merge: sync upstream through b73232bdd (15 commits)

[omegent-app]

Adopts upstream through b73232bdd — the 15 commits after 083fa4ab2.

Important

Merge with a merge commit, not squash. Squashing collapses the second parent and discards
upstream's commits as ancestors — the thing that keeps "commits behind upstream" honest.

Upstream commits entering the product

Commit
b73232bdd	feat(web): reset sidebar width on double click (pingdotgg#6320)
f131228a5	fix(web): theme Clerk surfaces (pingdotgg#6300)
e321667b1	fix(web): prevent changed files header overlap (pingdotgg#6314)
849bac894	fix(connect): preserve CLI OAuth parameters through browser sign-in (pingdotgg#6285)
6fd088af9	fix(web): align mobile onboarding header (pingdotgg#6293)
b54bfc931	feat(web): a better right panel empty state (pingdotgg#6258)
e1378a1f4	fix(mobile): keep ordered lists inside user bubbles (pingdotgg#6154)
5a8461480	fix(web): align the composer model picker (pingdotgg#6252)
d37a9b09b	feat(mobile): add thread title regeneration (pingdotgg#6253)
560d4a456	fix(web): keep sidebar wordmark visible at minimum width (pingdotgg#6246)
52e5a75a8	feat(web): compact sidebar footer actions (pingdotgg#6210)
c196f422e	fix(web): clean up composer resize animation (pingdotgg#6209)
f0b57ca23	feat(web): add Open VSX theme search (pingdotgg#5654)
2db08457f	fix(web): use upload icon for disabled push action (pingdotgg#6207)
63e6faef6	chore: add dara to vouched (pingdotgg#6259) — the fork's VOUCHED.td deletion stands

Resolutions

Nine conflicts, almost all of them pingdotgg#6253 (mobile title regeneration) landing in the same files the
fork has been extending. Three needed judgement:

Where	Situation	Resolution
thread-list-items.tsx	the union produced two menuActions declarations — the fork's settle-aware builder and upstream's regeneration list — and two stacked dependency arrays on one useCallback, which typechecked as a three-argument call rather than failing outright	buildThreadRowMenuActions now takes the regeneration items and splices them in, on both the settlement-supported and legacy branches, so a row has one menu carrying both features. Dependency arrays merged
HomeRouteScreen.tsx	upstream passes onEnvironmentChange={setSelectedEnvironmentId} — single-environment selection. The fork replaced that surface with a multi-select	upstream's prop dropped; the fork's onToggleEnvironment / onClearEnvironments are already wired beside it
useThreadListActions.ts	upstream extracted actionFailureMessage / ThreadListAction into threadActionMessages; the fork still declared them locally, so the import collided with the local declaration	local copies removed — the extracted module already carries the fork's settle / unsettle actions, so it is a true superset

ChangedFilesTree takes upstream's @container/changed-files inside the fork's cn() so the
caller-supplied className survives; the rest are additive unions.

Note on the Pull Requests button

pingdotgg#6210 reshapes the sidebar footer into compact icon buttons, and it landed here without a
conflict. I checked it deliberately, since that file holds the button you could not find: upstream's
change applied intact, and the gate is still usePrimaryEnvironment() only (SidebarChrome.tsx:131).
So the button now looks different, but the reason it stays hidden when your primary environment is
old or disconnected is unchanged — still worth its own fix.

Adversarial review

Both reviewers went through all the resolutions. Neither found a HIGH. Four findings were real
and are fixed in f8948145b and 3d263e1dc.

Finding	Disposition
both, MEDIUM — renderV2Item's dependency array lost nowMinute, which upstream listed. Unioning conflict hunks keeps both sides' code but silently keeps only one side's dependency array	Fixed. A screen left mounted across a minute boundary kept handing rows a stale minute, so snooze presets drift from the times they advertise
grok-4.5, MEDIUM — the same array also omitted snoozeEnvironmentIds, which the row reads for snoozeSupported	Fixed. A row could keep offering — or withholding — snooze after the environment's capability resolved
gpt-5.6-sol, LOW — the v2 layout memo reads props.threadGrouping but did not depend on it	Fixed. Switching grouping could keep recency headers until some other input changed identity
gpt-5.6-sol, LOW — CONTRIBUTING.md:13 still links to .github/VOUCHED.td, which the fork deletes. My merge message claimed nothing referenced it, and that was wrong	Fixed. The link is in the contributing guide, so anyone following the trust-status docs hit a 404
grok-4.5, MEDIUM — upstream's showDesktopUpdateDownloadedToast is never called	Not this merge. Verified against both parents: upstream calls it from LegacySidebar.tsx:3552, and the fork parent already did not — a pre-existing fork divergence, not something this weld dropped. Worth its own look
grok-4.5, LOW — regenerate sits above Archive on the settlement branch rather than upstream's exact relative slot	Cosmetic ordering inside a menu the fork already reorders

Both independently confirmed the parts I most wanted checked: every menu entry in
thread-list-items is reachable and dispatched with no duplicates or dead handlers; dropping
onEnvironmentChange is correct because HomeScreenProps requires the multi-select trio instead and
no upstream selection surface became unreachable; threadActionMessages is an exact superset of the
fork's local copies including the settle/unsettle wording; @container/changed-files appears once on
the real query container with the caller's className intact; and main.tsx uses both the fork's
preload recovery and upstream's Clerk appearance.

On SidebarChrome.tsx — the file behind the missing Pull Requests button, which did not
conflict: git diff HEAD^2 HEAD on it is empty, so the merge result is upstream's tip exactly.
pingdotgg#6210's compact footer landed intact, and the gate is still usePrimaryEnvironment()-only at
:131-133. Notably both reviewers confirmed upstream uses that same gate — so it is upstream's
own single-environment assumption, not a fork-side removal.

Verification

• Recursive typecheck clean across all 18 packages; vp build of the web app passes.
• 2,763 tests pass; the only failure is the pre-existing CodexTextGeneration launch-args one,
  which reproduces on unmerged fork/dev.
• pnpm-lock.yaml regenerated rather than hand-merged.

Note

A full-suite run surfaced the pre-existing browserHistoryStore.test.ts flake (~1 in 3). Not from
this branch — reproduced earlier on clean fork/dev and at f067b34a1.

Co-authored by @patroza

opened by Patrick Roza in chat thread Discord · Discord · T3