), so resume/startup input can still be skipped when the saved directory no longer exists.Patch
- pieces.append("cd -- \(quotedCwd)") + pieces.append("{ cd -- \(quotedCwd) 2>/dev/null || [ ! -d \(quotedCwd) ]; }") ... - lines.append("cd -- \(quotedCwd) || exit $?") + lines.append("{ cd -- \(quotedCwd) 2>/dev/null || [ ! -d \(quotedCwd) ]; } || exit $?")Also applies to: 205-208
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@CLI/CMUXCLI`+TmuxCompatSupport.swift around lines 49 - 52, Restore a guarded-cwd fallback instead of directly appending "cd -- …" so startup/resume won't fail when the saved directory is missing: where you currently build the tmux startup pieces (the block using trimmedCwd, tmuxShellQuote(resolvePath(trimmedCwd)) and pieces.append("cd -- …")), change the appended command to guard the cd with a directory existence check (e.g. append a shell fragment like `if [ -d "<quotedCwd>" ]; then cd -- "<quotedCwd>"; fi` or `test -d "<quotedCwd>" && cd -- "<quotedCwd>"`) and make the same change in the other tmux-startup path referenced around the 205-208 block so both code paths use the guarded cd.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@Sources/RestorableAgentSession.swift`:
- Around line 89-100: The current strippedSavedWorkingDirectoryOptions(...)
function reparses the command into argv and then reserializes tokens with
TerminalStartupShellQuoting.singleQuoted, which incorrectly turns shell
operators into literal arguments; instead, detect the tokens removed by
AgentLaunchSanitizer.removingSavedWorkingDirectoryOptions (or by comparing the
words from shellWords(command)), determine the byte/character ranges those
tokens occupy in the original command string, and produce the cleaned command by
removing those ranges from the original command (preserving surrounding
whitespace and operators). Update strippedSavedWorkingDirectoryOptions and the
similar code block at 114-159 to use this range-based splice approach rather
than token-by-token single-quoting, keeping shell syntax intact while removing
only the saved-cwd option substrings.
---
Duplicate comments:
In `@CLI/CMUXCLI`+TmuxCompatSupport.swift:
- Around line 49-52: Restore a guarded-cwd fallback instead of directly
appending "cd -- …" so startup/resume won't fail when the saved directory is
missing: where you currently build the tmux startup pieces (the block using
trimmedCwd, tmuxShellQuote(resolvePath(trimmedCwd)) and pieces.append("cd --
…")), change the appended command to guard the cd with a directory existence
check (e.g. append a shell fragment like `if [ -d "<quotedCwd>" ]; then cd --
"<quotedCwd>"; fi` or `test -d "<quotedCwd>" && cd -- "<quotedCwd>"`) and make
the same change in the other tmux-startup path referenced around the 205-208
block so both code paths use the guarded cd.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: f6af830e-514e-46e9-9bcf-47061cb11fe7
📒 Files selected for processing (7)
CLI/CMUXCLI+TmuxCompatSupport.swiftCLI/cmux.swiftPackages/CMUXAgentLaunch/Sources/CMUXAgentLaunch/AgentLaunchSanitizer.swiftPackages/CMUXAgentLaunch/Tests/CMUXAgentLaunchTests/AgentLaunchSanitizerTests.swiftSources/RestorableAgentSession.swiftcmuxTests/RestorableAgentHookProviderResumeTests.swiftcmuxTests/SessionPersistenceTests.swift
Comment on lines
205
to
208
| if let cwd = cwd?.trimmingCharacters(in: .whitespacesAndNewlines), !cwd.isEmpty { | ||
| lines.append("cd -- \(tmuxShellQuote(resolvePath(cwd))) || exit $?") | ||
| let quotedCwd = tmuxShellQuote(resolvePath(cwd)) | ||
| lines.append("cd -- \(quotedCwd) || exit $?") | ||
| } |
Contributor
There was a problem hiding this comment.
cd fails, so a deleted saved directory will kill the session before the agent can run. The PR description explicitly lists "tmux compatibility startup scripts" as fixed, and every other startup-script path in this PR (SessionRestoredTerminalCommandStore, Codex Teams in cmux.swift) now uses the guarded form, but this one does not. Workspace.swift now also suppresses passing requestedWorkingDirectory to Ghostty when restoredTmuxStartupScript != nil, which means neither Ghostty nor the script tolerates a missing cwd in this path.
Suggested change
| if let cwd = cwd?.trimmingCharacters(in: .whitespacesAndNewlines), !cwd.isEmpty { | |
| lines.append("cd -- \(tmuxShellQuote(resolvePath(cwd))) || exit $?") | |
| let quotedCwd = tmuxShellQuote(resolvePath(cwd)) | |
| lines.append("cd -- \(quotedCwd) || exit $?") | |
| } | |
| if let cwd = cwd?.trimmingCharacters(in: .whitespacesAndNewlines), !cwd.isEmpty { | |
| let quotedCwd = tmuxShellQuote(resolvePath(cwd)) | |
| lines.append("{ cd -- \(quotedCwd) 2>/dev/null || [ ! -d \(quotedCwd) ]; } || exit $?") | |
| } |
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit da06cd8. Configure here.
lawrencecchen
dismissed
coderabbitai[bot]’s stale review
May 27, 2026 15:50
Stale automated review on older commit. The shell reserialization finding was fixed by da06cd8/bb5c757e, and the tmux suggestion was intentionally not applied because explicit tmux -c must preserve failure semantics for missing directories.
Vangor
added a commit
to Vangor/most
that referenced
this pull request
May 27, 2026
* test: cover external dot path open * fix: open external path arguments without socket access * fix: preserve explicit socket path opens * fix: bound launchservices open helper * fix: avoid blocking primitive in open helper * fix: format localized path open errors * fix: complete path open localization * fix: scrub socket env for external path opens * fix: localize path open success output * Add cmux.xcworkspace with Packages visible alongside the project (manaflow-ai#4834) * Add cmux.xcworkspace with Packages visible alongside the project Top-level navigator shows cmux.xcodeproj plus every Packages/* SPM package as siblings, so local packages are editable in the same window without nesting under the project node. Additive only: existing -project build flows (reload.sh, CI) keep working unchanged. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Ignore xcshareddata under xcworkspaces Xcode auto-creates xcshareddata inside .xcworkspace bundles for WorkspaceSettings.xcsettings, IDEWorkspaceChecks.plist, and SwiftPM configuration. None of those carry intentional shared state for cmux today (build is driven by reload.sh and CI, no private SPM registry, no custom file headers). Ignoring them prevents Xcode bookkeeping churn in commits. Project-level xcshareddata (shared .xcscheme files) is unaffected because the pattern only matches inside .xcworkspace directories. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * fix: scrub legacy panel env for path opens * fix: limit path open locale entries * Redesign notifications popover: bigger, minimal, swipe to dismiss (manaflow-ai#4778) * Redesign notifications popover: bigger, minimal, swipe to dismiss - Larger frame (min 460x480, ideal 560x620, max 760x760) so more notifications are visible at once. - Compact rows with a thin accent bar for unread, tighter typography, hover-revealed clear button. No more boxed cards per row, just dividers between rows. - Drag a row left or right to dismiss; passing the threshold slides it out and removes the notification. Tap to open as before. - Empty/loading/populated states share the same outer frame so the popover stops resizing when notifications come and go. * Show jump-to-latest keyboard shortcut next to button label * Stabilize hover on notification rows: dedicated background tracking layer .onHover and .onTapGesture on the same SwiftUI node share AppKit's mouse-tracking pipeline and arbitrate against each other, producing flaky enter/exit events right after the popover opens and when crossing rows fast. Move hover detection to a transparent background layer using .onContinuousHover, scope the opacity animation to the background only so it does not re-animate dragOffset, and bump the hover opacity from 0.07 to 0.11 so it is visible against .windowBackgroundColor in light mode. * Use NSTrackingArea for notification row hover SwiftUI's .onHover and .onContinuousHover still arbitrate with the parent .onTapGesture in macOS popover content, leaving rows where hover never fires (notably on first popover open and when crossing between LazyVStack rows). Replace the SwiftUI hover modifier with an NSViewRepresentable that installs an NSTrackingArea (.mouseEnteredAndExited + .activeAlways + .inVisibleRect). The tracking NSView returns nil from hitTest so clicks pass through to the SwiftUI parent's tap gesture, and the view syncs current inside/outside state on tracking-area install for the case where the pointer is already inside the row when the popover opens. * Notifications popover: keep timestamp visible, drop hover animation, add right-click menu - Always show row timestamp (removed the hide-on-hover opacity). - Removed the hover background fade animation so it's an instant flip. - Right-click a row for Open / Mark as Read / Mark as Unread / Dismiss. - Added TerminalNotificationStore.markUnread(id:) sibling to markRead(id:). - Added en/ja localizations for the new menu strings. * Notifications popover: drop top margin, always semibold title * Make notifications popover resizable, taller default Default size bumped to 560x760 (was 560x620 ideal). User can drag a bottom-right resize handle to set their own size; the chosen width and height persist via @AppStorage. Bounds are 420x320 .. 1000x1200. * Notifications popover: invisible corner resize, drop swipe-to-dismiss NSPopover has no native resize chrome and there's no first-class SwiftUI resize API for it. The closest native-feeling thing is a hit-only corner region that drives a state-driven .frame: no visual button, just a resizeLeftRight cursor on hover (closest standard NSCursor to a diagonal resize). Also removed swipe-to-dismiss. The hover X button and the right-click Dismiss menu item are the dismiss surfaces; swiping rows in an NSPopover row list isn't a macOS convention. * Notifications popover: fix glitchy resize via AppKit screen coords SwiftUI's DragGesture reports translation in a local coordinate space that is literally being resized under the cursor as the popover grows. Each frame the gesture re-derives translation from a moving anchor, which produces dimension oscillation (the popover jumps between two sizes). Replace the SwiftUI gesture with an NSViewRepresentable that tracks NSEvent.mouseLocation in global screen coordinates. Screen coordinates are stable regardless of popover resize, so deltas are monotonic. Also disable implicit animation on the .frame width/height so the popover follows the cursor 1:1. * Notifications popover: use diagonal resize cursor on bottom-right corner * Notifications popover: address review feedback - unreadCount uses notificationMenuSnapshot.unreadCount instead of reducing the live list, so the badge stays consistent with the hasUnreadNotifications / hasNotifications guards that drive the Jump to Latest / Clear All disabled states. - Hide the visual keyboard-shortcut chip from accessibility; the button already exposes the shortcut via .accessibilityValue, so VoiceOver no longer reads it twice. - Add a row-level accessibility action to dismiss a notification, since the visible clear button is hover-only. Adds the localized string notifications.row.clear (en/ja). * markUnread(id:): clear manual workspace unread to avoid double-count When a user marks a read notification unread, the notification itself now provides the workspace unread indicator. If the workspace also had a manual unread flag set, the popover header, dock badge, and workspace badge would double-count that workspace because the count builders sum notification unread counts and workspace indicators. Mirrors what markLatestNotificationAsOldestUnread already does in the same file. Found by Codex review. * Restore Button wrapper on notification rows for keyboard access Replacing the original Button with .onTapGesture removed rows from the SwiftUI key-view loop, so keyboard-only users could tab to the header controls but not open a row. The original reason for dropping Button was that SwiftUI's .onHover arbitrated with the row's primary action — but hover is now driven entirely by an AppKit NSTrackingArea (HoverTrackingRepresentable), independent of the row's activation node. The Button wrapper is therefore safe again and restores space/return activation in the key-view loop. Found by Codex review. * Notifications popover: persist resize once on mouseUp, clamp drag baseline - Live resize uses @State (liveWidth/liveHeight); @AppStorage is written exactly once when the user releases the mouse, instead of on every mouseDragged event. Each @AppStorage write was hitting UserDefaults and broadcasting UserDefaults.didChangeNotification to every app-wide observer. - Drag baseline now comes from the clamped (currently displayed) size, not raw saved doubles, so a drag that starts with out-of-bounds stored values doesn't lose initial pointer travel re-entering the visible range. * Notifications popover: clear button as ZStack sibling, not nested in row Button Nesting clearButton inside the row's Button(action: onOpen) label created the classic SwiftUI nested-button hit-test problem on macOS: clicks on the inner X could be consumed by the outer row Button's tap area instead of clearing the notification. Move clearButton out of rowContent and into a ZStack(alignment: .trailing) at the body level so it's a sibling of the row Button with an independent hit target. Found by Codex review. * Notifications popover: clamp to screen, hide hover-only X from focus - Clamp the popover's frame against NSScreen.main?.visibleFrame minus an 80pt margin so the bottom-right resize handle stays reachable even when the saved size was captured on a larger display. - Mark the hover-only clear button .accessibilityHidden(!isHovering) so Full Keyboard Access / VoiceOver doesn't focus an invisible dismiss control. Dismissal is still exposed via the row's accessibility action and the right-click menu. Found by Codex. * CI: retry — GitHub Actions infra was returning 403/404 on codeload * CI: retrigger now that GitHub Actions infra is back * Mark-as-Read context menu: clear focused pane indicator too A user-initiated 'Mark as Read' should mirror the full dismissal intent, so the pane's focused-read indicator for that surface should clear alongside flipping the notification's isRead flag. Otherwise the unread count drops but the pane badge remains visible until the user later interacts with the terminal. Found by Codex review. * Notifications popover: clamp against host window's screen, fix hover sync race - Clamp popover size against the screen of NSApp.keyWindow (the popover's anchor window), not NSScreen.main. On multi-monitor setups the popover may appear on a different display than 'main'. - Remove the DispatchQueue.main.async hop in HoverTrackingNSView's updateTrackingAreas; the queued onChange(true) could land after mouseExited's synchronous onChange(false), leaving rows stuck in the hovered state. updateTrackingAreas already runs on the main thread, so the call is fine synchronously. Found by Cursor Bugbot. * Mark-as-Read: only clear focused pane indicator for pane-scoped notifications clearFocusedReadIndicator treats surfaceId == nil as 'clear any pane indicator for this tab', so passing nil for a workspace-level notification would wipe an unrelated pane's badge. Gate the call on notification.surfaceId being non-nil. Found by Codex review. * Notifications popover: add a11y label/hint to resize handle The localized notifications.resize key existed but was never wired to the actual control. Adds .accessibilityLabel/.accessibilityHint so VoiceOver and Full Keyboard Access users can discover the resize affordance. Found by Cursor Bugbot. * Notifications popover: move accessibility identifier back onto the row Button XCUITests query rows with app.buttons["NotificationPopoverRow.<id>"]. The previous pass put .accessibilityIdentifier on the combined outer ZStack, exposing the row as a container rather than a button, which breaks both XCUITest lookups and the button accessibility role for assistive tech. Move identifier + primary action + clear action back onto the inner Button. Found by Codex review. * Notifications popover: snapshot list before LazyVStack; clear restored unread on markUnread - ForEach now iterates an immutable snapshot captured outside the LazyVStack instead of reading notificationStore.notifications inside the row builder; the previous code reintroduced a store dependency below the list boundary, which CLAUDE.md flags as the same anti-pattern that caused the LazyLayoutViewCache spin-loop in the sessions panel (manaflow-ai#2586). - markUnread(id:) now also clears the restored workspace unread indicator for the tab, so a session-restored unread hint plus a user-toggled unread notification no longer double-count in the header/dock/workspace badges. Found by Codex review. --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Remove History from right sidebar (manaflow-ai#4785) * Remove History mode from right sidebar Drops the History tab from the right sidebar mode picker along with its keyboard shortcut, command palette entries, and the 'Open Full History' menu button. Recently Closed/Focused submenus and Reopen Last Closed still work from the menu bar. * Drop tests for deleted HistoryDayGrouping helper * Tolerate obsolete history mode and restore focus-history menu key - SessionRightSidebarToolPanelSnapshot now decodes an unknown mode (e.g. legacy 'history') as nil instead of failing the entire snapshot, so upgraded users can still restore prior sessions. - Restore the 'menu.history.showFullFocusHistory' key still referenced by AppDelegate+FocusHistoryContextMenu for non-sidebar focus history. * Remove duplicate @mainactor attribute left by prior edit --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Launch restored agent sessions via startup commands (manaflow-ai#4777) * Test agent restore startup command launch * Launch restored agents with startup commands * Tighten restored startup launch state * Preserve remote startup for agent restore * Relax remote startup restore assertion * Fix startup restore lifecycle * Remove unused restore launch flag * Run startup resumes in login shell * Tighten remote restore assertion * Handle csh agent restore launchers --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Wrap workspace titles in sidebar (manaflow-ai#4848) * Wrap workspace titles in sidebar * Add workspace title wrap toggle * Localize workspace title wrap setting * Align wrapped workspace title accessories * Reset workspace title wrap preference --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: cover bare relative external path open * fix: open bare relative path arguments externally * fix: preserve refresh surfaces command precedence * fix: keep bare path command matching case-sensitive * Add secure cmux navigation links (manaflow-ai#4857) * Add secure cmux navigation links * Localize cmux navigation link labels --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Fix matched sidebar terminal background (manaflow-ai#4780) * test: cover matched sidebar background tint * fix: match sidebar terminal background * fix: remove dead sidebar overlay path * test: cover matched sidebar boundary separators * fix: keep matched sidebar separators visible * test: cover matched sidebar chrome borders * test: sample sidebar boundary resizers --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Fix JSONC comment skipper for CRLF line endings (manaflow-ai#4869) * Fix JSONC // comment skipper for CRLF line endings Swift treats \r\n as a single extended grapheme cluster, so the existing source[index] != "\n" check inside the three // line-comment skippers never matches a CRLF line ending. On a file with CRLF throughout, the loop runs past the rest of the file looking for a standalone LF and strips everything after the first //. preprocess() then produces truncated JSON and loadCmuxSettingsRoot() / jsonObject() throw "Unexpected end of file". This was breaking the tmux corpus terminal-nightly job's testSurfaceResumeApprovalWritesRecordsIntoCmuxJSON, which explicitly exercises the CRLF path. Match any character whose first scalar is CR or LF so we stop at CR, LF, or CRLF correctly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Allow workflow_dispatch to run tmux corpus terminal-nightly So this workflow's macOS test job is reproducible on demand for branches under review, not only on the nightly schedule. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Fix JSONC editor indent detection for CRLF line endings indentationBeforeLine walked back to find the start of the line containing a given index by stopping at "\n" or "\r" characters, but on CRLF input Swift treats "\r\n" as a single extended grapheme cluster that equals neither. The loop then walked past every line break to the start of the file and returned an empty indent, so newly inserted properties (e.g. resumeCommands) ended up at the wrong indentation level. Reuse the isLineTerminator helper that already handles CRLF correctly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Open forked conversations without palette action probe (manaflow-ai#4852) * Open forked conversations without palette action probe * Fix fork probe snapshot selection * Refresh fork probe cache per palette session * Preserve verified fork probe visibility * Preserve fork snapshot cache during live refresh * Allow verified fallback fork execution * Reprobe fork cache after fallback clears * Preserve focus after instant fork commands * Clear fork cache on execution failure * Avoid stale fork cache on palette reopen * Clear stale fallback fork cache before reprobe * Preserve verified fork cache on palette reopen * Require verified fork snapshot before fallback execution * Track actual fork fallback usage * Preserve verified fork cache fallback flag --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: cover bonsplit tab indicator drift * fix: update bonsplit tab indicator handling * Skip Cmd+Shift key forwarding test when Ghostty surface init fails (manaflow-ai#4871) The tmux corpus terminal-nightly runner cannot initialize a Metal-backed Ghostty surface; embedded_window logs "error initializing surface err=error.OutOfMemory" for every surface the suite creates, and ghostty_surface_new returns nil. The test then fails XCTAssertTrue on performKeyEquivalentAfterMenuMiss because ensureSurfaceReadyForInput cannot return a live surface, which looks like a key-forwarding regression rather than the environment issue that it is. Skip the test (via XCTSkipUnless on TerminalSurface.hasLiveSurface) when the surface fails to initialize, so CI surfaces the real problem (Metal unavailable on this runner) without masking it as a key-bind regression. The test still runs end-to-end on developer machines and on runners with a logged-in GUI session. Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Add algorithmic complexity review rule (manaflow-ai#4866) * Add algorithmic complexity review rule * Align algorithmic complexity review hint --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Narrow AppDelegate CI quarantine (manaflow-ai#4874) Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore browser devtools config CI coverage * Add prompt and rules deeplinks (manaflow-ai#4839) * Add prompt and rules deeplinks * Preserve encoded punctuation in text deeplinks * Localize text deeplink dialogs * Support Freestyle SSH link users * Harden prompt and rules deeplinks * Suppress welcome for text deeplinks * Restore SSH deeplink trust gate * Preserve plus signs in text deeplinks * Respect no-focus for text deeplink target selection * Use deterministic text deeplink target * Start background text deeplink targets * Report text deeplink send failures * Handle mixed deeplink batches safely * Gate mixed external deeplink batches * Report failed external text deeplink sends * Stop after handling external cmux links * Allow multiline cmux text links * Reject control characters in cmux text links * Polish external cmux link handling * Label sendTextWhenReady callbacks --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Make session index backfill linear (manaflow-ai#4868) * Make session index backfill linear * Stabilize session index backfill tie breaks * Fix session index tie-break test setup --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Optimize batch workspace sidebar actions (manaflow-ai#4865) * Optimize batch workspace sidebar actions * Preserve workspace selection after drag reorder * Harden workspace batch action coverage --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: harden bonsplit scroll indicator check * Restore browser remote store CI coverage * Make browser search providers configurable (manaflow-ai#4849) * feat: make browser search providers configurable * fix: keep parsing browser settings after invalid custom search config * test: harden custom search config regression * fix: handle custom search edge cases * fix: validate custom search fallbacks * fix: localize search engine labels * fix: gate stale search suggestions * fix: allow blank custom search names * fix: add exa search provider --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore browser lifecycle CI coverage * docs: add cmux ssh deep links (manaflow-ai#4833) * docs: add cmux ssh deep links * docs: add cmux deeplink fallback pages * docs: add deeplink locale fallbacks * docs: tighten deeplink fallback validation * docs: mirror native ssh deeplink validation * docs: localize deeplink messages * docs: canonicalize rules deeplink alternates * Allow Freestyle SSH deeplink users * docs: validate text deeplink fallback params * docs: ignore blank optional deeplink params * docs: address deeplink review comments --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore file preview review CI coverage * Pin Xcode 26 (objectVersion 60) and add pbxproj normalizer + CI guard (manaflow-ai#4836) * Add deterministic normalizer for cmux.xcodeproj/project.pbxproj scripts/normalize-pbxproj.py sorts the high-churn sections (PBXBuildFile, PBXFileReference, and the files = (...) arrays inside Sources / Resources / Frameworks / CopyFiles build phases) into a deterministic order keyed on the entry comment plus UUID. The Xcode build does not care about the order of these flat dictionary sections; sorting them just kills the nondeterministic diff noise Xcode generates on every UI touch. Does not touch UUIDs, comments, or PBXGroup children = (...) arrays (navigator order is intentional). Idempotent: a second run produces zero diff. Standalone in this commit so the diff is just the script. The next commit applies the script and bumps objectVersion in one shot, so the resulting churn is contained and never repeated. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Pin objectVersion = 60 and normalize pbxproj Bumps objectVersion from 56 to 60 (the format Xcode 16+ and Xcode 26 write by default) and runs scripts/normalize-pbxproj.py once to establish the deterministic baseline. After this commit, future diffs to project.pbxproj show only real changes, not Xcode's nondeterministic section reordering. One-time large diff. No semantic changes to targets, sources, build phases, or settings: pure sort + version pin. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add tracked pre-commit hook that normalizes pbxproj scripts/git-hooks/pre-commit calls scripts/normalize-pbxproj.py on cmux.xcodeproj/project.pbxproj when it is staged and re-stages the result. scripts/install-git-hooks.sh points the clone at this directory via `git config core.hooksPath scripts/git-hooks`, and scripts/setup.sh auto-runs it so devs get the hook without a separate manual step. After this, Xcode's nondeterministic reordering of build-file and file-reference sections is canceled out at commit time. The CI guard in the next commit enforces the rule for anyone who bypasses the hook with --no-verify or who never ran setup. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add CI guard for objectVersion pin and pbxproj normalization scripts/check-pbxproj.sh asserts cmux.xcodeproj/project.pbxproj has objectVersion = 60 (Xcode 26 default) and that the file is normalized per scripts/normalize-pbxproj.py. Wired as a step in the workflow-guard-tests job so every PR is gated. This catches anyone who bypasses the pre-commit hook with --no-verify or who never ran scripts/setup.sh. The error message points at the exact fix path. To bump the pin (e.g., when the team adopts a newer Xcode major), edit EXPECTED_OBJECT_VERSION in this script and the matching line in CLAUDE.md. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add .xcode-version and document Xcode 26 pin in CLAUDE.md .xcode-version records the major (26.0) for tooling that reads it (xcodes CLI, some CI helpers). CLAUDE.md gains an Xcode toolchain section explaining the pin, the normalizer + pre-commit hook + CI guard mechanics, and the procedure for bumping the pin in the future. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Read .xcode-version as the source of truth in check-pbxproj.sh scripts/check-pbxproj.sh now reads .xcode-version and maps the Xcode major to the expected objectVersion via a one-entry case statement. Bumping the team's Xcode pin becomes a one-file edit (.xcode-version), with a script update only required when Apple actually changes objectVersion in a new Xcode major. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address CodeRabbit findings on check-pbxproj.sh and pre-commit hook scripts/check-pbxproj.sh now passes "$PBXPROJ" explicitly to normalize-pbxproj.py instead of letting it default to a path relative to the current working directory, so the guard works regardless of where CI invokes it. scripts/git-hooks/pre-commit refuses to run when the working-tree pbxproj has unstaged changes. Previously the hook would normalize the working-tree file and `git add` the result, which silently staged any unstaged hunks the user had deliberately left out of the commit. The hook now exits non-zero with a clear message telling the user to either stage the whole file or stash the unstaged hunks first. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address Greptile findings: misleading comment + bump-step docs scripts/normalize-pbxproj.py: the comment said "preserve empty lines exactly where they are" but the implementation collapses blanks to a trailing group. Reworded the comment to match the actual behavior. CLAUDE.md: the bump procedure now mentions opening cmux.xcodeproj in the new Xcode so objectVersion gets rewritten automatically. Without that step a developer following the docs alone would update only the pin file and the script case, and the CI guard would fail on their next commit. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * Add scripts/cleanup-dev-builds.sh for safely reclaiming tagged DerivedData (manaflow-ai#4837) * Add scripts/cleanup-dev-builds.sh Removes tagged dev-build artifacts produced by scripts/reload.sh: DerivedData/cmux-<tag>/ (multi-GB each), /tmp/cmux-<tag>/, the per-tag debug socket and logs, the reload log, and the App Support cmuxd dev socket. Defaults to dry-run; pass --apply to delete. Safety rules always on: - Skip the tag of any running `cmux DEV <tag>` app - Skip the tag pointed at by /tmp/cmux-last-cli-path - Skip any tag tied to a live git worktree Filters: --older-than DAYS, --keep TAG (repeatable). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Drop "worktree exists" safety rule in cleanup-dev-builds.sh Existence of a git worktree with the same name is a weak signal of active use, and HQ tends to accumulate worktrees long after the work is done. The rule made cleanup over-protective for the typical case (worktree still around from a merged or abandoned PR). The remaining safety rules (skip running app, skip the tag pointed at by /tmp/cmux-last-cli-path) plus --keep TAG and --older-than DAYS cover what we actually want without false positives. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add /cleanup-builds slash command Wraps scripts/cleanup-dev-builds.sh with the standard preview -> confirm -> apply flow. Sits alongside the existing .claude/commands (pull, sync-branch, release, etc.) and enforces user confirmation before --apply. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address review feedback on cleanup-dev-builds CodeRabbit + Greptile findings, all real: - Active-tag extraction from /tmp/cmux-last-cli-path now uses a regex match on /cmux-<tag>/ anywhere in the path, not just a strict $DERIVED_DATA_ROOT/cmux- prefix. Also avoids the unquoted parameter expansion that could be sensitive to glob metacharacters in DERIVED_DATA_ROOT. - discover_tags switched from find | xargs basename to a shell glob loop. Cleaner, works on macOS regardless of xargs flavor, handles the empty case naturally. - --older-than no longer skips tags whose DerivedData was already deleted (age == -1 sentinel). Orphan sockets/logs for those tags now get cleaned instead of being silently retained. - "freed" label reworded as "freed (estimated)" because the byte count is measured during planning, not after rm. - .claude/commands/cleanup-builds.md: blank lines around fenced blocks (MD031). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * Fix agent resume when saved cwd is deleted (manaflow-ai#4859) * test: cover agent resume with deleted cwd * fix: resume agents when saved cwd is gone * fix: harden restored cwd guard * fix: skip ghostty cwd for guarded restore commands * fix: drop duplicate cwd args on restore * fix: preserve shell args during cwd cleanup * fix: preserve custom resume cwd arguments * fix: preserve shell syntax during cwd cleanup * chore: remove unused shell word helper --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore file preview text saving CI coverage * Restore browser session history CI coverage * chore(rename): swap bundle id, product name, config dir to most Visual rename pass — upstream-merge friendly. Touches user-visible surfaces only; keeps Swift identifiers, package dirs, env vars, and on-disk xcodeproj/test target names so cmux upstream merges remain conflict-free. - pbxproj: PRODUCT_BUNDLE_IDENTIFIER com.cmuxterm.* → com.4etverg.most* - pbxproj: PRODUCT_NAME cmux → most ("cmux DEV" → "most DEV") - Entitlements: app group ids release/nightly → com.4etverg.most[.nightly] - CLI config path: ~/.config/cmux/cmux.json → ~/.config/most/most.json - One-shot legacy config migration in KeyboardShortcutSettingsFileStore - README.md (English) user-visible refs flipped; upstream URLs preserved Co-Authored-By: OpenAI Codex (gpt-5.4-mini) <noreply@openai.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(configuration): rename to most.json + document migration Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Abdulaziz Albahar <67667005+azooz2003-bit@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Lawrence Chen <54008264+lawrencecchen@users.noreply.github.com> Co-authored-by: Konstantin <your@email.ar> Co-authored-by: OpenAI Codex (gpt-5.4-mini) <noreply@openai.com>
hhsw2015
pushed a commit
to hhsw2015/cmux
that referenced
this pull request
May 28, 2026
Resolved conflicts: - AppDelegate / Workspace / RestorableAgentSession: keep fork's top tab controller and quick terminal wiring while adopting upstream's dead-cwd fix (PR manaflow-ai#4859) and SharedLiveAgentIndex Fork Conversation availability. - pbxproj: union ref entries, bump objectVersion 56 -> 60 to match Xcode 26 pin, add ProcessPipeReader.swift + ProcessPipeReadCrashRegressionTests.swift to main app, cli, and unit-test targets (PR manaflow-ai#4800). - Web messages and CI workflow files: take upstream verbatim. Verified: - Debug build green - ForkRegressionTests: 9/9 pass - Release build green at /tmp/cmux-release-merge/Build/Products/Release/cmux.app
Vangor
added a commit
to Vangor/most
that referenced
this pull request
May 30, 2026
* test: cover external dot path open * fix: open external path arguments without socket access * fix: preserve explicit socket path opens * fix: bound launchservices open helper * fix: avoid blocking primitive in open helper * fix: format localized path open errors * fix: complete path open localization * fix: scrub socket env for external path opens * fix: localize path open success output * Add cmux.xcworkspace with Packages visible alongside the project (manaflow-ai#4834) * Add cmux.xcworkspace with Packages visible alongside the project Top-level navigator shows cmux.xcodeproj plus every Packages/* SPM package as siblings, so local packages are editable in the same window without nesting under the project node. Additive only: existing -project build flows (reload.sh, CI) keep working unchanged. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Ignore xcshareddata under xcworkspaces Xcode auto-creates xcshareddata inside .xcworkspace bundles for WorkspaceSettings.xcsettings, IDEWorkspaceChecks.plist, and SwiftPM configuration. None of those carry intentional shared state for cmux today (build is driven by reload.sh and CI, no private SPM registry, no custom file headers). Ignoring them prevents Xcode bookkeeping churn in commits. Project-level xcshareddata (shared .xcscheme files) is unaffected because the pattern only matches inside .xcworkspace directories. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * fix: scrub legacy panel env for path opens * fix: limit path open locale entries * Redesign notifications popover: bigger, minimal, swipe to dismiss (manaflow-ai#4778) * Redesign notifications popover: bigger, minimal, swipe to dismiss - Larger frame (min 460x480, ideal 560x620, max 760x760) so more notifications are visible at once. - Compact rows with a thin accent bar for unread, tighter typography, hover-revealed clear button. No more boxed cards per row, just dividers between rows. - Drag a row left or right to dismiss; passing the threshold slides it out and removes the notification. Tap to open as before. - Empty/loading/populated states share the same outer frame so the popover stops resizing when notifications come and go. * Show jump-to-latest keyboard shortcut next to button label * Stabilize hover on notification rows: dedicated background tracking layer .onHover and .onTapGesture on the same SwiftUI node share AppKit's mouse-tracking pipeline and arbitrate against each other, producing flaky enter/exit events right after the popover opens and when crossing rows fast. Move hover detection to a transparent background layer using .onContinuousHover, scope the opacity animation to the background only so it does not re-animate dragOffset, and bump the hover opacity from 0.07 to 0.11 so it is visible against .windowBackgroundColor in light mode. * Use NSTrackingArea for notification row hover SwiftUI's .onHover and .onContinuousHover still arbitrate with the parent .onTapGesture in macOS popover content, leaving rows where hover never fires (notably on first popover open and when crossing between LazyVStack rows). Replace the SwiftUI hover modifier with an NSViewRepresentable that installs an NSTrackingArea (.mouseEnteredAndExited + .activeAlways + .inVisibleRect). The tracking NSView returns nil from hitTest so clicks pass through to the SwiftUI parent's tap gesture, and the view syncs current inside/outside state on tracking-area install for the case where the pointer is already inside the row when the popover opens. * Notifications popover: keep timestamp visible, drop hover animation, add right-click menu - Always show row timestamp (removed the hide-on-hover opacity). - Removed the hover background fade animation so it's an instant flip. - Right-click a row for Open / Mark as Read / Mark as Unread / Dismiss. - Added TerminalNotificationStore.markUnread(id:) sibling to markRead(id:). - Added en/ja localizations for the new menu strings. * Notifications popover: drop top margin, always semibold title * Make notifications popover resizable, taller default Default size bumped to 560x760 (was 560x620 ideal). User can drag a bottom-right resize handle to set their own size; the chosen width and height persist via @AppStorage. Bounds are 420x320 .. 1000x1200. * Notifications popover: invisible corner resize, drop swipe-to-dismiss NSPopover has no native resize chrome and there's no first-class SwiftUI resize API for it. The closest native-feeling thing is a hit-only corner region that drives a state-driven .frame: no visual button, just a resizeLeftRight cursor on hover (closest standard NSCursor to a diagonal resize). Also removed swipe-to-dismiss. The hover X button and the right-click Dismiss menu item are the dismiss surfaces; swiping rows in an NSPopover row list isn't a macOS convention. * Notifications popover: fix glitchy resize via AppKit screen coords SwiftUI's DragGesture reports translation in a local coordinate space that is literally being resized under the cursor as the popover grows. Each frame the gesture re-derives translation from a moving anchor, which produces dimension oscillation (the popover jumps between two sizes). Replace the SwiftUI gesture with an NSViewRepresentable that tracks NSEvent.mouseLocation in global screen coordinates. Screen coordinates are stable regardless of popover resize, so deltas are monotonic. Also disable implicit animation on the .frame width/height so the popover follows the cursor 1:1. * Notifications popover: use diagonal resize cursor on bottom-right corner * Notifications popover: address review feedback - unreadCount uses notificationMenuSnapshot.unreadCount instead of reducing the live list, so the badge stays consistent with the hasUnreadNotifications / hasNotifications guards that drive the Jump to Latest / Clear All disabled states. - Hide the visual keyboard-shortcut chip from accessibility; the button already exposes the shortcut via .accessibilityValue, so VoiceOver no longer reads it twice. - Add a row-level accessibility action to dismiss a notification, since the visible clear button is hover-only. Adds the localized string notifications.row.clear (en/ja). * markUnread(id:): clear manual workspace unread to avoid double-count When a user marks a read notification unread, the notification itself now provides the workspace unread indicator. If the workspace also had a manual unread flag set, the popover header, dock badge, and workspace badge would double-count that workspace because the count builders sum notification unread counts and workspace indicators. Mirrors what markLatestNotificationAsOldestUnread already does in the same file. Found by Codex review. * Restore Button wrapper on notification rows for keyboard access Replacing the original Button with .onTapGesture removed rows from the SwiftUI key-view loop, so keyboard-only users could tab to the header controls but not open a row. The original reason for dropping Button was that SwiftUI's .onHover arbitrated with the row's primary action — but hover is now driven entirely by an AppKit NSTrackingArea (HoverTrackingRepresentable), independent of the row's activation node. The Button wrapper is therefore safe again and restores space/return activation in the key-view loop. Found by Codex review. * Notifications popover: persist resize once on mouseUp, clamp drag baseline - Live resize uses @State (liveWidth/liveHeight); @AppStorage is written exactly once when the user releases the mouse, instead of on every mouseDragged event. Each @AppStorage write was hitting UserDefaults and broadcasting UserDefaults.didChangeNotification to every app-wide observer. - Drag baseline now comes from the clamped (currently displayed) size, not raw saved doubles, so a drag that starts with out-of-bounds stored values doesn't lose initial pointer travel re-entering the visible range. * Notifications popover: clear button as ZStack sibling, not nested in row Button Nesting clearButton inside the row's Button(action: onOpen) label created the classic SwiftUI nested-button hit-test problem on macOS: clicks on the inner X could be consumed by the outer row Button's tap area instead of clearing the notification. Move clearButton out of rowContent and into a ZStack(alignment: .trailing) at the body level so it's a sibling of the row Button with an independent hit target. Found by Codex review. * Notifications popover: clamp to screen, hide hover-only X from focus - Clamp the popover's frame against NSScreen.main?.visibleFrame minus an 80pt margin so the bottom-right resize handle stays reachable even when the saved size was captured on a larger display. - Mark the hover-only clear button .accessibilityHidden(!isHovering) so Full Keyboard Access / VoiceOver doesn't focus an invisible dismiss control. Dismissal is still exposed via the row's accessibility action and the right-click menu. Found by Codex. * CI: retry — GitHub Actions infra was returning 403/404 on codeload * CI: retrigger now that GitHub Actions infra is back * Mark-as-Read context menu: clear focused pane indicator too A user-initiated 'Mark as Read' should mirror the full dismissal intent, so the pane's focused-read indicator for that surface should clear alongside flipping the notification's isRead flag. Otherwise the unread count drops but the pane badge remains visible until the user later interacts with the terminal. Found by Codex review. * Notifications popover: clamp against host window's screen, fix hover sync race - Clamp popover size against the screen of NSApp.keyWindow (the popover's anchor window), not NSScreen.main. On multi-monitor setups the popover may appear on a different display than 'main'. - Remove the DispatchQueue.main.async hop in HoverTrackingNSView's updateTrackingAreas; the queued onChange(true) could land after mouseExited's synchronous onChange(false), leaving rows stuck in the hovered state. updateTrackingAreas already runs on the main thread, so the call is fine synchronously. Found by Cursor Bugbot. * Mark-as-Read: only clear focused pane indicator for pane-scoped notifications clearFocusedReadIndicator treats surfaceId == nil as 'clear any pane indicator for this tab', so passing nil for a workspace-level notification would wipe an unrelated pane's badge. Gate the call on notification.surfaceId being non-nil. Found by Codex review. * Notifications popover: add a11y label/hint to resize handle The localized notifications.resize key existed but was never wired to the actual control. Adds .accessibilityLabel/.accessibilityHint so VoiceOver and Full Keyboard Access users can discover the resize affordance. Found by Cursor Bugbot. * Notifications popover: move accessibility identifier back onto the row Button XCUITests query rows with app.buttons["NotificationPopoverRow.<id>"]. The previous pass put .accessibilityIdentifier on the combined outer ZStack, exposing the row as a container rather than a button, which breaks both XCUITest lookups and the button accessibility role for assistive tech. Move identifier + primary action + clear action back onto the inner Button. Found by Codex review. * Notifications popover: snapshot list before LazyVStack; clear restored unread on markUnread - ForEach now iterates an immutable snapshot captured outside the LazyVStack instead of reading notificationStore.notifications inside the row builder; the previous code reintroduced a store dependency below the list boundary, which CLAUDE.md flags as the same anti-pattern that caused the LazyLayoutViewCache spin-loop in the sessions panel (manaflow-ai#2586). - markUnread(id:) now also clears the restored workspace unread indicator for the tab, so a session-restored unread hint plus a user-toggled unread notification no longer double-count in the header/dock/workspace badges. Found by Codex review. --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Remove History from right sidebar (manaflow-ai#4785) * Remove History mode from right sidebar Drops the History tab from the right sidebar mode picker along with its keyboard shortcut, command palette entries, and the 'Open Full History' menu button. Recently Closed/Focused submenus and Reopen Last Closed still work from the menu bar. * Drop tests for deleted HistoryDayGrouping helper * Tolerate obsolete history mode and restore focus-history menu key - SessionRightSidebarToolPanelSnapshot now decodes an unknown mode (e.g. legacy 'history') as nil instead of failing the entire snapshot, so upgraded users can still restore prior sessions. - Restore the 'menu.history.showFullFocusHistory' key still referenced by AppDelegate+FocusHistoryContextMenu for non-sidebar focus history. * Remove duplicate @mainactor attribute left by prior edit --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Launch restored agent sessions via startup commands (manaflow-ai#4777) * Test agent restore startup command launch * Launch restored agents with startup commands * Tighten restored startup launch state * Preserve remote startup for agent restore * Relax remote startup restore assertion * Fix startup restore lifecycle * Remove unused restore launch flag * Run startup resumes in login shell * Tighten remote restore assertion * Handle csh agent restore launchers --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Wrap workspace titles in sidebar (manaflow-ai#4848) * Wrap workspace titles in sidebar * Add workspace title wrap toggle * Localize workspace title wrap setting * Align wrapped workspace title accessories * Reset workspace title wrap preference --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: cover bare relative external path open * fix: open bare relative path arguments externally * fix: preserve refresh surfaces command precedence * fix: keep bare path command matching case-sensitive * Add secure cmux navigation links (manaflow-ai#4857) * Add secure cmux navigation links * Localize cmux navigation link labels --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Fix matched sidebar terminal background (manaflow-ai#4780) * test: cover matched sidebar background tint * fix: match sidebar terminal background * fix: remove dead sidebar overlay path * test: cover matched sidebar boundary separators * fix: keep matched sidebar separators visible * test: cover matched sidebar chrome borders * test: sample sidebar boundary resizers --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Fix JSONC comment skipper for CRLF line endings (manaflow-ai#4869) * Fix JSONC // comment skipper for CRLF line endings Swift treats \r\n as a single extended grapheme cluster, so the existing source[index] != "\n" check inside the three // line-comment skippers never matches a CRLF line ending. On a file with CRLF throughout, the loop runs past the rest of the file looking for a standalone LF and strips everything after the first //. preprocess() then produces truncated JSON and loadCmuxSettingsRoot() / jsonObject() throw "Unexpected end of file". This was breaking the tmux corpus terminal-nightly job's testSurfaceResumeApprovalWritesRecordsIntoCmuxJSON, which explicitly exercises the CRLF path. Match any character whose first scalar is CR or LF so we stop at CR, LF, or CRLF correctly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Allow workflow_dispatch to run tmux corpus terminal-nightly So this workflow's macOS test job is reproducible on demand for branches under review, not only on the nightly schedule. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Fix JSONC editor indent detection for CRLF line endings indentationBeforeLine walked back to find the start of the line containing a given index by stopping at "\n" or "\r" characters, but on CRLF input Swift treats "\r\n" as a single extended grapheme cluster that equals neither. The loop then walked past every line break to the start of the file and returned an empty indent, so newly inserted properties (e.g. resumeCommands) ended up at the wrong indentation level. Reuse the isLineTerminator helper that already handles CRLF correctly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Open forked conversations without palette action probe (manaflow-ai#4852) * Open forked conversations without palette action probe * Fix fork probe snapshot selection * Refresh fork probe cache per palette session * Preserve verified fork probe visibility * Preserve fork snapshot cache during live refresh * Allow verified fallback fork execution * Reprobe fork cache after fallback clears * Preserve focus after instant fork commands * Clear fork cache on execution failure * Avoid stale fork cache on palette reopen * Clear stale fallback fork cache before reprobe * Preserve verified fork cache on palette reopen * Require verified fork snapshot before fallback execution * Track actual fork fallback usage * Preserve verified fork cache fallback flag --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: cover bonsplit tab indicator drift * fix: update bonsplit tab indicator handling * Skip Cmd+Shift key forwarding test when Ghostty surface init fails (manaflow-ai#4871) The tmux corpus terminal-nightly runner cannot initialize a Metal-backed Ghostty surface; embedded_window logs "error initializing surface err=error.OutOfMemory" for every surface the suite creates, and ghostty_surface_new returns nil. The test then fails XCTAssertTrue on performKeyEquivalentAfterMenuMiss because ensureSurfaceReadyForInput cannot return a live surface, which looks like a key-forwarding regression rather than the environment issue that it is. Skip the test (via XCTSkipUnless on TerminalSurface.hasLiveSurface) when the surface fails to initialize, so CI surfaces the real problem (Metal unavailable on this runner) without masking it as a key-bind regression. The test still runs end-to-end on developer machines and on runners with a logged-in GUI session. Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Add algorithmic complexity review rule (manaflow-ai#4866) * Add algorithmic complexity review rule * Align algorithmic complexity review hint --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Narrow AppDelegate CI quarantine (manaflow-ai#4874) Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore browser devtools config CI coverage * Add prompt and rules deeplinks (manaflow-ai#4839) * Add prompt and rules deeplinks * Preserve encoded punctuation in text deeplinks * Localize text deeplink dialogs * Support Freestyle SSH link users * Harden prompt and rules deeplinks * Suppress welcome for text deeplinks * Restore SSH deeplink trust gate * Preserve plus signs in text deeplinks * Respect no-focus for text deeplink target selection * Use deterministic text deeplink target * Start background text deeplink targets * Report text deeplink send failures * Handle mixed deeplink batches safely * Gate mixed external deeplink batches * Report failed external text deeplink sends * Stop after handling external cmux links * Allow multiline cmux text links * Reject control characters in cmux text links * Polish external cmux link handling * Label sendTextWhenReady callbacks --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Make session index backfill linear (manaflow-ai#4868) * Make session index backfill linear * Stabilize session index backfill tie breaks * Fix session index tie-break test setup --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Optimize batch workspace sidebar actions (manaflow-ai#4865) * Optimize batch workspace sidebar actions * Preserve workspace selection after drag reorder * Harden workspace batch action coverage --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * test: harden bonsplit scroll indicator check * Restore browser remote store CI coverage * Make browser search providers configurable (manaflow-ai#4849) * feat: make browser search providers configurable * fix: keep parsing browser settings after invalid custom search config * test: harden custom search config regression * fix: handle custom search edge cases * fix: validate custom search fallbacks * fix: localize search engine labels * fix: gate stale search suggestions * fix: allow blank custom search names * fix: add exa search provider --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore browser lifecycle CI coverage * docs: add cmux ssh deep links (manaflow-ai#4833) * docs: add cmux ssh deep links * docs: add cmux deeplink fallback pages * docs: add deeplink locale fallbacks * docs: tighten deeplink fallback validation * docs: mirror native ssh deeplink validation * docs: localize deeplink messages * docs: canonicalize rules deeplink alternates * Allow Freestyle SSH deeplink users * docs: validate text deeplink fallback params * docs: ignore blank optional deeplink params * docs: address deeplink review comments --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore file preview review CI coverage * Pin Xcode 26 (objectVersion 60) and add pbxproj normalizer + CI guard (manaflow-ai#4836) * Add deterministic normalizer for cmux.xcodeproj/project.pbxproj scripts/normalize-pbxproj.py sorts the high-churn sections (PBXBuildFile, PBXFileReference, and the files = (...) arrays inside Sources / Resources / Frameworks / CopyFiles build phases) into a deterministic order keyed on the entry comment plus UUID. The Xcode build does not care about the order of these flat dictionary sections; sorting them just kills the nondeterministic diff noise Xcode generates on every UI touch. Does not touch UUIDs, comments, or PBXGroup children = (...) arrays (navigator order is intentional). Idempotent: a second run produces zero diff. Standalone in this commit so the diff is just the script. The next commit applies the script and bumps objectVersion in one shot, so the resulting churn is contained and never repeated. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Pin objectVersion = 60 and normalize pbxproj Bumps objectVersion from 56 to 60 (the format Xcode 16+ and Xcode 26 write by default) and runs scripts/normalize-pbxproj.py once to establish the deterministic baseline. After this commit, future diffs to project.pbxproj show only real changes, not Xcode's nondeterministic section reordering. One-time large diff. No semantic changes to targets, sources, build phases, or settings: pure sort + version pin. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add tracked pre-commit hook that normalizes pbxproj scripts/git-hooks/pre-commit calls scripts/normalize-pbxproj.py on cmux.xcodeproj/project.pbxproj when it is staged and re-stages the result. scripts/install-git-hooks.sh points the clone at this directory via `git config core.hooksPath scripts/git-hooks`, and scripts/setup.sh auto-runs it so devs get the hook without a separate manual step. After this, Xcode's nondeterministic reordering of build-file and file-reference sections is canceled out at commit time. The CI guard in the next commit enforces the rule for anyone who bypasses the hook with --no-verify or who never ran setup. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add CI guard for objectVersion pin and pbxproj normalization scripts/check-pbxproj.sh asserts cmux.xcodeproj/project.pbxproj has objectVersion = 60 (Xcode 26 default) and that the file is normalized per scripts/normalize-pbxproj.py. Wired as a step in the workflow-guard-tests job so every PR is gated. This catches anyone who bypasses the pre-commit hook with --no-verify or who never ran scripts/setup.sh. The error message points at the exact fix path. To bump the pin (e.g., when the team adopts a newer Xcode major), edit EXPECTED_OBJECT_VERSION in this script and the matching line in CLAUDE.md. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add .xcode-version and document Xcode 26 pin in CLAUDE.md .xcode-version records the major (26.0) for tooling that reads it (xcodes CLI, some CI helpers). CLAUDE.md gains an Xcode toolchain section explaining the pin, the normalizer + pre-commit hook + CI guard mechanics, and the procedure for bumping the pin in the future. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Read .xcode-version as the source of truth in check-pbxproj.sh scripts/check-pbxproj.sh now reads .xcode-version and maps the Xcode major to the expected objectVersion via a one-entry case statement. Bumping the team's Xcode pin becomes a one-file edit (.xcode-version), with a script update only required when Apple actually changes objectVersion in a new Xcode major. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address CodeRabbit findings on check-pbxproj.sh and pre-commit hook scripts/check-pbxproj.sh now passes "$PBXPROJ" explicitly to normalize-pbxproj.py instead of letting it default to a path relative to the current working directory, so the guard works regardless of where CI invokes it. scripts/git-hooks/pre-commit refuses to run when the working-tree pbxproj has unstaged changes. Previously the hook would normalize the working-tree file and `git add` the result, which silently staged any unstaged hunks the user had deliberately left out of the commit. The hook now exits non-zero with a clear message telling the user to either stage the whole file or stash the unstaged hunks first. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address Greptile findings: misleading comment + bump-step docs scripts/normalize-pbxproj.py: the comment said "preserve empty lines exactly where they are" but the implementation collapses blanks to a trailing group. Reworded the comment to match the actual behavior. CLAUDE.md: the bump procedure now mentions opening cmux.xcodeproj in the new Xcode so objectVersion gets rewritten automatically. Without that step a developer following the docs alone would update only the pin file and the script case, and the CI guard would fail on their next commit. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * Add scripts/cleanup-dev-builds.sh for safely reclaiming tagged DerivedData (manaflow-ai#4837) * Add scripts/cleanup-dev-builds.sh Removes tagged dev-build artifacts produced by scripts/reload.sh: DerivedData/cmux-<tag>/ (multi-GB each), /tmp/cmux-<tag>/, the per-tag debug socket and logs, the reload log, and the App Support cmuxd dev socket. Defaults to dry-run; pass --apply to delete. Safety rules always on: - Skip the tag of any running `cmux DEV <tag>` app - Skip the tag pointed at by /tmp/cmux-last-cli-path - Skip any tag tied to a live git worktree Filters: --older-than DAYS, --keep TAG (repeatable). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Drop "worktree exists" safety rule in cleanup-dev-builds.sh Existence of a git worktree with the same name is a weak signal of active use, and HQ tends to accumulate worktrees long after the work is done. The rule made cleanup over-protective for the typical case (worktree still around from a merged or abandoned PR). The remaining safety rules (skip running app, skip the tag pointed at by /tmp/cmux-last-cli-path) plus --keep TAG and --older-than DAYS cover what we actually want without false positives. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Add /cleanup-builds slash command Wraps scripts/cleanup-dev-builds.sh with the standard preview -> confirm -> apply flow. Sits alongside the existing .claude/commands (pull, sync-branch, release, etc.) and enforces user confirmation before --apply. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * Address review feedback on cleanup-dev-builds CodeRabbit + Greptile findings, all real: - Active-tag extraction from /tmp/cmux-last-cli-path now uses a regex match on /cmux-<tag>/ anywhere in the path, not just a strict $DERIVED_DATA_ROOT/cmux- prefix. Also avoids the unquoted parameter expansion that could be sensitive to glob metacharacters in DERIVED_DATA_ROOT. - discover_tags switched from find | xargs basename to a shell glob loop. Cleaner, works on macOS regardless of xargs flavor, handles the empty case naturally. - --older-than no longer skips tags whose DerivedData was already deleted (age == -1 sentinel). Orphan sockets/logs for those tags now get cleaned instead of being silently retained. - "freed" label reworded as "freed (estimated)" because the byte count is measured during planning, not after rm. - .claude/commands/cleanup-builds.md: blank lines around fenced blocks (MD031). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> * Fix agent resume when saved cwd is deleted (manaflow-ai#4859) * test: cover agent resume with deleted cwd * fix: resume agents when saved cwd is gone * fix: harden restored cwd guard * fix: skip ghostty cwd for guarded restore commands * fix: drop duplicate cwd args on restore * fix: preserve shell args during cwd cleanup * fix: preserve custom resume cwd arguments * fix: preserve shell syntax during cwd cleanup * chore: remove unused shell word helper --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> * Restore file preview text saving CI coverage * Restore browser session history CI coverage * chore(rename): swap bundle id, product name, config dir to most Visual rename pass — upstream-merge friendly. Touches user-visible surfaces only; keeps Swift identifiers, package dirs, env vars, and on-disk xcodeproj/test target names so cmux upstream merges remain conflict-free. - pbxproj: PRODUCT_BUNDLE_IDENTIFIER com.cmuxterm.* → com.4etverg.most* - pbxproj: PRODUCT_NAME cmux → most ("cmux DEV" → "most DEV") - Entitlements: app group ids release/nightly → com.4etverg.most[.nightly] - CLI config path: ~/.config/cmux/cmux.json → ~/.config/most/most.json - One-shot legacy config migration in KeyboardShortcutSettingsFileStore - README.md (English) user-visible refs flipped; upstream URLs preserved Co-Authored-By: OpenAI Codex (gpt-5.4-mini) <noreply@openai.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(configuration): rename to most.json + document migration Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Lawrence Chen <lawrencecchen@users.noreply.github.com> Co-authored-by: Abdulaziz Albahar <67667005+azooz2003-bit@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Lawrence Chen <54008264+lawrencecchen@users.noreply.github.com> Co-authored-by: Konstantin <your@email.ar> Co-authored-by: OpenAI Codex (gpt-5.4-mini) <noreply@openai.com>
This was referenced May 30, 2026
austinywang
added a commit
that referenced
this pull request
Jun 10, 2026
PR #4859 (May 27) changed agent resume cwd handling to the `{ cd -- '<dir>' 2>/dev/null || [ ! -d '<dir>' ]; } &&` guard so resume survives a deleted working directory; every sibling agent expectation in this test asserts that form. The kiro expectation merged May 30 from a branch written against the old plain `cd '<dir>' &&` prefix and has failed ever since — masked because the CI unit-test job times out before reaching this suite. No production behavior change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
austinywang
added a commit
that referenced
this pull request
Jun 10, 2026
…5639) (#5721) * Add failing regression test: resumed claude must execute through cmux wrapper The close-&-reopen restore launcher runs the resumed agent in a fresh $SHELL -lic login shell where cmux's shell integration (claude() function + per-surface PATH shim) is not active, and the resume command is `env … claude …` — `env` resolves `claude` via execvp, bypassing any shell function. So bare `claude` resolves to the user's real binary, the cmux wrapper is bypassed, no hook --settings is injected, and SessionStart/Stop/Notification stay dead on resume. #5430 only asserted the argv string, never the executed shell resolution. This test runs the real resumeCommand through `zsh -lic` (as the launcher does) in a hermetic sandbox whose login profile clobbers PATH and shadows claude with a function, and asserts the launched invocation routed through the wrapper (its injected --settings). It fails on current main (no fix yet). Refs #5639 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Resolve resumed claude through the wrapper shim env var, not bare PATH #5430 fixed the resume argv string but the executed command still bypassed the cmux `claude` wrapper. The close-&-reopen restore launcher runs the resumed agent via `$SHELL -lic <cmd>` (SessionPersistence.commandThenReturnLines), a fresh login shell where cmux's shell integration (PATH shim + `claude()` function) is not active, and the command is `env … claude …` — `env` resolves `claude` via execvp, bypassing shell functions, and the user's login profile rebuilds PATH, dropping the shim dir. So bare `claude` hit the user's real binary, no hook --settings was injected, and SessionStart/Stop/Notification stayed dead on resume. Render the claude executable as the wrapper shim token "${CMUX_CLAUDE_WRAPPER_SHIM:-claude}" instead of bare `claude`. CMUX_CLAUDE_WRAPPER_SHIM is a managed terminal env var (TerminalStartupEnvironment / GhosttyTerminalView) inherited by every descendant shell regardless of PATH/function shadowing, so the executed resume/fork command execs the per-surface shim → wrapper → re-injected hook --settings, and falls back to bare `claude` outside cmux. This is immune to the env-bypass / PATH-clobber that defeat integration-sourcing fixes. Applied at the one shared value-layer seam (AgentResumeArgv.renderingClaudeWrapperExecutable) across all three claude resume render paths: the app resume/fork builder (AgentResumeCommandBuilder), the cmux-cli surface-restore publisher (agentSurfaceResumeCommand), and the session-index manual resume (SessionEntry). Other agents resumed through the same launcher replay an absolute executable path and are unaffected; grok still relies on its function-only wrapper (no shim env var) and is out of scope. Makes the failing regression test green. Fixes #5639 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Bump Swift file length budget for #5639 changes Account for the new regression test and the wrapper-shim render additions across the four files this PR grows. Scoped to changed files only. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * Add failing regression tests: claude resume must parse in tcsh and fish The wrapper-shim token "${CMUX_CLAUDE_WRAPPER_SHIM:-claude}" is POSIX parameter expansion, but resume commands are dispatched through the user's login shell (commandThenReturnLines runs `"$_cmux_resume_shell" -c <cmd>` for csh|tcsh and fish) and typed into the user's interactive shell. tcsh rejects the token with "Bad : modifier in $"; fish rejects ${…} outright — and the entire pre-#5639 command was valid fish, so fish users regress from working resume to a hard parse error. Red on this commit by design (regression test commit policy); the portable render fix lands in the follow-up commit. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Wrap token-bearing claude resume commands in /bin/sh -c for any shell The wrapper-shim token is POSIX parameter expansion, but rendered resume commands are parsed by the user's shell: the restore launcher dispatches "$_cmux_resume_shell" -c <cmd> for csh|tcsh and fish, session-drop types the command into the user's interactive shell, and Copy Resume Command pastes anywhere. fish rejects ${…} outright and csh/tcsh have no `:-` modifier, so the raw token regressed those shells from working resume to a hard parse error. `/bin/sh -c '<command>'` is the one spelling every dispatching shell parses identically: the user's shell still sources its own config, sh inherits CMUX_CLAUDE_WRAPPER_SHIM from the managed terminal environment and resolves the token, and outside cmux the unset variable still falls back to bare claude. The wrap is applied at the shared seam (AgentResumeArgv.renderedPortableClaudeResumeShellCommand) only when the token was actually substituted — claude-teams resumes that resolve to cmux's own CLI stay unwrapped — and before the cwd guard so cd-prefix rewriting keeps composing. Non-ASCII cwd commands now exceed the inline startup-input budget and use the established launcher-script form; the escaping test resolves through the script. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Bound the launcher regression-test shell waits (review feedback) process.waitUntilExit() on the dispatched login shells is unbounded: a stalled subprocess (missing shebang interpreter, prompting profile) would hang the test until the CI harness kills the job instead of failing with a clear message. Wait on a termination-handler semaphore with a 30s deadline and fail crisply on timeout. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Align stale kiro resume expectation with the #4859 cwd guard form PR #4859 (May 27) changed agent resume cwd handling to the `{ cd -- '<dir>' 2>/dev/null || [ ! -d '<dir>' ]; } &&` guard so resume survives a deleted working directory; every sibling agent expectation in this test asserts that form. The kiro expectation merged May 30 from a branch written against the old plain `cd '<dir>' &&` prefix and has failed ever since — masked because the CI unit-test job times out before reaching this suite. No production behavior change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Move posixSingleQuoted to file scope per package design policy Pure helpers belong as file-scope private funcs rather than private static members on the type (cmux Aziz package-design policy). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Guard the wrapper-shim token on executability, not bare expansion macOS reaps idle temporary-directory contents after ~3 days, so a long-idle surface can hold CMUX_CLAUDE_WRAPPER_SHIM while the shim file is gone. Bare ${VAR:-claude} expansion only falls back when the variable is unset/empty — a set-but-dead path would exec "No such file or directory" and hard-fail resume, worse than the pre-#5639 graceful degradation. The token now guards on [ -x … ] and falls back to PATH-resolved claude (hooks lost, resume works), identical to the unset-variable case outside cmux. The regression test runs the real resume command under zsh -lic with the env var pointing at a reaped path and asserts the PATH fallback executed (shell-level red proof for the old token: sh execs the dead path and exits 127 without launching anything). Test expectations now compose the token from AgentResumeArgv.claudeWrapperShellExecutableToken instead of duplicating the literal. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jun 13, 2026
This was referenced Jun 22, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
cd ... &&startup prefixes with a cd-first guard:{ cd -- <cwd> 2>/dev/null || [ ! -d <cwd> ]; } && ....commandreads see the guarded form.Regression test:
f2f1302baaddsSessionPersistenceTests.testAgentHookSurfaceResumeStartupInputRunsWhenSavedWorkingDirectoryWasDeleted.739e349afmakes the red test pass by replacing hard cwd preconditions with guarded cwd attempts.05146ab25makes the guard race-safe, canonicalizes persisted commands on decode, and covers non-ASCII legacy prefixes.Verification:
git diff --checkpassed../scripts/lint-pbxproj-test-wiring.shpassed../scripts/reload.sh --tag restcwdpassed and builtcmux DEV restcwd.app.Cloud proof:
cloud-mac-26494038050./Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/before-window/recording.mov| TLDR: old restore command exits withzsh:cd:1: no such file or directorybeforecodex resumecan run./Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/before-window/frames/02.png,/Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/before-window/frames/tail.png./Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/after-final-window/recording.mov| TLDR: final cd-first restore guard skips the missing cwd, runscodex resume session-duplicate-turn --yolo, and exits 0./Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/after-final-window/frames/02.png,/Users/lawrence/fun/cmuxterm-hq/cmux-assets/issue-session-restore-missing-cwd/auto-issue/20260526-230750/after-final-window/frames/tail.png.Note: full-display recordings were captured too, but macOS displayed a private-window-picker permission prompt over them. The accepted proof videos above are prompt-free TextEdit window recordings, with clean checked frames and metadata.
Note
Medium Risk
Changes shell startup and session-restore paths used on every agent resume; behavior is well-covered by new tests but mistakes could affect restore or wrong cwd.
Overview
Agent and terminal resume/restore no longer fail when the saved working directory was deleted. Generated startup commands use a cd-first guard (
{ cd -- <cwd> 2>/dev/null || [ ! -d <cwd> ]; } && …) instead of a hardcd … &&that aborts before the agent runs.Resume command building centralizes that prefix in
TerminalStartupWorkingDirectoryPrefix, strips legacycdprefixes and duplicate--cd/--cwd/--workspaceoptions (viaAgentLaunchSanitizer.removingSavedWorkingDirectoryOptions), and applies the same pattern across agent resume/fork, Codex Teams/tmux launcher scripts, and restored terminal scripts.Persistence & restore:
SurfaceResumeBindingSnapshotsanitizes storedcommandon init/decode for agent-hook bindings;WorkspaceomitsrequestedWorkingDirectorywhen startup already handlescd(tmux scripts, agent resume, agent-hook bindings) so Ghostty does not fail early on a missing path.Tests were updated/added for missing cwd, duplicate cwd flags, non-ASCII paths, and shell operators.
Reviewed by Cursor Bugbot for commit bb5c757. Bugbot is set up for automated code reviews on this repo. Configure here.
Summary by CodeRabbit
Bug Fixes
New Features
Tests