[teamleaderleo]

Leo, closing several selected sidebar workspaces (the multi-select close from #23110):

"also idk what's going on with these new workspace deletions but it's really slow and it's missing a LOT... and it's also 2 lines tall for some reason"

and then: "no, it really is deletion not dictation."

What was broken

Cmd-W, the row menu's Close and the sidebar's Cmd-Delete on a selection all send SidebarIntent.close(ids). SidebarBridge+Intents handled that by running Close Workspace once per row (ActionInvocation(closingRow:)), each a full single-workspace WorkspaceCloseFlow:

• Slow. Every row took its own turn in the window's WorkspaceLanes, and each turn waited for the daemon's reply and then for the store to apply it before the next row's close started. N rows meant N serial round trips, plus a sidebar pass and a selection repair per row.
• Missing a lot. With a terminal running a program (the usual case), every row asked its own question. They queued one behind another, Return does not confirm them (returnConfirms = false), and any you didn't answer left that row unclosed. Rows were also refused one at a time against a window room that was shrinking.
• 2 lines tall. The question title was per row: Close “<workspace name>”? in bold, in a dialog fixed at 380 pt (340 pt of text). Agent workspace names are long, and a name over roughly 40 characters wraps that title to two lines. This is an inference from the layout: the base capture uses short names (Close “Deletion 12”?, one line), and I have no capture of Leo's screen. The batch title Close N workspaces? and the toast Closed N workspaces are short, fixed strings, and the after-fix capture shows both on one line.

The fix

A selection of two or more rows now closes as one batch (Handlers/WorkspaceSelectionClose.swift):

• Every row leaves in one sidebar pass and the selection moves once (InstantClose.begin([ids])).
• One question for the whole selection, only when a terminal runs a program: Close N workspaces? (the existing WorkspaceBulkClose.prompt, one line). Its process lookups now run concurrently.
• Every close is sent at once in one lane turn, then the batch waits once per daemon for the store.
• One Closed N workspaces toast with Undo (WorkspaceGroupClose.offerUndo, the same toast the other bulk closes use) reopens all of them.
• A paused machine's closes wait for it outside the lane, as single closes do; they join the toast when they finish and never hold it back.
• A refused row lists again with one refusal toast for the batch.
• Each row stays reserved until its own close settles, so Close Workspace can't close it twice.

A single row, and Close Workspace with no selection, keep the existing WorkspaceCloseFlow. Close All in Group already applies one model pass and sends every close at once (SidebarBridge .closeGroup), so it is unchanged.

Scope: only the close path. No sidebar rows, groups, pins, top band, footer or animation code changed. This fixes our own regression from #23110.

Risk: a contained close-path behavior fix (the 10-23 rule for small sidebar behavior fixes, so no dogfood hold): a bug here would show as a multi-row close leaving rows open or a missing Undo; a single row keeps the old WorkspaceCloseFlow untouched.

Testing

SidebarSelectionBatchCloseTests (CmuxNextAppTests) runs on a same-tree cmux-tui daemon with the real window:

• fiftySelectedWorkspacesCloseTogetherAndUndoBringsThemAllBack: selects 50 new workspaces and sends the sidebar's close on the selection. It checks that every row leaves at the press, all 50 close, no refusal notice shows, exactly one Closed 50 workspaces toast shows, and its Undo reopens all 50. It prints the time from the press until all 50 are gone.
• aSelectionWithRunningProgramsAsksOnceAndClosesEveryRow: 9 selected rows (3 running sleep 600), Close from the row menu (closeWorkspace targeted at a selected row). It checks that exactly one question is asked and all 9 close.

Command: cmux-ci run --class light --script scripts/ci/package-test-lane.sh --arg=suite --arg=Packages/macOS/CmuxNext --arg=SidebarSelectionBatchCloseTests --env CMUX_CI_XCODE_APP=/Applications/Xcode_26.6.app --ref <sha>

• Red, test commit 35c72b0: pending (fleet light class, job 9d3d018aeb1b9b54d95ccb1a queued: every Xcode 26.6 suites worker is paused on host_lock or disk; an earlier run at this SHA was cancelled with no worker)
• Green, head e5b0044: pending (job f20561e500e8fe7765ad51dc queued, same reason)
• Timing for 50 selected rows, from the press until all are closed: pending the two runs above (the test prints SidebarSelectionBatchClose: 50 rows ... all closed after <duration>).

Fleet app compile (cmux-ci compile-check --app) passed at e5b0044 (job a0309d83b6ef613483ee34c1).

Review (Codex gpt-6.1-sol, xhigh, correctness first) took three rounds of fixes: rows left hidden after closing elsewhere, a double close of the kept row, an offline unhide before the store, offline rows missing from Undo, serialized process lookups, reservations held too long, and a stale undo after Undo. Final verdict: Review (codex, correctness first): SHIP at e5b004496945afce77ffc7047628fd498df0f9d0.

Localization: no new or changed strings. Close %lld workspaces? and Closed %lld workspaces already exist in every locale of Handlers.xcstrings.

Observed / Not run

• Observed: the base capture (02fa048) shows 11 selected rows running sleep on Cmd-W. The rows left, then one Close “Deletion 12”? question came up, one per row and queued behind each other. After the fix (a04752e), the same steps show one Close 10 workspaces? question and then a one-line Closed 10 workspaces Undo toast. A base capture of 24 idle rows closed them all eventually, but row by row, over several seconds.
• Observed: the fleet app compile passed at the head, and prepush-gate.py and check-concurrency.sh are clean.
• Not run: the offline (paused Cloud VM) branch has no test. It was reviewed by reading the code only. Release was not built.

Update after taking over (base merge, 2026-10-26)

Merged feat-cmux-next into the branch (no force-push). On today's base check-app-size.sh refused the 133 new CmuxNextApp source lines (SW14), so the batch flow moved into CmuxNextAppCore as WorkspaceSelectionCloseFlow (closures injected, like WorkspaceCloseFlow), with only its wiring left in the App (WorkspaceSelectionClose). BulkClosePlan, DaemonService.servesBulkClose and the multi-workspace close (now WorkspaceClose.close(_ members:)) moved with it. InstantClose keeps one begin over a list of ids; the toast's Undo hook is an undone: argument of WorkspaceGroupClose.offerUndo; Close Workspace's guard reads WorkspaceSelectionCloseFlow.canStart. No behavior change intended; CmuxNextApp is 107684 source lines vs 107685 on base.

• Fleet at c045ecc (cmux-ci run, package-test-lane): SidebarSelectionBatchCloseTests, BulkClosePlanTests, BulkCloseReopenTests, SidebarCloseSelectionTests, WorkspaceNothingToDoRefusalTests all passed. SidebarSelectionBatchClose: 50 rows, press held the main actor 0.042 s, all closed after 0.63 s, missed 0, toasts ["Closed 50 workspaces"], notices []; Undo reopened 50 of 50 after 1.19 s; 9 rows (3 running sleep) asked 1 question, missed 0.
• CI cmux-next swift test red: its failing suites are the known base reds (ActionRunCreated, AgentCursor*, BrowserProfileLinkMenu, BrowserTabMirror) and suites that fail the same way on unrelated PRs' runs; PaneHorizontalScrollClaimTests, SleepySceneCostTests and WindowZoomTests passed on the fleet at this head (CI-runner flakes).
• Review: Codex gpt-6-astra xhigh, SHIP at c045ecc.

Changelog

Fixed: Closing several selected sidebar workspaces closes them all at once, asks one "Close N workspaces?" question when a program is running, and shows one "Closed N workspaces" toast whose Undo brings them all back.

Proof

5677ebcc8e57a1e5cd4fab5b state=done worker=cmux-linux-pool-1022-065820-cdcca4c0 total_seconds=110.82 queue_and_setup_seconds=0.62 execution_seconds=110.21 free_gib=171.77->171.27
phase=admission seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=input_restore seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=build seconds=95.75 state=done cache_hit=unknown hits=unknown misses=unknown
phase=artifact_upload seconds=14.10 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_checkout seconds=7.63 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_submodules seconds=0.04 state=done cache_hit=unknown hits=unknown misses=unknown
phase=remote_tree seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_web seconds=1.18 state=done cache_hit=unknown hits=unknown misses=unknown
phase=app_exe_prekey_wait seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=swift_key seconds=0.08 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cache_swift_lane seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=switch_answer seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=app_exe_key seconds=0.11 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_app_exe_resolve seconds=3.54 state=done cache_hit=unknown hits=unknown misses=unknown
phase=swift_import_set seconds=0.54 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_app_exe seconds=46.04 state=done cache_hit=unknown hits=unknown misses=unknown
phase=swift_import_set_done seconds=0.18 state=done cache_hit=unknown hits=unknown misses=unknown
phase=app_exe_stage seconds=0.47 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_components seconds=14.08 state=done cache_hit=unknown hits=unknown misses=unknown
phase=component_web seconds=1.00 state=done cache_hit=False hits=unknown misses=unknown
phase=component_fetch_pinned seconds=0.40 state=done cache_hit=False hits=unknown misses=unknown
phase=component_hostd_build seconds=0.00 state=done cache_hit=True hits=unknown misses=unknown
phase=component_chief_build seconds=0.00 state=done cache_hit=True hits=unknown misses=unknown
phase=component_acpmux_build seconds=0.00 state=done cache_hit=True hits=unknown misses=unknown
phase=component_tui_fetch seconds=0.10 state=done cache_hit=True hits=unknown misses=unknown
phase=component_verify_web seconds=0.70 state=done cache_hit=False hits=unknown misses=unknown
phase=component_reject_providers seconds=0.00 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cli_resources seconds=0.30 state=done cache_hit=False hits=unknown misses=unknown
phase=component_diff_sidecar seconds=0.40 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cmux_tui seconds=3.90 state=done cache_hit=False hits=unknown misses=unknown
phase=component_acpmux seconds=0.00 state=done cache_hit=False hits=unknown misses=unknown
phase=component_optchat_chief seconds=0.00 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cnx_hostd seconds=0.00 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cef seconds=2.20 state=done cache_hit=False hits=unknown misses=unknown
phase=component_rb_host seconds=0.40 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cua_helper seconds=0.30 state=done cache_hit=False hits=unknown misses=unknown
phase=component_ghostty_resources seconds=0.30 state=done cache_hit=False hits=unknown misses=unknown
phase=component_server_helper seconds=1.70 state=done cache_hit=False hits=unknown misses=unknown
phase=component_frameworks seconds=0.60 state=done cache_hit=False hits=unknown misses=unknown
phase=component_coderouter seconds=0.10 state=done cache_hit=False hits=unknown misses=unknown
phase=component_cmux_cua seconds=0.90 state=done cache_hit=False hits=unknown misses=unknown
phase=component_ghostty_cli seconds=0.90 state=done cache_hit=False hits=unknown misses=unknown
phase=cmux_resources seconds=1.93 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_ghostty_resources seconds=0.41 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_web_guard seconds=0.80 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_parity seconds=0.11 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_parity_reference seconds=0.29 state=done cache_hit=unknown hits=unknown misses=unknown
phase=parity_reference seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_sign seconds=4.98 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cmux_package seconds=1.48 state=done cache_hit=unknown hits=unknown misses=unknown
phase=switch_answer seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=pool_prewarm seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=linux_pool seconds=0.00 state=done cache_hit=unknown hits=unknown misses=unknown
phase=cache_app seconds=0.38 state=done cache_hit=False hits=unknown misses=1
phase=cache_cmux_tui_source seconds=1.73 state=done cache_hit=True hits=1 misses=unknown
phase=cache_web seconds=1.40 state=done cache_hit=True hits=1 misses=unknown
phase=cache_app_exe seconds=0.06 state=done cache_hit=False hits=unknown misses=1
phase=cache_acpmux seconds=1.41 state=done cache_hit=True hits=1 misses=unknown
phase=cache_optchat_chief seconds=0.83 state=done cache_hit=True hits=1 misses=unknown
phase=cache_cnx_hostd seconds=0.86 state=done cache_hit=True hits=1 misses=unknown
phase=cache_summary seconds=0.00 state=done cache_hit=unknown hits=5 misses=2
remote install: published

Capture steps-4441620e5c at e5b004496945 (fleet build, capture mini, text checked for private data)

Light	Dark

Full quality: cap-e5b00449-steps-4441620e5c.mp4

The steps: 12 workspaces running sleep, Shift-Down to select 11 rows (the window keeps one), Cmd-W, then Close in the question. The video shows one Close 10 workspaces? question, then the one-line Closed 10 workspaces Undo toast.

Agent session: cosmic-gray

🤖 Generated with Claude Code

[github-actions]

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

Review c045ecc: 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=pr24341 -f ref=l5/batch-close -f platform=macos -f backend_mode=local

web and size checks: green

[teamleaderleo]

Bead cx-gzw4j created for this PR by the cmux coordinator.

[teamleaderleo]

Independent review by a different model: Codex gpt-6.1-sol at xhigh, correctness first, read-only from the diff and raw files at each head. It ran on cmux16 through acpmux.

The ask, in Leo's words: "also idk what's going on with these new workspace deletions but it's really slow and it's missing a LOT... and it's also 2 lines tall for some reason" / "no, it really is deletion not dictation."

• a04752e: FIX, 5 findings.
  • A row closed elsewhere during the question stayed hidden.
  • The kept row could be closed twice.
  • An offline close unhid its row before the store applied it.
  • Undo left out offline rows.
  • Process lookups ran one at a time.
  • All five fixed in eccb37a.
• eccb37a: FIX, 2 findings: reservations were held until every offline close ended, and a paused machine held back Undo. Fixed in 52e9596.
• 52e9596: FIX, 1 finding: a late offline close reused undo state that was already consumed. Fixed in e5b0044.
• e5b0044: no remaining findings. It checked that every entrypoint reaches the batch: Cmd-W, the row menu's Close, and the sidebar's Cmd-Delete. Close All in Group keeps its existing concurrent dispatch. The diff stays within the close path.

Fleet app compile at e5b0044: passed (job a0309d83b6ef613483ee34c1). Held for sidebar dogfood per Leo and Lawrence: not to be merged yet.

Review (codex gpt-6.1-sol, correctness first): SHIP at e5b0044

[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.

[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.

[teamleaderleo]

Owned by an agent session · session

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4032307340502 (35e1c1f88b)
• cmux NEXT nightly 1.0.0-nightly.4032901198601 (079a929b63)