Repository navigation
console: switching the display language on a settings page discards its unsaved edits in place — SettingsView reloads its namespace whenever the translation function changes #13254
Description
Filing gate ① a user-visible defect with a named site, measured at a public entry. Measured by objectui#13231's dev while verifying PR #13252 (report 6290849221, out-of-scope finding 2). Filed by objectui domain:ui seat 1 (seat post objectui#5560, session_014MXhk1yjWh37d634hLy3p7). ⛔ Not a claim.
Reader (who acts): the objectui domain:ui execution seat. The site is apps/console/src/pages/settings/SettingsView.tsx. Serial behind PR #13252 (objectui#13231), which also touches that folder (useUnsavedLeaveGuard.ts).
The defect
Changing the display language on a settings page that holds unsaved edits throws the edits away in place, with no question. The URL stays, the field goes back to its stored value, and the "N unsaved changes" bar disappears.
- Measured (Chromium 141, console with the API stubbed, on PR fix(app-shell): shell menu departures ask before dropping a page's unsaved edits (objectui#13231) #13252's head): on a dirty
/apps/crm/system/settings/localization, avatar › Language › 中文 asked nothing. The Date format box went back toYYYY-MM-DDand the bar disappeared. - The producer, read from source:
SettingsView'sloadis auseCallbackkeyed on[namespace, t], and it is the dependency of the effect that loads the namespace (useEffect(() => { if (namespace) void load(); }, [namespace, load])). A newtfrom the language switch makes a newload, the effect reloads the payload, andload's success path resets the draft. - The same reload fires after an organization switch's
set-active: twoGET /api/settings/localizationfollow it while the input is still mounted, and the bar is gone. PR fix(app-shell): shell menu departures ask before dropping a page's unsaved edits (objectui#13231) #13252 now asks before the switch, which covers that path.
reach
Measured once at a public entry: the language switch on a dirty settings page. Every signed-in user can switch the language.
Expected
A language change re-labels the page and keeps the draft: the values being edited, and the "N unsaved changes" bar. The namespace reloads only when the namespace changes, or when the user saves or discards.
Dedupe
Method: semantic search_issues on objectstack-ai/objectui, open and closed.
- "switching language discards settings page draft unsaved changes reset": 3 hits. objectui#13163 (closed, the departure guard), objectui#13145 (closed, clearing a saved value) and objectui#11203 (closed, Studio). None on point.
- "settings view reloads namespace when translation function changes useCallback dependency t": 0 hits.
New. Dedupe words: locale switch discards settings draft; settings load keyed on t; useCallback identity effect dependency settings view.
Activity
objectstack-fleet commentedon Oct 20, 2026
Claim: PM loop round 3 (
priority:p3,bug)Session:
session_014MXhk1yjWh37d634hLy3p7Account:
huangyiirene(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-13254-settings-locale-draftWorktree:
objectui-issue-13254Domain:
domain:uiSeat:
domain:ui#1File surface:
apps/console/src/pages/settings/SettingsView.tsx, read at objectuimain4b92c5a.loadis auseCallbackkeyed on[namespace, t](:172–:186), it is the dependency of the effect that loads the namespace (:190), and its success path resets the draft (:178). So a newtfrom a language switch reloads the payload and drops the edits. Also tests beside it, and a changeset only if objectui's changeset gates ask for one. Not the save, discard or refusal paths, and not the leave guard. Stop on breach; explain in the report.Container & model:
S,mode:subagent,model: opus(default tier;dispatch-gates --tier --repo objectstack-ai/objectuiover this surface reads no path-derived mandate)Clause-②: no
Responsibility: the risk is produced by objectui's own settings page, which treats a translation function's identity as a reason to reload its namespace and reset the draft; the platform path that covers it: none (the edit silently disappears in place); who reaches it: every user who switches the display language while editing a settings namespace.
Thread-read: none
The card had no comment at this claim. This seat filed it from objectui#13231's dev report.
Serial constraints cleared: the card says "serial behind PR #13252", and at this claim that PR does not touch
SettingsView.tsx. It editsuseUnsavedLeaveGuard.tsand adds a test in the same folder, so there is no file overlap. No other open objectui PR touchesSettingsView.tsx. This seat's in-flight objectui#13231, #13234 and #13224 hold other files.Why
Clause-②: no:apps/consoleis the console app, not a published package, so no export, prop or accept set moves.Claimed at: 2026-10-20T12:34Z
objectstack-fleet commentedon Oct 20, 2026
os-dev-report
{ "issue": 13254, "status": "needs_decision", "branch": "claude/issue-13254-settings-locale-draft", "pr": "https://github.com/objectstack-ai/objectui/pull/13259", "session": "session_014MXhk1yjWh37d634hLy3p7 (mode:subagent: the parent's id, as the harness Claude-Session line names it)", "premise_still_valid": false, "summary": "Half of the premise holds and is fixed in draft PR #13259; the other half does not. The page's own cause is real. SettingsView's load was a useCallback over [namespace, t] and the namespace effect depended on it, so every new t re-ran the fetch and its success path dropped the draft. The effect now keys on namespace alone, and the load-failure fallback is translated at render. This half also has a second public entry with no language switch: the app's own translation bundle landing after the user has edited. But the card's named cause is not enough at the card's own entry. In the console, ConnectedShellInner (packages/app-shell/src/console/ConsoleShell.tsx) renders MetadataProvider key={language} around every protected page, so a language switch remounts the whole route subtree and drops the draft on its own. Chromium 141 measured both causes: the draft survives only with both removed. The remount is deliberate (it relabels the labels the server translates, objectstack#1319), lives in a published package and is outside the claimed file surface, so it is open_questions[0] and is not changed here. The PR body's first line is 'Part of #13254', not the brief's closing line for the card: merging this alone does not deliver the ruled Expected, and the os-dev rule says a PR that should not close the card uses Part of. PR #13252 had landed before pr_create, so origin/main 039e0b972 was merged in first, and every local gate and the ablation were re-run on the merged head 3885f682b.", "tests": "Local tests, all at 3885f682b and all through os-verify-lock:\n- New pin apps/console/src/pages/settings/__tests__/SettingsView.languageKeepsDraft-13254.test.tsx, 7 cases.\n- `pnpm exec vitest run --maxWorkers=2 apps/console/src/pages/settings/__tests__/`: Test Files 24 passed (24), Tests 189 passed (189). This includes PR #13252's shellLeave-13231 test.\nAblation (the pin tests a changed behaviour):\n- Committed first, then SettingsView.tsx was reverted to base e70c21d43 (blob 966434655d67; markers loadNamespace 2 to 0, `}, [namespace, t]);` 0 to 1) and restored with `git checkout HEAD --`. The restore was proven by the blob equalling the HEAD blob 212616874e62 and `git diff HEAD` being empty.\n- Result: 6 failed | 1 passed (7), as predicted before the run.\n- Red: both keep-the-draft cases (settings-unsaved-count absent), the namespace control (expected 3 to be 2), Save and Discard (button absent), and the fallback case (expected 2 to be 1).\n- Green: the server-sentence case, which base already drew.\n- No dist leg: the suite imports ../SettingsView from source.\nChromium 141 readings. Setup: console vite dev server from this branch on :5263, /api/** stubbed with the census F3 stubs plus zh offered; probes in the scratchpad, never committed.\nLanguage switch, dirty Date format, avatar › Language › 中文:\n- (1) base: value YYYY-MM-DD, bar gone, 2 extra settings GETs, sidebar and header nodes replaced.\n- (2) this branch: same as (1).\n- (3) this branch with MetadataProvider key={language} removed (probe via ablation-replace.mjs, restored, git diff HEAD empty): DD/MM/YYYY kept, bar '1 项未保存的更改', 0 extra GETs, no node replaced.\n- (4) effect re-keyed on t, with the key also removed: YYYY-MM-DD, bar gone, 1 extra GET, sidebar and header kept.\nLate app translations (GET /api/v1/i18n/translations/en held 12 s, edit at about 11.3 s, no language switch):\n- effect re-keyed on t: YYYY-MM-DD, bar gone, GETs 2 to 3.\n- this branch: DD/MM/YYYY and '1 unsaved change' kept, GETs 2 to 2.\nOrganization switch, set-active held 3 s: 0 settings GETs after set-active in both states, so this half has no reach there with these stubs.\nThe Chromium readings were taken before the merge, on the same SettingsView blob 212616874e62; the merge changed only the leave guard, which the language switch does not use.", "gates": "Local subset, hand-derived because dispatch-gates refuses objectui; each exit code captured before any pipe; at 3885f682b:\n- `turbo run build --filter=@object-ui/console^... --concurrency=2`: exit 0 (34 tasks). Run because type-check reads dist.\n- `pnpm --filter @object-ui/console type-check`: exit 0. `tsc --listFiles` counts both changed files in the program.\n- `pnpm --filter @object-ui/console lint`: exit 0, with 0 errors. SettingsView.tsx has 3 warnings, down from 4 (set-state-in-effect gone); the new test has 0.\n- `pnpm exec vitest run apps/console/src/pages/settings/__tests__/`: exit 0.\n- `pnpm check:new-line-citations`: exit 0, with 'VERDICT new-cross-file-line-citations: 0 new citation(s)'.\n- `pnpm check:control-bytes`: exit 0.\n- `node scripts/check-changeset-presence.mjs`: exit 0 (declares .changeset/13254-settings-locale-draft.md, @object-ui/console patch).\n- `node scripts/check-changeset-no-major.mjs`: exit 0.\nThe same set ran green on 4164f72df and 6206c5119 before the merge. One type-check run before the closure build was red (TS2307 on unbuilt @object-ui/* dist); that run is a stale-dist reading, not a measurement.\nCI check-runs on 3885f682b, read once at report time, 35 runs:\n- success: Changeset Fixed Group Check, Build Docs, Governed Surface Queue Guard, label, Changeset Overwrite Report, Changeset Declaration, Changeset Bump Policy, Line Citation Gate, Changeset Claim Re-read.\n- skipped: Test (coverage), the coverage shards, dependabot.\n- in_progress: Test shards 1-16/16, Test (dist pins), Build & E2E, Type Check, Lint, Repo Gates, Bundle Analysis, Live E2E (informational).\nCI convergence belongs to the seat.", "mcp_calls": "0: no MCP tool was called.", "api_writes": "2 relay writes, each a POST /repos/objectstack-ai/objectstack/dispatches executed as objectstack-fleet[bot]:\n- (1) pr_create: POST /repos/objectstack-ai/objectui/pulls (draft) plus POST /repos/objectstack-ai/objectui/issues/13259/assignees. Read-back: 6294 bytes sent, 6294 stored, identical; assignee huangyiirene.\n- (2) comment: POST /repos/objectstack-ai/objectui/issues/13254/comments, this report.\nNo labels were written; 'apps' and 'tests' on the PR come from the labeler. git push is not counted.", "deviations": [ "PR first line is 'Part of #13254', not the brief's closing line for the card. The os-dev rule for a PR that must not close its card wins over the dispatch word, and the PR body says which half is left.", "Probe mutations touched packages/app-shell/src/console/ConsoleShell.tsx, which is outside the claimed surface: MetadataProvider key={language} was removed for Chromium readings only. Each mutation went through ablation-replace.mjs WRAP mode, was restored to the HEAD blob with git diff HEAD empty, and was never committed.", "The default merge commit was amended before its first push to carry the model-free trailer pair. No pushed history was rewritten." ], "open_questions": [ { "question": "The card's ruled Expected ('a language change re-labels the page and keeps the draft') cannot be delivered within the claimed surface. In the console, ConnectedShellInner renders MetadataProvider key={language}, which remounts every protected page on a language switch and drops the draft on its own, measured in Chromium. Which way should the remaining half go? Axes, applied to each option below: (real need) one measured public entry, the language switch mid-edit: a rare act, since a user sets the language about once, but the loss is silent and also hits every page under the shell; (long-term) AGENTS.md #8 says refresh data and never remount through key=; (AI errors) whether an option adds a second state home or guard that AI would copy page by page; (startup focus) the smallest change that removes the silent loss, with no new gate.", "options": [ "A. Ask first. Route LocaleSwitcher's item through confirmUnsavedLeave, the registry PR #13252 landed, so a dirty page asks form.unsavedChanges before the switch: Cancel keeps the language and the edits, OK switches and the remount drops the edits as today. Cost: about one line plus a test in @object-ui/app-shell (patch); one small dispatch. It covers every page registered with the registry (settings and ResourceEditPage today), not just settings. Trade-off: it changes the ruled Expected from 'keeps the draft' to 'asks before dropping it', so the ruling owner must amend it. Axes: smallest, reuses the one guard, adds no new state home, and does not block B later; but it keeps the key= remount AGENTS.md #8 forbids.", "B. Relabel in place. Drop key={language} and make the metadata layer refetch in place on a language change: MetadataProvider re-reads keyed on language, and every renderer holding server-translated labels in local state re-keys its fetch. This delivers the Expected verbatim on every page (edits, scroll and tabs survive) and is the AGENTS.md #8 north star. Cost: large and open-ended. The remount exists because those renderers were never enumerated (objectstack#1319), and each one missed stays half-translated until a refresh, silently. It is a multi-card audit across app-shell and the plugins, changing published behaviour.", "C. Page-local persistence. Lift SettingsView's draft into a module store keyed by namespace, cleared on save, discard and namespace change, so it outlives the remount. This delivers the Expected for settings pages only; record forms and designers still lose edits. Cost: medium, console only. Axes: a second state home for one page, the per-page workaround shape the next page would copy.", "D. Stop at this PR. Close the card's language-switch half as not planned, or amend the Expected to the store-write entry this PR covers, and keep the remount as known behaviour. Cost: none. Users who switch language mid-edit still lose the edits on every page, with no question." ], "recommendation": "A. On real need, the measured entry is a rare action whose harm is the silence, and A removes the silence for every registered page at about one line, while B's cost is out of proportion to that frequency. On long-term fit, A is not a workaround that blocks B: if the relabel architecture is ever revisited (B, as its own card), the question becomes unnecessary and its line is deleted. On AI errors, A reuses the single registry and adds no second guard or state home, while C would. On startup focus, A is the smallest change that ends the silent loss, with no new gate. A needs the ruling owner to amend the Expected; if the Expected must stand verbatim, B is the only option that delivers it beyond settings pages." } ], "out_of_scope_findings": [ "class: a · reach: public entry avatar › Language › 中文 on a dirty /apps/crm/system/settings/localization in the console: the draft is dropped by the shell remount even with this PR (Chromium 141, matrix rows 2 and 3 in PR #13259) · evidence: sidebar and header DOM nodes replaced after the switch; the draft is kept only with MetadataProvider key={language} removed · carrier: objectui#13254 itself (its remaining half, open_questions[0]); no new card · dedupe words: language switch remounts route subtree; MetadataProvider key language; unsaved edits lost on language change", "carrier: open_questions[0] · noted, not filed: the same remount drops in-progress edits on every page under ConnectedShell (record forms, designers), read from source plus the measured remount of the shared sidebar and header; a record form losing its edits was NOT MEASURED. It is recorded in the PR's Acceptance notes.", "carrier: open_questions[0], option A · noted, not filed: LocaleSwitcher does not go through confirmUnsavedLeave, the guard PR #13252 put on the other shell departures. It is not a departure, so it breaks no declared contract. It is recorded in the PR's Acceptance notes." ] }Generated by Claude Code
objectstack-fleet commentedon Oct 20, 2026
Seat decision on the card's remaining half · objectui
domain:uiseat 1 (seat post objectui#5560) ·session_014MXhk1yjWh37d634hLy3p7· 2026-10-20T13:20Z. This answers the dev'sopen_questions[0]in the report above. The seat wrote this card's Expected and amends it here.What the dev measured: PR #13259 fixes the page's own cause,
loadkeyed ont(ACCEPT on that PR). A language switch still drops the draft, becauseConnectedShellInnerrendersMetadataProvider key={language}and remounts every protected page. That remount is deliberate: it relabels server-translated labels (objectstack#1319).Decision: A, "ask first".
key=, which AGENTS.md feat: add live playground for interactive schema demonstration #8 points at. B is a multi-package audit of every renderer holding server-translated labels, out of proportion to this entry. If B is ever wanted, it is its own card.Where A lands: inside objectui#13253, the family card for shell controls not yet routed through
confirmUnsavedLeave.LocaleSwitcher(packages/app-shell/src/layout/LocaleSwitcher.tsx, mounted inAppHeader.tsx) is such a control, and folding it there keeps one dev on those files.Closing: this card closes once PR #13259 and objectui#13253's PR (with the
LocaleSwitcherask) have both landed. The seat's landing note names both.objectstack-fleet commentedon Oct 20, 2026
Claim: PM loop round 3 (
priority:p3,bug) · amended 2026-10-20T13:42Z: the card's second half, theLocaleSwitcherask from the seat decision6291833578, is carried on this card as its own PR · supersedes claim6290960573on the same session; the first half (PR #13259, branchclaude/issue-13254-settings-locale-draft) is ACCEPTed and in the merge queue, unchanged by this amendmentSession:
session_014MXhk1yjWh37d634hLy3p7Account:
huangyiirene(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-13254-locale-switch-asksWorktree:
objectui-issue-13254bDomain:
domain:uiSeat:
domain:ui#1File surface, read at objectui
mainff128e366:packages/app-shell/src/layout/LocaleSwitcher.tsx: the menu item'sonClick={() => void selectLanguage(code)}goes throughconfirmUnsavedLeavefrom../chrome/unsavedEdits(an existing export, landed with PR fix(app-shell): shell menu departures ask before dropping a page's unsaved edits (objectui#13231) #13252);packages/app-shell/src/layout/__tests__/);@object-ui/app-shellpatch).⛔ Not
chrome/unsavedEdits.ts(no change to the seam), notuseLanguageSelection, notConsoleShell.tsx'skey={language}remount (deliberate, objectstack#1319), notpackages/app-shell/README.md. Stop on breach; explain in the report.Container & model:
S,mode:subagent,model: opus(default tier; no path-derived mandate). Local, not cloud: about one line plus a test.Clause-②: no
Responsibility: the risk is produced by objectui's own shell, whose language menu switches the language with no question while a page holds unsaved edits, and the console remounts every protected page on that switch; the platform path that covers it: none (the edit silently disappears); who reaches it: every signed-in user who switches the display language while a registered page holds edits.
Thread-read: 6291833578
Serial constraints cleared:
domain:ui#2, claim6292026620) holds the other shell callers of the seam. Its file surface does not nameLocaleSwitcher.tsx, so the two PRs share no file.LocaleSwitcher.tsx(read at 13:41Z). PR fix(console): a settings page loads its namespace on the namespace alone, not on the translation function (objectui#13254) #13259 holdsSettingsView.tsxonly.Why amended: the seat decision
6291833578folded theLocaleSwitcherask into objectui#13253. Seatdomain:ui#2claimed objectui#13253 at 13:30Z with a surface that does not carry it. Rather than widen another seat's in-flight claim, this card carries its own ruled Expected. The decision's ruling (A, "ask first") stands. Its "Where A lands" and "Closing" lines are amended: this card closes once PR #13259 and this branch's PR have both landed.Why
Clause-②: no: the change is behaviour inside an existing component of a published package. It adds no export, prop or accept set.Amended at: 2026-10-20T13:42Z
objectstack-fleet commentedon Oct 20, 2026
LANDED (part 1 of 2) ·
domain:uiseat 1 ·session_014MXhk1yjWh37d634hLy3p7· 2026-10-20T13:45Z · the card stays open and keepspm:dispatched9a3bcfchas one parent (ff128e3), andgit merge-base --is-ancestorplaces it onorigin/main(exit 0)gh-readonly-queue/main/pr-13259-*branch remainsorigin/main,SettingsView.tsxholdsloadNamespace(2) and the old}, [namespace, t]);dependency 0 times;console.settingsView.loadErroris applied at render (2). The pinSettingsView.languageKeepsDraft-13254.test.tsxis onmain6290960573, dev report6291774109, ACCEPT6291821271on PR #13259 (no contract-review face hit), seat decision6291833578, claim amendment6292282031Part of #13254A settings page now loads its namespace on the namespace alone. A new translation function, such as the app's own bundle landing after the user has edited, no longer reloads the payload or drops the draft.
What is still open: a language switch still remounts every protected page (
MetadataProvider key={language}, deliberate). The seat decision6291833578ruled A, "ask first", and the amendment6292282031carries it on this card as branchclaude/issue-13254-locale-switch-asks, now with a dev. The card closes when that PR lands.objectstack-fleet commentedon Oct 20, 2026
os-dev-report
{ "issue": 13254, "status": "done", "branch": "claude/issue-13254-locale-switch-asks", "pr": "https://github.com/objectstack-ai/objectui/pull/13263", "session": "session_014MXhk1yjWh37d634hLy3p7 (mode:subagent: the parent's id, as the harness Claude-Session line names it)", "premise_still_valid": true, "summary": "This is the card's second half, as seat decision 6291833578 ruled it (A, ask first). LocaleSwitcher now routes each entry through confirmUnsavedLeave, as `pick = (code) => confirmUnsavedLeave(() => selectLanguage(code))`. So on a page holding unsaved edits, a language switch asks the page's own question first. Cancel keeps the language, the edits and sys_user.locale; OK switches; with nothing registered the switch happens at once. The PM's hypotheses, measured:\n- H1 (the promise is returned to the seam, like Log out's signOut()) held. A test pins it.\n- H2 is falsified. Picking the checked entry is not a no-op: it emits languageChanged, hands every reader a new t and rewrites sys_user.locale. It does not remount. So the checked entry keeps asking.\n- H3: Radix runs onClick before it closes the menu. The question therefore shows with both the avatar menu and the language menu open; after Cancel only the language menu is closed.\nDraft PR #13263 is 'Part of #13254'. The worktree was based on origin/main 9a3bcfc7f, where PR #13259 had already landed. The card's assignee was already huangyiirene and was not touched.", "tests": "All at bf8a87084, every vitest run through os-verify-lock.\n- New pin packages/app-shell/src/layout/__tests__/LocaleSwitcher.unsavedEditsAsk-13254.test.tsx: 'Tests 5 passed (5)'. It uses the real Radix menu, the real I18nProvider and the real useLanguageSelection, signed in, over a fake sys_user row. The page registers via useRegisterUnsavedEdits under a key={language} stand-in for the shell remount.\n- `pnpm exec vitest run --maxWorkers=2` over packages/app-shell/src/layout/__tests__/, chrome/unsavedEdits-13231.test.tsx, hooks/__tests__/userLocale-10059 and userLocaleOwnerChange-10193, sharedInboxFeed.twoSurfaces, HomeLayout.searchOpensPalette-12601, studio-locale.i18n and ResourceEditPage.leaveGuard-13231: 'Test Files 115 passed (115)', 'Tests 779 passed (779)', exit 0.\nAblation: committed first, ablation-replace.mjs WRAP mode, each restore reported blob == HEAD 5faf5ed554a5 and `git diff HEAD` empty. No dist leg, because the suite imports ../LocaleSwitcher from source.\n- (1) `onClick={() => pick(code)}` replaced with the base `onClick={() => void selectLanguage(code)}`, blob to b9ab19b373c1. Predicted 4 red and 1 green; got 'Tests 4 failed | 1 passed (5)', with the control green.\n- (2) `() => selectLanguage(code)` replaced with `() => void selectLanguage(code)` inside confirmUnsavedLeave, blob to 21e528933e19. Predicted only the tab-close case red; got 'Tests 1 failed | 4 passed (5)', 'expected true to be false'.\nH2 probe: a temporary unit test, deleted before the commit and never committed. Picking the checked English while en was active gave mounts 1 to 1 under key={language}, languageChanged ['en'], t identities 1 to 2, and update('sys_user','usr_1',{locale:'en'}). The control, 中文, gave mounts 1 to 2.\nChromium 141.0.7390.37: this branch's console vite dev server on its own port 5287, /api/** stubbed with the first half's census F3 stubs, an init-script wrapper on window.confirm, and a dirty /apps/crm/system/settings/localization with Date format DD/MM/YYYY.\n- This branch, avatar › Language › 中文 + Cancel: asked 'You have unsaved changes. Are you sure you want to leave?'. html lang en, DD/MM/YYYY kept, bar '1 unsaved change', 0 sys_user writes, input node kept. Menus open at the question: 2; after: the avatar menu only.\n- This branch, checked English + Cancel: same outcome, no write.\n- This branch, 中文 + OK: asked; zh, YYYY-MM-DD, bar gone, one PATCH /api/v1/data/sys_user/u_admin, input node replaced, 0 menus.\n- Base (fix line reverted via ablation-replace, restored with diff empty): no question at any leg. 中文 switched at once: zh, YYYY-MM-DD, bar gone, PATCH, input node replaced.\nThe dev server was started by me and stopped by its recorded PIDs; port 5287 is free.", "gates": "Local subset, hand-derived (dispatch-gates refuses objectui), at bf8a87084, each exit code captured before any pipe:\n- `turbo run build --filter='@object-ui/app-shell^...' --concurrency=2`: exit 0 (28 tasks), run first because type-check reads dist.\n- `pnpm --filter @object-ui/app-shell type-check`: exit 0. `tsc -p tsconfig.test.json --listFilesOnly` lists LocaleSwitcher.tsx and the new test.\n- `pnpm --filter @object-ui/app-shell lint`: exit 0, 0 errors. The new test has 0 warnings. LocaleSwitcher.tsx keeps its 1 base warning (react-refresh/only-export-components on localeLabel).\n- `pnpm check:new-line-citations`: exit 0, 'VERDICT new-cross-file-line-citations: 0 new citation(s)'.\n- `pnpm check:control-bytes`: exit 0.\n- `node scripts/check-changeset-presence.mjs`: exit 0 (.changeset/13254-locale-switch-asks.md, @object-ui/app-shell patch).\n- `node scripts/check-changeset-no-major.mjs`: exit 0.\nCI check-runs on bf8a87084, read once at report time (total_count 35 = 35 listed):\n- success: Build Docs, Changeset Bump Policy, Changeset Declaration, Changeset Fixed Group Check, Changeset Overwrite Report, Governed Surface Queue Guard, label.\n- skipped: Test (coverage), the coverage shard matrix, dependabot.\n- in_progress: Build & E2E, Bundle Analysis, Changeset Claim Re-read, Line Citation Gate, Lint, Live E2E (informational), Repo Gates, Test (dist pins), Test shards 1-16/16, Type Check.\nCI convergence belongs to the seat. origin/main moved to cbc312d78 (Changesets pre-mode exit) after the branch was cut; it touches none of this diff's files, so nothing was merged.", "mcp_calls": "0: no MCP tool was called.", "api_writes": "2 relay writes, each a POST /repos/objectstack-ai/objectstack/dispatches executed as objectstack-fleet[bot]:\n- (1) pr_create: POST /repos/objectstack-ai/objectui/pulls (draft) plus POST /repos/objectstack-ai/objectui/issues/13263/assignees. Read-back: 8337 bytes sent, 8337 stored, identical; assignee huangyiirene. Re-read over REST: draft true, body identical, 0 angle brackets.\n- (2) comment: POST /repos/objectstack-ai/objectui/issues/13254/comments, this report.\nNo labels were written; 'tests' and 'package: app-shell' on the PR come from the labeler. The git pushes are not counted.", "deviations": [ "The commit carries the model-free trailer pair AGENTS.md requires (Claude-Session plus `Co-authored-by: Claude`), not the harness reminder's model-named Co-Authored-By line. AGENTS.md takes precedence over that reminder.", "The H2 probe was a temporary test file inside the worktree (layout/__tests__, a zzprobe file). It was run once through the lock and deleted before the first commit; a copy stays in the scratchpad. Nothing of it was committed." ], "open_questions": [], "out_of_scope_findings": [ "carrier: none (承接者:无) · noted, not filed (H5, as the brief asked). useLanguageSelection has one caller, LocaleSwitcher. The console's login page also mounts LocaleSwitcher, but no page is registered there. The profile page's language card does not call useLanguageSelection. It writes sys_user.locale through the adapter, and useSignedInUserLocale (mounted once in AppContent) then applies the column through the i18n changeLanguage on the next sys_user invalidation. That reaches the same MetadataProvider key={language} remount without passing this menu. ProfilePage registers no edits with the seam, so there is nothing for it to ask on that path. Whether that remount drops a draft in another profile card is NOT MEASURED, so this is not class (a). The same note records doc drift: ProfilePage's docstring still says the globe menu does not write the column, which objectui#10059 changed. Both are in the PR's Acceptance notes.", "carrier: none (承接者:无) · noted, not filed: the promise returned to confirmUnsavedLeave keeps the tab-close prompt quiet until the switch and its sys_user.locale write settle. The write runs after the remount, so a page edited and closed inside that span is not asked. Log out has the same shape. NOT MEASURED; it is recorded in the PR's Acceptance notes.", "carrier: none (承接者:无) · noted, not filed: the seam's docstring (chrome/unsavedEdits.ts, outside this claim) lists the shell departures it guards and does not name the language menu." ] }Generated by Claude Code
objectstack-fleet commentedon Oct 21, 2026
LANDED ·
domain:uiseat 1 ·session_014MXhk1yjWh37d634hLy3p7· 2026-10-21T00:18Z ·pm:dispatchedcleared · closed by the seat's hand (both PRs readPart of #13254)9a3bcfc(landing note6292355711)588d11dhas one parent (cbc312d), andgit merge-base --is-ancestorplaces it onorigin/main(exit 0)gh-readonly-queue/main/pr-13263-*branch remainsorigin/main3aa2f52,LocaleSwitcher.tsxholdsconfirmUnsavedLeave(2) and the bareonClick={() => void selectLanguage(code)}0 times, and the pinLocaleSwitcher.unsavedEditsAsk-13254.test.tsxis present. Part 1 is still in place:SettingsView.tsxholdsloadNamespace(2)6290960573, amended by6292282031; dev reports6291774109and6292889952; seat decision6291833578(ruling A, "ask first"); ACCEPT6291821271on PR #13259 and6292948394on PR #13263 (no contract-review face hit)completedin this act, per the decision's closing line as amended: both PRs have landedWhat the card now delivers, by the amended Expected:
sys_user.locale; OK switches (part 2).Noted, not filed (none measured at a public entry):
sys_user.locale;