[lawrencecchen]

Summary

Cmd-T, then Cmd-W soon after, sometimes closed nothing: the New Tab page stayed. This turned the tip smoke red at b732874 (STATUS 4852, step 66 "Cmd-W closes that tab").

Cmd-T shows the New Tab page as a stand-in tab (pending_tab_...). When the daemon's reply comes, the store replaces the stand-in with its own tab in one step. The pane's strip follows the store one observation step later. A Cmd-W in that step reads the strip's selected id, which is still the stand-in. PaneController.tab(_:) found no tab with that id, so Close Tab (and every other action on the selected tab) did nothing, with no message.

The fix: the store's intent log keeps, for the last 64 confirmed creations, which daemon surface each stand-in tab became. PaneController.tab(_:) resolves a stand-in id that the pane no longer lists to that tab while the pane lists it (DaemonCreations.replacement(ofStandIn:in:)). A stand-in that is still shown keeps the cancel path ("closed before created"); a cancelled or refused stand-in still resolves to nothing. CmuxNextApp lines: +0 (one line changed).

This race is old: it is not a commit in the tip range. The same failure shows at d15622d (2026-11-01) and on the last smoke-green build (7563ccd).

Testing

Regression test (committed red before the fix): CmdWOnReplacedStandInTests (CmuxNextAppTests, live same-tree cmux-tui). It shows a New Tab stand-in, lets the daemon make the tab (hidden behind the stand-in), runs the reply's turn as AgentTabStore.open does, and presses Cmd-W in that same turn, before the strip's observation step.

Fleet command (red at 9fe44fb, green at 2f4997e):
cmux-ci run --class light --script scripts/ci/package-test-lane.sh --ref SHA --arg=suite --arg=Packages/macOS/CmuxNext --arg=CmdWOnReplacedStandInTests,CloseTabKeepsPaneTests,RapidCloseTabTests,NewTabReplaceInPlaceTests --env CMUX_CI_XCODE_APP=/Applications/Xcode_26.6.app

• Red on the base + test only (cf90261, branch fix/cmdw-stale-standin-red, fleet step 9c180fae, aws-m4pro-1): CmdWOnReplacedStandInTests failed at the Cmd-W check: "Cmd-W closed nothing: the pane holds [...], the New Tab tab is tab_6080d827...". The other 4 suites passed.
• Green on the head (934d51b, fleet step dd7aae92, aws-m4pro-8): CmdWOnReplacedStandInTests, CloseTabKeepsPaneTests, RapidCloseTabTests, NewTabReplaceInPlaceTests and TabSpamRaceTests (it covers the intent log) all passed.
• The first version of the test (9fe44fb) waited for the store to pair the daemon's raw tab-added with the stand-in. On the base run (step ed4e88f8) that wait timed out before the Cmd-W, so that run is no red evidence. 934d51b sets the same state without that wait.
• scripts/cmux-next/prepush-gate.py: no new violations; arch ratchet: no count rises.

Observed / Not run

Observed:

• Fleet diag journey (smoke journey to the browser page, then 16 Cmd-T/Cmd-W pairs 0-0.25 s apart, hog +15 GB, debug.key, cmux21): pair 2 (Cmd-W about 45 ms after Cmd-T) left the New Tab tab open on the tip build 84b90c53 (2 of 2 rounds) and on the last smoke-green build 45613f0e (1 of 1 round). Pair 1 (Cmd-W at once) closed the stand-in through the cancel path.
• The tip smoke round's state: Cmd-W ran closeTab (handled_by app), no "closed before created" in new_tab_key_log, the page then reported ready at +134 ms and stayed; 25 s later the pane still had the New Tab tab selected.
• Fix build 027c40e1 (2f4997e), diag round 1 (cmux24): the round failed earlier, at step 23 ("Cmd-D twice fast": both typed lines went to one new terminal), before the Cmd-T/Cmd-W pairs. That step has failed the same way on several tips before (Oct 31 tip-watcher rounds), and the fix does not touch the split path (.splitPane creations are not recorded). More rounds on this build are queued through the mini scheduler.

Not run: real HID Cmd-W rounds (the smoke's real-key rounds stop earlier at the New Column item, STATUS 1712/1724).

Changelog

Fixed: Cmd-W right after Cmd-T closes the new tab every time, instead of sometimes doing nothing.

Proof

No capture: no visible UI change; the evidence is the regression test and the fleet journey above.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Fixes a race where Cmd-W right after Cmd-T closed nothing, leaving the New Tab page open.

Cmd-T shows the New Tab page as a stand-in tab. When the daemon's reply arrives, the store replaces the stand-in with its own tab, but the pane's strip only catches up one observation step later. A Cmd-W in that step still targets the stand-in id, which the store no longer lists, so Close Tab silently found no tab.

The intent log now keeps, for the last 64 confirmed creations, which daemon surface each stand-in became. PaneController.tab(_:) resolves a stand-in id the pane no longer lists to that tab while the pane lists it. Cancelled or refused stand-ins still resolve to nothing.

Also adds a regression test that replays the racy turn and verifies Cmd-W closes the created tab.

^{Written for commit 934d51b. Summary will update on new commits.}

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 934d51b: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr31837 -f ref=fix/cmdw-stale-standin -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4139510839501 (d47dcbddc6)
• cmux NEXT nightly 1.0.0-nightly.4139567428201 (14a5e1b153)
• cmux NEXT nightly 1.0.0-nightly.4139686277401 (4e9b96d7dd)
• cmux NEXT nightly 1.0.0-nightly.4139750725501 (947bb5e00d)

[lawrencecchen]

Review (smoke-cmdw reviewer): SHIP at 934d51b

Non-blocking notes:

• The test creates the daemon's tab before intend, so it shows beside the stand-in (production hides it via guess), and it sets pendingSelectSurface by hand instead of running open/tabCreated. It covers the close lookup, not the exact strip state. An #expect that the browser page is selected again would cover the second symptom.
• The test is skipped without SameTreeDaemonBinary: please name the fleet run that shows red on base and green on head.
• Strip-id consumers that bypass tab(_:) (cache.browsers[id] in livePageDecidesClose, pageClosing, the renameTab ask) still see the stale stand-in id for one step. For a frontend-browser stand-in, a Cmd-W in that step skips beforeunload. The New Tab page is not affected.
• PaneController.swift:573 is now 160 characters.

[blacksmith-sh]

Found 6 test failures on Blacksmith runners:

Failures

Test	View Logs
main.DeveloperIdAppTests.test_download_prints_the_active_profile_only/ test_download_prints_the_active_profile_only	View Logs
stubs: /System/Library/Frameworks/Security.framework/Versions/A/Security has no stub/ Security has no stub	View Logs
stubs: CoreFoundation.tbd: install-name /System/Library/Frameworks/ CoreFoundation.framework/Versions/A/CoreFoundation, missing 1: ['_CFRetain']/ CoreFoundation, missing 1: ['_CFRetain']	View Logs
stubs: CoreFoundation.tbd: install-name /System/Library/Frameworks/ CoreFoundation.framework/Versions/A/CoreFoundation, missing 1: ['_log2']/ CoreFoundation, missing 1: ['_log2']	View Logs
test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)/ test_every_compile_job_names_its_warm_sources (main.ThisTree.test_every_compile_job _names_its_warm_sources)	View Logs
test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)/ test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}