[teamleaderleo]

HOLD: structural sidebar change: waits for a dogfood build for Leo to try

Leo-ask: ask-0dbbb0b936

Carry of #24287 (#24287).

Original purpose: “cmux NEXT sidebar: agent-started sessions nest under their parent by default.”

This carry keeps the original commits and authorship, integrates the current feat-cmux-next, and resolves the current sidebar feed/layout conflicts against the base's host-scoped session-parent and gap-handling changes. It preserves the default lineage-nesting setting, avoids GitHub reads when only lineage nesting is enabled, and retains tree fold/roll-up metadata for grouped rows.

Validation: merge completed cleanly; git diff --check on the carry delta has no new whitespace errors. Native validation belongs on the fleet.

Leo 2026-10-21 (ask-18a79e54a3, recorded in design/sidebar.md as default-on): "A LOT OF THESE SESSIONS ID IMAGINE DIDNT NECESSSARILY NEED TO SHOW UP ON THE SIDE BY DEFAULT". Lineage nesting stays on by default.

Drag fix (Astra review BLOCK): SidebarLayout.dragUnfolds returned false whenever the session tree was set, but the Done fold now runs with the tree and moves whole runs, so a drag of a Done row kept the folds and the drop gap opened where the row would not show. It now understands the tree: a drag unfolds when it carries a Done row, a row under a Done ancestor, or the parent of a Done row (the drag's base rows lack that parent, so the Done row would draw top-level and fold). The walk follows parents across lists, so it may unfold a drag the fold would not move; that drag draws as it did before drags kept the folds. Regression: SidebarLineageNestingTests.aDragOfADoneRowUnfoldsWithTheTree, committed red first ("test(sidebar): a drag of a Done row unfolds with lineage nesting (red)"), then the fix. plans/cmux-next/sidebar-owner.md gets the lineage entry and logs f0803b4, which had no Sidebar-Look: trailer.

Defaults: this PR intentionally changes tunable.sidebar.lineageNesting from none to true. design/sidebar.md records it as default-on from Leo's ask-18a79e54a3: "A LOT OF THESE SESSIONS ID IMAGINE DIDNT NECESSSARILY NEED TO SHOW UP ON THE SIDE BY DEFAULT". The defaults guard's change list matches this.

Reveal fix (router review BLOCK at 0614c4a): revealsSelection forced a reveal whenever the whole parents map changed, so an agent started or closed anywhere reopened a fold the person made above the selected row. Now only three things reopen folds above the selection: a new selection, lineage arriving for the first time (no parents before, some now), or a change in the selected row's own drawn ancestor chain (SidebarSelectionTreeReveal: a reparent, or a move to another list). Tests: SidebarLineageNestingTests.anUnrelatedSpawnOrCloseKeepsAFoldAboveTheSelection (committed red first), reparentingTheSelectionUnderAFoldRevealsIt, theFirstLineageRevealsTheSelection.

Performance: SidebarHugeGroupTests.oneGroup10k opened slower at 0614c4a than on base (862 ms vs 693 ms). That group has no lineage. The nesting path in it builds treeRows and a member lookup per layout, and a reveal pass after each structural change, which called listsHolding and rebuilt the model's row index (O(rows)). Two cheap trims are in: the reveal skips its walk while there is no lineage, and the group's tree lookup holds only parents (rows with a chevron), not all 10k members. treeRows stays O(members) per layout. Neither trim is measured yet; recheck this bench at the head after #28556 lands.

Merge of feat-cmux-next (9f6d9cd)

Merged origin/feat-cmux-next (17b7593) as a merge commit. Seven files conflicted; each keeps base's newer sidebar work beside lineage nesting:

• SidebarSessionTreeFeed: base's pull request capture fixture (sample badges, no GitHub reads) plus the nesting-only pass (shaped, no GitHub reads when only nesting is on).
• SidebarLayout+DoneFold: base's "Archived (N)" fold and Archive-section drag unfold, now folding whole runs as the Done fold does; dragUnfolds keeps its lineage walk and folds archived rows too.
• SidebarLayout+Fold: base's archived-group fast-path exits plus the lineage one.
• SidebarRowIndex / SidebarSessionTree: base's perf(sidebar): a folded session's step lays nothing out, only its roll-up state does #31783 (a folded session's step or title lays nothing out) with this PR's per-list fold check (patchChangesRollUp(..., lists:)).
• SidebarTreeRollUpView: base's running spinner ring (cmux NEXT: the working spinner is a faint ring with a turning sweep, not the 3/4 arc or spokes #31455) beside this PR's total-count chip.
• SidebarNestedFixture: both soleMember and lineageGroup scenarios.
• SidebarListView+Layout: a drag unfolds Done, Archived and the Archive section, lineage-aware.

One behavior call: base turned the Archive section off whenever the session tree was set. With nesting on by default the tree is set for everyone, so that would have removed the Archive section for every user. It now draws with the tree; a row whose parent moved to the archive draws as a root of its list. Test: SidebarArchiveSectionTests.theSessionTreeKeepsTheArchiveRow.

Fleet at 9f6d9cd: cmux-ci compile-check --package CmuxNext passed (job ff5be5652660ddb5941bf5ef). Swift suites on a fleet Mac (step 0eabab2bd3b8fc10e8beffb6), all passed: SidebarLineageNestingTests, SidebarSessionTreeTests, SidebarArchiveSectionTests, SidebarArchivedFoldTests, SidebarDeepTreeTests, SidebarDoneFoldTests, SidebarDragKeepsDoneFoldTests, SidebarLayoutFoldTests, SidebarFoldKeyboardTests, SidebarFoldAccessibilityTests.

Drop-gap fix (router review BLOCK at 9f6d9cd)

With the session tree a list draws its rows in tree preorder, but the drop gap for a tab dropped from a pane opened at the raw model index. Model [parent, other, child] draws [parent, child, other]. A drop before child opened its gap there, but the new row is a root, so it landed after other. SidebarLayout.gapEntry now covers nesting lists as well as folded ones. The gap opens before the first drawn top-level row at or past the gap's index; failing that, before the first fold row; failing that, after the list's last entry (gapAtEnd). This works for loose rows and group members. Test: SidebarLineageNestingTests.aDropGapOpensWhereTheDroppedRowWillDraw checks that the gap's y equals the dropped row's y after the drop, for every index, loose and grouped. It failed at 46d224f (red commit, fleet step bc265bac1251980375083cda: 52 pt apart) and passes at 4a49300.

Fleet at 4a49300: compile-check --package CmuxNext passed (job 6cb56d6194019113441d448f). Swift suites (step 062a61e564cf192a227610d0) all passed: the ten above plus SidebarUniformRowSpacingTests and SidebarSingleMemberGroupTests.

Needs-you tray fix (router review BLOCK at 4a49300)

Lineage nesting sets the session tree for everyone, and the needs-you tray drew whenever a tree was set. So with sidebar.sessionTree off, blocked, failed or unread sessions added the tray and its bottom inset. SidebarNeedsYou.trayTree now passes the tree to the tray only while sidebar.sessionTree is on. Test: SidebarNeedsYouTests.nestingAloneDrawsNoTray.

Fleet at 5d04973: compile-check --package CmuxNext passed (job 00a879738ce5979d57d6e788). Swift suites (step 13a7bdee9db6412b9a77b8da) all passed: the twelve above plus SidebarNeedsYouTests.

Observed / Not run

• Observed: prepush-gate.py --only sidebar over the PR's commits reports 0 findings (it named f0803b4 before); the full pre-push gate reports no new violations.
• Not run: the new capture state (fleet only). Swift compile and the sidebar fold/lineage suites ran on the fleet at 9f6d9cd (above).

Captures

Viewed after image: lineage-after-17c0923e-light.png below. Release week shows its two agent-started children, Settings pass and Release notes, one level in; each folds its own child behind a 1 count chip with the state roll-up. The before (4c12180f, nesting off) lists the same rows flat. These are real fleet captures from the carried PR #24287, at 17c0923 (a commit in this branch), of the needsYou fixture's loose rows, with sidebar.sessionTree=false. They are not stills of this head. This lane VM can't run cmux-ci capture.

Fresh stills of lineage nesting in a group are pending a fleet capture. The new state shows a group with a parent and two children, one folded, first with sidebar.lineageNesting off and then on, in one build:
cmux-ci capture <head sha> --state scripts/cmux-next/media/states/sidebar-lineage-group.json --pr 28103
The sidebar is native AppKit, so the webview gallery doesn't cover it.

Before (4c12180, flat)	After (17c0923, nested)

Performance

SidebarHugeGroupTests.oneGroup10k, head cc6384a vs base 7050ae30, fleet runs (head and base could not share a host: --target-worker pins fall back after about 10 minutes queued).

metric	before (base)	after (head)	delta	harness
open	base host (M4 Pro)	head host (M4 Pro, matched)	-0.7%	oneGroup10k, matched M4 Pro hardware
fold	base host (M4 Pro)	head host (M4 Pro, matched)	+0.3%	oneGroup10k, matched M4 Pro hardware
unfold	base host (M4 Pro)	head host (M4 Pro, matched)	+0.2%	oneGroup10k, matched M4 Pro hardware
reorder	base host (M4 Pro)	head host (M4 Pro, matched)	-2.4%	oneGroup10k, matched M4 Pro hardware
all four	a Mac mini	second head host	head 7-12% faster	within the 6-12% host-to-host spread

The head is no slower. The earlier 27-35% gap was the host, not this change. The four ceilings (open<540, unfold<175, fold<80, reorder<290 ms) fail on both head and base on every host, so they are a base red. The four group tests (groupHeaderPlusMakesTheNewWorkspaceInThatGroup, pinGroupFromTheHeaderMenuPinsAndUnpins, groupMovesAtTheEdgeSayWhy, newWorkspaceInGroupIsAwaitedByTheCLI) and the GroupHeaderNewWorkspace, WorkspaceGroupPinLive and Sidebar/WorkspaceNothingToDoRefusal suites pass at head and base. The base results come from one run on a Mac mini. The earlier full-suite reds were load flakes.

Agent session: agent session
Agent session: agent session

[github-actions]

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

Review 10382f9: 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=pr28103 -f ref=lane/carry-24287-sidebar -f platform=macos -f backend_mode=local

red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on 4165f6f to compare; do not merge until green; see lane-rules

Default changes

This pull request changes 2 defaults at 10382f922ea3 (what everyone gets without configuring anything):

• tunable-source.CmuxNextSidebar/Logic/SidebarTunables.swift:lineageNesting: (none) -> "5c13d502a1e4479873eaf73635b82a2e119850427c8983e8307dce69f0f08ece"
• tunable.sidebar.lineageNesting: (none) -> true

Signed off on this commit by review (codex, router-review), carried from 5d049730188d across base merges.

Regenerate the snapshot with python3 scripts/ci/defaults_guard.py snapshot.

[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]

Review (codex gpt-6-astra high, next-review): BLOCK at 96f7c85

• Packages/macOS/CmuxNext/Sources/CmuxNextSidebar/Logic/SidebarLayout+DoneFold.swift:88: dragUnfolds still returns false whenever o.sessionTree is non-nil, but this PR now applies Done folding with the session tree. Dragging a Done workspace therefore removes it from the tree and recomputes/removes the Done (N) row during the drag, shifting rows and invalidating the drag geometry/drop gap. The session-tree guard must be removed or made tree-aware.

[teamleaderleo]

Owned by an agent session · took over from an earlier session · session

[teamleaderleo]

Awareness note from an agent lane that does not own this PR; it is not a review or a request to take the PR over.

---

This PR is the build for Leo's ask ask-0dbbb0b936 (2026-10-21): he saw sub-sessions nested under their parent in the acpmux dashboard and asked why his actual cmux NEXT nightly sidebar doesn't do that. The ask is 12 days past its 1 h SLA. Current state: CONFLICTING with feat-cmux-next, no review at head cc6384a after the 0614c4a BLOCK (manual folds reopened by unrelated lineage updates), last commit 2026-10-30 22:51Z. Owner: please merge base, address the BLOCK, get a review at the new head and land it, or release it (pr-owner reports you alive, so nobody else can claim it). Add Leo-ask: ask-0dbbb0b936 to the body.

[teamleaderleo]

Review (codex gpt-6-astra high, router-review): BLOCK at 9f6d9cd

BLOCK at 9f6d9cd

Visual verdict: supplied before/after pairs show the intended nesting; no fresh head capture [follow-up].

1. [P1][blocking] Tree ordering breaks external drop-gap placement. treeRows now emits preorder (SidebarLayout+SessionTree.swift:110), while SidebarLayout.swift:416 and :462 compare raw sibling indices as if drawn order were model order. With model order [parent, unrelated, child], the tree draws [parent, child, unrelated]; dropping a new tab before child resolves to index 2, so the gap opens before child, but insertion at raw index 2 yields [parent, unrelated, new, child], which draws [parent, child, unrelated, new]. The row jumps away from the visible gap on drop. Resolve gaps at tree-run boundaries or translate the target index through the drawn tree.

Follow-ups:

• [follow-up] Add a fresh capture for head 9f6d9cda787fe5b9a08c7b3d1967e84728a91793.

[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]

Review (codex gpt-6.1-sol xhigh, router-review): BLOCK at 4a49300

BLOCK at 4a49300

Visual verdict: no images; Leo tries the build

1. [P1][blocking] SidebarSessionTreeFeed.shaped now publishes a nonnil tree whenever lineage nesting is enabled (SidebarSessionTreeFeed.swift:91-94). The existing SidebarView+NeedsYou.refreshNeedsYou treats any nonnil tree as enabling the Needs You tray, so with the default lineageNesting=true and sidebar.sessionTree=false, blocked, failed, or unread sessions now add the tray and bottom inset. The design specifies this tray is sidebar.sessionTree-only, and the PR keeps the rest of that feature opt-in. Gate the tray on the session-tree preference (or provide a separate capability).

[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]

Review (codex, router-review): SHIP at 5d04973

Reviewer: codex gpt-6-astra high, fast tier.

SHIP at 5d04973

Visual verdict: stale pre-head after images show the expected nesting; no fresh head capture, Leo tries the build.

Reviewed the changed lineage, fold, drag-gap, tray, feed, and selection-reveal paths plus targeted tests. No correctness findings.

Follow-ups: none

[blacksmith-sh]

Found 43 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp1uceec3f/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpjkhaedua/summary-missing.json (live):/summary-missing.json (live):	View Logs
[new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows/ [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows	View Logs
[new_workspaces] the window draws no frames/[new_workspaces] the window draws no frames	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
main.DeveloperIdAppTests.test_download_prints_the_active_profile_only/ test_download_prints_the_active_profile_only	View Logs
convention violations found (ERROR lines above)./ convention violations found (ERROR lines above).	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

...and 33 more test failures. View all on Blacksmith

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