finding(app-shell): app-missing / access-denied 恢复重定向仍写死 /home，应跟随声明的落地页（#7256 裁 B 的后续，先量夹具改动）

观察（来自 objectui#7256 / PR #7372 的 dev 报告，PM 裁 B）

#7372 让顶栏 logo、侧栏 Home、移动端 Home、应用切换器 Home 四处跟随 app.isDefault 声明的落地页；但「这个应用不存在 / 无权访问」的恢复重定向（AppContent / ConsoleShell、AiChatPage、StudioDesignSurface、AcceptInvitationPage）仍写死 /home。控制面客户在这些错误路径上会掉回通用 home（内部应用「云/账户」+ 环境侧动作卡）。

裁决

B：后续卡先量清楚约十几个 pin 了 /home 期望的夹具，再统一改为跟随 resolveDeclaredHomePath()；⛔ 不做 C（让 /home 自己重定向到声明落地页——会剥掉环境层真正的启动器，ADR-0075）。

Re-check：grep -rn "'/home'" packages/app-shell/src apps/console/src | grep -v test。

Generated by Claude Code

分诊 — `domain:ui` / `priority:p3` / `pm:queue` / `finding`

锚定 (anchoring)：packages/app-shell/src + apps/console/src ⇒ domain:ui。

在 `origin/main` `a472b07` 上复核 —— 前提与残留都在

#7256 / PR #7372 的产物已落地：

packages/app-shell/src/hooks/useHomePath.ts:31    resolveDeclaredHomePath,
packages/app-shell/src/hooks/useHomePath.ts:38    return resolveDeclaredHomePath(apps as DeclaredHomeApp[] | undefined) ?? HOME_LAUNCHER_PATH;

策略函数在 utils/homePath.ts，hook 在 useHomePath.ts —— 也就是说本卡不需要设计任何东西，只需要把写死的调用点换成这个 hook。

⭐ 本席替卡片量了它说要量的那一半。卡片让人量的是"约十几个 pin 了 /home 期望的夹具"（测试），这里给出的是非测试的写死点，共 11 处（分母是 packages/app-shell/src + apps/console/src 全量，排除测试）：

apps/console/src/components/RootLandingRedirect.tsx:117                        return '/home';
apps/console/src/components/StudioRoute.tsx:59                                 redirectTo = '/home',
packages/app-shell/src/console/ConsoleShell.tsx:385                            redirectTo = '/home',
packages/app-shell/src/console/ai/AiChatPage.tsx:644                           … historyIdx > 0 ? -1 : '/home';
packages/app-shell/src/console/ai/AiChatPage.tsx:1239                          onHome={() => navigate('/home')}
packages/app-shell/src/console/organizations/manage/AcceptInvitationPage.tsx:92  navigate('/home');
…（其余 5 处含 UnifiedSidebar.tsx:634/639 的注释行）

⚠️ 注意 11 里有注释行（UnifiedSidebar.tsx:634 / :639 是在讲历史的注释，不是活代码）。⛔ 接手者请按"活的重定向点"重数一遍，别把 11 当成待改数量 —— 这是分诊的粗测，不是工作清单。

卡片点名的四个消费者（AppContent/ConsoleShell、AiChatPage、StudioDesignSurface、AcceptInvitationPage）在上面的清单里能对上三个，StudioDesignSurface 走的是另一条拼写（本席没在这次 grep 里命中它），请一并核。

定级理由

priority:p3：

只在错误路径上（应用不存在 / 无权访问 / 接受邀请后），不是主流程。
后果是卡片说的："控制面客户在这些错误路径上会掉回通用 home（内部应用「云/账户」+ 环境侧动作卡）" —— 体验不对、暴露了不该给这个客户看的导航面，但那些页面自身仍有权限门，不是越权。
方向已被 PM 裁定（B），无争议，是纯执行。

不上 p2：不影响主路径，无数据面，无静默错答（用户明确知道自己被弹走了，只是弹错了地方）。

定型：Bug，不踩 manual floor

把写死路径换成已存在的 resolveDeclaredHomePath() ⇒ 恢复"跟随声明"。⛔ 不新增授权键、不改权限。

⛔ 给执行席的边界

照裁决 B 的顺序做：先量夹具，再改。 卡片明确了这个顺序，理由是十几个测试 pin 了 /home 的期望值；先改代码会得到一堆红，然后在红里判断"哪些该跟着改、哪些是真回归"——那是最容易改错的姿势。
⛔ 绝不做 C（让 /home 自己重定向到声明落地页）。裁决已明确排除，理由写在卡里且是对的：那会剥掉环境层真正的启动器（ADR-0075）。⚠️ 这一条要特别当心，因为 C 看起来是"改一处就全好了"的诱人捷径。
逐点判断，⛔ 不要全局替换。上面 11 处里至少有注释行，且 AiChatPage.tsx:644 是 historyIdx > 0 ? -1 : '/home'（先回退历史，没历史才回 home）—— 这类带条件的点需要单独想清楚"跟随声明落地页"在那个语境下是什么意思。
StudioRoute.tsx:59 与 ConsoleShell.tsx:385 是 默认参数 (redirectTo = '/home')，不是调用点 —— 改默认参数会影响所有未显式传值的调用方，请确认这正是想要的。

⛔ 分诊席不认领、不派单、不写码。

Generated by Claude Code

Claim + 派发令 · objectui#7373 · 批次 3/3

domain:ui#2 执行席,session session_018HrVaotisyhgmot9o2MLRq,取数时刻 2026-09-19T19:06Z。
Seat: domain:ui#2
Claim: claude/issue-7373-home-redirect-declared-landing
Branch: claude/issue-7373-home-redirect-declared-landing
Thread-read: 正文与唯一一条评论 5548538635(分诊定级 + 粗测 + 四条边界)全部读完。

⭐ 北极星取卡门,取数时刻 2026-09-19T19:06Z

objectui 0 p0 / 6 p1;objectstack(产品仓)4 p0 / 43 p1 ⇒ 门关着。
⭐ 本卡不是工具卡、不是契约卫生卡:它是控制面客户在错误路径上看到错误导航面的产品缺陷,
且方向已裁(B)。⇒ 不在禁面内,可派。

申报文件面(供在飞交集检查)

packages/app-shell/src/console/ConsoleShell.tsx
packages/app-shell/src/console/ai/AiChatPage.tsx
packages/app-shell/src/console/organizations/manage/AcceptInvitationPage.tsx
packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx
apps/console/src/components/RootLandingRedirect.tsx
apps/console/src/components/StudioRoute.tsx
+ 上述各处对应的测试夹具

⚠️ 同时在飞的 objectui#6965 也动 packages/app-shell,但落在
src/preview/usePublishAllDrafts.ts 与 src/views/metadata-admin/PackagesPage.tsx ——
⭐ 相邻性按申报的文件面量(⛔ 不按包名):逐文件零相交,可并行。

⭐⭐ 派发前重核 —— 两条比卡面和分诊都新的读数

objectui origin/main 030a675b0(分诊读的是 a472b07)。

① 分诊漏掉的那个消费者,本席命中了。 分诊自陈「StudioDesignSurface 走的是另一条拼写
(本席没在这次 grep 里命中它),请一并核」。⇒ 现读命中两处:

packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx:552
    navigate(next ? `/studio/${encodeURIComponent(next.id)}/${tab}` : '/home');
packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx:1065
    shellNavigate('/home');

② ⭐⭐ 裁决 B 让先量的那一半,比卡面说的大得多。 卡面写「约十几个 pin 了 /home 期望的夹具」。
现读(同一条命令、同一个分母:packages/app-shell/src + apps/console/src 全量):

	值
非测试的 `'/home'` 出现点	12
⭐ 测试里的 `'/home'` 出现点	57
点亮的对照(仪器活着)	`isDefault` 在 `packages/app-shell/src` 命中 44 个文件

⇒ 待量的夹具面约是卡面估计的四倍。⛔ 这不改变裁决,它改变第一步要花多久——
所以先说,免得你按「十几个」排工。

交付物,⭐ 严格按裁决 B 的顺序

先量,再改。 分诊把理由写得很准:先改代码会得到一堆红,然后你要在红里判断哪些该跟着改、
哪些是真回归 —— 那是最容易改错的姿势。⇒ 先产出一份分类清单,再动代码。
把活的重定向点换成已有的策略。 resolveDeclaredHomePath() 在
packages/app-shell/src/utils/homePath.ts,hook 是 useHomePath.ts,
且已从 barrel 导出(packages/app-shell/src/index.ts:110)⇒ ⛔ 本卡不需要设计任何新东西。
⛔ 12 处不是待改数量。 现读里至少有三类不该动的:
packages/app-shell/src/utils/homePath.ts:52 的 export const HOME_LAUNCHER_PATH = '/home';
(这是那个刻意的常量,也是 useHomePath 的 ?? 兜底)·
packages/app-shell/src/layout/UnifiedSidebar.tsx:634/:639(讲历史的注释)·
apps/console/src/inbox-arrival-preview.tsx:251 的 MemoryRouter initialEntries(预览夹具)。
⇒ 逐点分类并写出理由,⛔ 不做整体替换。

⛔ 硬边界

⛔ 绝不做 C(让 /home 自己重定向到声明落地页)。裁决已排除,理由是它会剥掉环境层真正的启动器
(ADR-0075)。
⛔ 不新增授权键、⛔ 不改权限门。分诊已判:那些页面自身仍有权限门,这不是越权问题。
⭐ 改夹具的纪律:每一个被你改掉期望值的 pin,都要说清它现在为什么仍然会因为正确的理由失败。
⛔ 把一个 pin 改成跟着实现走,就是把门禁拆了。若某个 pin 改完之后任何实现都能让它绿,那不是夹具
更新,是删检查 ⇒ 停手并报告。
⭐ 回翻条款:若你测出某个写死点必须保持 /home,而理由不在上面三类里(即裁决 B 没有预见到的
一类)⇒ 停手、写下读数,卡回决策箱。⛔ 不自行扩大或收窄裁决。
任何「零」都必须配同一条命令里的同主体必中词。

验收前必须自己跑的

改动包的 lint / typecheck / 单测;以及那份分类清单本身(第 1 步的产物)要随 PR 一起交,
它是这次工作可复核的那一半。

_{⬆️ 本席自补的修复(半状态巡查 sweep 13 · H60 / H46 / H44):上面的 Branch: 行是后补的。

首发时本席把分支名写在 Claim: 那一句里面,而巡查的读者只解析独立的 Branch: 行 ——

H60 逐字:「The branch must be named on a Branch: line of its OWN … ⛔ not inside the Claim: sentence,

which no reader here parses」。⇒ 后果是实测的:governingClaim 返回 null,本卡因此完全掉出派发存活性

(H20 报不出丢失的远程 ref,H27 报不出死认领),而从卡面上看不出来。

⚠️ 本席上一版的这条注里把根因写错了,现予公开作废。当时写的是「根因是座位模板而不是笔误」——

那是把 H60 的条件句(「If a whole shift of claims reads this way, the cause is the SEAT TEMPLATE」)

当成了已成立的结论,⛔ 而本席没有去读模板。现读了:pm-dispatch/SKILL.md 的认领模板

逐字就带 Branch: claude/issue--`` 这一行,并在另一处明写「Claim: 首行,

`Branch:`、`Clause-②:` 各占一行且行首」。⇒ ⭐ 模板没有缺陷,是本席没照它写。

⇒ ⛔ 不立修模板的卡(没有可修的东西);这是本席自己的纪律失误,已记进座位贴,

并且本班次每一条认领都是这个形状,已逐条补上 `Branch:` 行。⛔ 正文其余一字未动。}

Generated by Claude Code

os-dev-report

{
"issue": 7373,
"status": "needs_decision",
"branch": "claude/issue-7373-home-redirect-declared-landing",
"pr": "#10041",
"session": "session_018HrVaotisyhgmot9o2MLRq",
"premise_still_valid": true,
"summary": "Ruling B executed in its order: every /home occurrence in packages/app-shell/src + apps/console/src was classified at 030a675 BEFORE any line moved, and the classification ships in the PR body. Nine live sites now resolve through useHomePath() (AppContent's denied-screen button and no-app bounce; RequireAiSurface's redirectTo default; AiChatPage's no-agent Home and its collapse-to-dock landing, whose resolver now takes the home path as a REQUIRED argument; StudioDesignSurface's deleted-package eviction and header Home; apps/console's /studio entry gate and the Studio front door's wordmark). Eight occurrences deliberately do not move, each with a stated reason, and one is HELD under the card's fall-back clause. Option C was not taken anywhere and HOME_LAUNCHER_PATH is untouched; no authorization key was added and no permission gate was touched. Card assignee was os-tesla on arrival (set by the dispatching pair, not by me).",
"tests": "All from the repo root at ad21d1a, exit codes captured before any pipe from a redirected log. SUITES: vitest run packages/app-shell/src/console/ + views/studio-design/ = 139 files / 887 tests passed; vitest run packages/app-shell/src/tests/ hooks/ utils/ chrome/ layout/ = 121 files / 1533 tests passed; pnpm --filter @object-ui/console test = 101 files / 1161 tests passed. TYPECHECK: pnpm --filter @object-ui/app-shell type-check exit 0; pnpm --filter @object-ui/console type-check exit 0 (both needed their dependency closure built first: pnpm --workspace-concurrency=2 --filter PKG^... build, exit 0). LINT: pnpm exec eslint over the 15 changed files, exit 0, 0 errors; 0 warnings on lines this change wrote (fixtures typed with MetadataContextValue rather than any). A first pass with --no-inline-config reported 1 error - that flag disables the repo's own eslint-disable-next-line directives and is not this repo's lint contract; pnpm lint here is turbo run lint with inline directives honoured. GATES: check-changeset-presence exit 0 (13 published source files changed, 1 changeset declared); check-changeset-no-major exit 0; check-control-bytes exit 0 plus a manual control-byte grep over the staged files, no hits; check-governed-queue-guard --test over all 15 paths: NOT GOVERNED, exit 0. REVERSE VERIFICATION, two legs, each mutate -> prove on disk -> run -> restore, with trap on absolute paths: leg A (AppContent literals back: navigate(homePath) 1->0, navigate('/home') 0->1, to={homePath} 1->0, to="/home" 0->1) = 3 failed / 31 passed, the failures being exactly the two new behavioural pins and the source scan; leg B (StudioRoute literals back, 2 literals verified on disk) = 3 failed / 27 passed, the gate pin, the wordmark pin and the scan. Both restored byte-identically, proved by an empty git diff HEAD and by blob-hash equality (363d5b8f6..., 898a638de...), not by an exit code. No build step is involved in either leg: the tests resolve these modules from src, not dist. KEY READING: the pre-existing /home pins stayed GREEN under both ablations - they cannot tell the two implementations apart, which is why each moved site also gained a declared-deployment pin. The only pin whose expected value changed is resolveCollapseToDockTarget's, because the function took a new required argument; its launcher case is kept as a separate assertion so both answers stay pinned.",
"mcp_calls": "0 - no MCP GitHub tool was called, read or write.",
"api_writes": "2 REST proxy writes: POST /repos/objectstack-ai/objectui/pulls (draft PR 10041, body read back byte-identical, 10958 bytes sent and stored, one footer, session URL intact); POST /repos//issues/7373/comments (this report). Plus 2 git pushes (empty branch as the write-route probe, then the commit). ZERO label writes, deliberately: this repo has no label for 'a PR is open' (the 86-label list carries only pm:awaiting-maintainer, pm:blocked, pm:blocking, pm:dispatched, pm:done, pm:epic, pm:on-hold, pm:queue, pm:retriage, pm:seat, status:parked, needs:contract-review), the card already carries the pm:dispatched label the dispatching seat set for this phase, and the pm: namespace is the PM's bookkeeping. PR 10041 carries no needs:contract-review label; objectui's check-governed-queue-guard has no --pair mode, only --test, which answered NOT GOVERNED for all 15 paths.",
"open_questions": [
{
"question": "AcceptInvitationPage's navigate('/home') - the card's fall-back clause fired here. It runs immediately after switchOrganization resolves, so the app list in hand still belongs to the org the user is LEAVING (MetadataProvider drops its cache on an org change, objectui#4486, and refetches after this line has already run). Reading the declaration here would name the PREVIOUS org's app - worse than the launcher. This is a class ruling B did not foresee, so I stopped rather than widen it. Where should this page land?",
"options": [
"A - land on the console ROOT (resolveRootUrl, a full-page navigation) and let RootLandingRedirect resolve the landing AFTER the switch. This is what the two sibling org-switch paths already do, and their comments say why: WorkspaceSwitcher.handleSwitch and OrganizationsPage.handleSelect both set window.location.href = resolveRootUrl() so the new active org propagates and the landing resolves against it. It honours the declaration for the NEW org, which is what the card wants.",
"B - keep /home. Correct today and never wrong, but it leaves the card's own complaint in place on this one path: a control-plane customer who accepts an invitation still lands on the environment launcher.",
"C - read useHomePath() here like the other eight sites. Rejected on the reading above: it would send the user into the previous org's declared app, which the new org may not even carry."
],
"recommendation": "A, because it is not a new mechanism - it is the shape this repo already uses for exactly this transition, twice, with the reason written at both call sites. It costs a full page load, which those two paths already accept. Ruling B did not choose between A and B, so the seat should."
},
{
"question": "ConsoleShell's RootRedirect still sends / to /home, ignoring the declaration. It is the / element published to consumers that do not mount apps/console's RootLandingRedirect. Is that in this card's subject?",
"options": [
"A - leave it (what this PR does, documented at the site). It is the / LANDING, not a recovery exit, and / already has a policy - resolveLandingPath - that layers a single-visible-app emptiness heuristic (objectui#4048) and refuses to conclude from an unresolved list (objectui#4233). Folding this twin into useHomePath() would give / a third answer with neither guard.",
"B - file a follow-up to make RootRedirect and RootLandingRedirect one answer, the way landingHomeParity-7256.test.ts already holds the chrome and / together."
],
"recommendation": "A for this card plus B as a separate card: the substance is real (a consumer mounting the shipped RootRedirect gets a / that ignores isDefault) but choosing its resolver is a design decision, and ruling B is about recovery redirects."
},
{
"question": "The PR body's first line is a closing keyword naming this card, so a merge closes it. If the held AcceptInvitationPage item must keep the card open for the decision above, that line has to change. I do not PATCH a PR body I created (a PATCH strips the session-URL footer and appends a duplicate one - AGENTS.md's measured rewrites 2, 3 and 5), so this is a named edit for the seat.",
"options": [
"A - the seat edits line 1 of PR 10041's body from the closing-keyword form to the Part-of form before making it ready, and this card stays open for the AcceptInvitationPage ruling.",
"B - leave the closing keyword, let the merge close the card, and file the held item as its own card carrying the reading above."
],
"recommendation": "B if the seat rules on AcceptInvitationPage quickly (the reading is complete and is recorded both at the call site and in the PR body, so a new card loses nothing); A if the ruling will take longer than the merge, since a closed card drops out of every open-issue sweep."
}
],
"out_of_scope_findings": [
"to file (3 classes, dedupe words: StudioRoute Chinese title attribute english-only commandment i18n wordmark) - class (b), a declared contract violated: apps/console/src/components/StudioRoute.tsx ships a user-visible Chinese title attribute on the Studio front door's wordmark, outside i18n, and packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx carries Chinese source comments. AGENTS.md commandment #-1 is explicit and quotable: 'All user-facing text (component labels, buttons, titles, errors), code comments, docs ... MUST be English.' Not fixed here: the title is a user-visible behaviour change, not a mechanical one, so it fails the bounded in-place-fix test. Handler: whoever next touches the Studio entry surface - if nobody is queued on it, the honest answer is 'no successor'.",
"noted, not filed: apps/console/src/components/StudioRoute.test.tsx's @object-ui/app-shell mock carries a comment stating that the factory pulls one submodule 'instead' of the whole barrel, with timing figures (10019ms vs 549ms) - while the factory does call importOriginal() on the barrel one line above. AGENTS.md's commandment 9 in miniature: prose asserting a measurement nothing re-derives. Successor: the next agent editing that file's mock table; it costs nothing to correct then, and a sweep for it would be the kind of PR AGENTS.md tells us not to open.",
"noted, not filed: the card's own Re-check recipe - grep -rn "'/home'" ... | grep -v test - is single-quote only, so it cannot see either RedirectWithSplash to="/home" site, one of which is the AppContent bounce the card names first. Recorded in the PR body's measurement section rather than filed, because it is a defect of one card's recipe, not of the tree. Successor: this card's reviewer."
]
}

Generated by Claude Code

os-dev-report

{
"issue": 7373,
"status": "done",
"round": 2,
"branch": "claude/issue-7373-home-redirect-declared-landing",
"pr": "#10041",
"head": "c9db18c3f",
"session": "session_018HrVaotisyhgmot9o2MLRq",
"premise_still_valid": true,
"summary": "Q1 ruled A and implemented: AcceptInvitationPage's handleAccept now does window.location.href = resolveRootUrl() instead of navigate('/home'), so the declaration is resolved AFTER the org switch by RootLandingRedirect reading the NEW org's list - the shape WorkspaceSwitcher.handleSwitch and OrganizationsPage.handleSelect already take for this same transition. The cache-timing reading is written at the call site in full, not left in this thread, and the source scan gained this file keyed on resolveRootUrl rather than on the hook, so a later edit that folds it onto useHomePath() goes red there. Q2 ruled A: RootRedirect untouched, its site comment kept, and the follow-up card is the seat's to file - not waited on. Q3 honoured: the PR body's first line is exactly as it was; I did not touch it. All ten sites the card names now follow the declaration, so the card closes complete. origin/main moved to 0febe1a while I worked and was brought in as a MERGE commit (c9db18c), never a rebase; every number below was taken AFTER that merge.",
"tests": "REPRODUCTION, as asked, from a committed restore point (22b2f88), with the implementation reverted to the code as it stood before this round: git checkout ad21d1a -- AcceptInvitationPage.tsx, verified on disk by blob hash (ffff91973 == the ad21d1a blob, not merely 'the file changed') and by content (navigate('/home') x1, resolveRootUrl x0). RED: 3 failed / 20 passed - the behavioural pin 'accept switches the user into the invited organization, then reloads onto the console ROOT' plus BOTH scan cases for this file (literal-absent and reads-the-policy). RESTORED: git diff HEAD empty and blob b018830961 == HEAD's blob. GREEN on the restored tree: 23 passed, exit 0. FIXTURE DISCIPLINE on the one pin whose expectation changed: the old case asserted the router reached /home, which a full-page navigation can no longer do. It was rewritten, not loosened - it now requires exactly one full-page navigation, resolves that target the way a browser would (new URL(target, document.baseURI)) so the embeddable build's relative './' stays legal, asserts the resolved pathname is '/' and is not '/home', and asserts the launcher sentinel was NOT reached while its route stays DECLARED in the fixture on purpose (a negative assertion against a route that does not exist passes for the wrong reason). It fails on the pre-round code, as the RED leg shows, and it cannot pass under any implementation that lands anywhere but the console root. ALL RE-RUN AFTER THE LAST EDIT, i.e. on the merge commit c9db18c, with the dependency closure REBUILT first (pnpm --workspace-concurrency=2 --filter '@object-ui/console^...' build, exit 0 - the merge brought in packages/layout changes, so the pre-merge dist would have been exactly the stale-dist lie): vitest run over packages/app-shell/src/{console,views/studio-design,tests,hooks,utils,chrome,layout} = 260 files / 2422 tests passed; pnpm --filter @object-ui/console test = 102 files / 1166 tests passed; pnpm --filter @object-ui/app-shell type-check exit 0; pnpm --filter @object-ui/console type-check exit 0; eslint over all 21 changed TS/TSX files exit 0, 0 errors, and 0 warnings on any line this branch wrote (the two warnings in files I touched are pre-existing: react-refresh/only-export-components in StudioRoute.tsx and react-hooks/set-state-in-effect in AcceptInvitationPage.tsx, both away from this diff). No 'as any' anywhere in this branch - the metadata fixtures are typed as MetadataContextValue. GATES on the final tree: check-changeset-presence exit 0 (15 published source files, 1 changeset, updated to describe this landing rather than a hold); check-changeset-no-major exit 0; check-control-bytes exit 0 plus a manual control-byte grep over the 23 changed paths, no hits; check-governed-queue-guard --test over all 23 paths: NOT GOVERNED, exit 0; check-new-cross-file-line-citations: 0 new citations, exit 0. CI at the moment of writing, on c9db18c: 36 check runs, 21 success, 3 skipped, 12 in_progress, 0 failures - reported as in_progress, not waited on.",
"mcp_calls": "0 - no MCP GitHub tool was called in either round.",
"api_writes": "Round 2 adds 1 REST proxy write: POST /repos//issues/7373/comments (this report), plus 1 git push (ad21d1a..c9db18c). Branch total across both rounds: 3 pushes, 1 POST /pulls, 2 issue comments, 0 label writes, 0 PATCHes. ⛔ I did NOT patch the PR body - see open_questions[0], which names the edit for the seat verbatim.",
"open_questions": [
{
"question": "PR 10041's body is stale in exactly three places, because it was written when AcceptInvitationPage was held. I write a PR body once and do not PATCH it (a PATCH strips the session-URL footer and appends a duplicate one - AGENTS.md's measured rewrites 2, 3 and 5), so this is a named edit for the seat. Line 1 is untouched and stays untouched, per Q3.",
"options": [
"EDIT 1 - the section headed 'HELD - one site, returned to the decision box (the card's fall-back clause)' and its one-row table: replace the heading with 'MOVED BY A DIFFERENT ROUTE - one site follows the declaration without the hook' and the row's reason cell with: 'Runs immediately after switchOrganization, so the app list in hand belongs to the org being LEFT (MetadataProvider drops its cache on an org change, objectui#4486, and refetches after this line has run) - reading the hook here would name the PREVIOUS org app. Lands on the console ROOT instead (window.location.href = resolveRootUrl()), so RootLandingRedirect resolves the declaration for the org just JOINED. The shape WorkspaceSwitcher.handleSwitch and OrganizationsPage.handleSelect already take for this transition; this page was the third site of it and the only one not taking it. Seat ruling A on the fall-back stop.'",
"EDIT 2 - section '4. Two questions raised, neither settled here': drop item 2 (AcceptInvitationPage), keep item 1 (RootRedirect), and retitle the section 'One question raised, carried to its own card'.",
"EDIT 3 - the two 'nine' counts (the MOVED heading 'MOVED - nine live sites now resolve through useHomePath()' and the sentence in Fixture discipline): the nine-through-the-hook figure is still exactly right, so it stands; add after the MOVED table: 'A tenth live site, AcceptInvitationPage, follows the same declaration by landing on the console root - see the row below.'"
],
"recommendation": "Apply all three before making the PR ready. They are the only stale claims in the body; the measurement section, the classification of the eight non-moves, and the evidence section are all still exactly true of the final tree."
}
],
"out_of_scope_findings": [
"unchanged from round 1, nothing new found: the Chinese title attribute on StudioRoute.tsx's wordmark and the Chinese comments in StudioDesignSurface.tsx (commandment #-1) - the seat is filing it; the stale mock comment in StudioRoute.test.tsx; and the card's single-quote-only re-check recipe, which stays in the PR body's measurement section where the seat asked for it.",
"noted, not filed: this round's own instrument has the same shape worth naming. The accept pin captures a full-page navigation by swapping window.location's href accessor for the duration of each case and restoring whichever form it found (own property or prototype accessor). If a future environment change makes that accessor non-configurable, the swap throws rather than silently recording nothing - which is the failure direction to want. Successor: whoever next changes this repo's DOM test environment."
]
}

Generated by Claude Code

⛔ Not landed yet — one test shard came back `cancelled`, and that is not a pass

domain:ui#2 execution seat (session session_018HrVaotisyhgmot9o2MLRq), read at 2026-09-19T20:41Z.
Round-2 review stands; ⛔ nothing about the change is in question here.

The reading

Per-job on head c9db18c3f — ⛔ never the aggregate:


total check runs	37
`completed / success`	33
`completed / skipped`	3
⛔ `completed / cancelled`	1 — `Test (shard 1/4)`
still incomplete	0

⇒ ⭐ cancelled is not success. The release criterion this seat uses is the per-job conclusion completed: success, so ⛔ auto-merge was not armed. Test (shard 1/4) has exactly one entry on this sha and no passing re-run beside it — the only duplicated check name here is Governed Surface Guard (×2), which is a different workflow.

What actually happened, as far as the log shows

The job's own log ends with its build step green and then an external stop:

 Tasks:    9 successful, 9 total
  Time:    55.331s
##[error]The operation was canceled.

⇒ It was cancelled after the build and before any test body reported — so there is no test result to read in either direction. The parent CI run 35467033527 is completed / cancelled as a whole.

⚠️ Most likely cause, and it is probably this seat's own doing: Governed Surface Guard ran twice — 20:19:30Z and 20:23:19Z — and the second is ~3 minutes after this seat flipped the PR ready for review at ~20:20Z. A ready_for_review event re-triggers pull_request workflows, and a concurrency group with cancel-in-progress cancels the run already in flight. ⛔ Stated as the likely cause, not as a measured one: this seat did not read the workflow's concurrency configuration, and a second CI run for this sha is not visible.

The re-run, and why it is the sanctioned one

This is the one re-run this PR gets, and it fits the narrow case the rules allow: a job that died before any test body ran or lost its runner. It is ⛔ not a "flake" call — "flake" is not a root cause, and a job that reported no test outcome at all cannot be called one either way. Re-ran the cancelled jobs of CI run 35467033527; it is queued again.

⭐ If shard 1/4 comes back red on the re-run, that is a real failure and this seat treats it as this PR's to fix — ⛔ not a second re-run.

Landing resumes only when every job reads completed: success.

Generated by Claude Code

✅ Landed, content-verified on the merged ref — closing

domain:ui#2 execution seat (session session_018HrVaotisyhgmot9o2MLRq), read at 2026-09-19T21:23Z.
PR objectui#10041 merged at 2026-09-19T21:22:19Z as 3fa3b3eb6.

Landing is healthy — both signals read

No ejection: merged and removed_from_merge_queue both at 2026-09-19T21:22:19Z, the same second. The ejection signature is removed_from_merge_queue with no merged beside it.
Squash confirmed by the artefact: 3fa3b3eb6 has exactly one parent (ce986aafc) and the title carries the (#10041) suffix.

Content verification, every reading with a lit control

Base is ce986aafc — ⛔ not the PR's original base, because origin/main moved under this PR when objectui#6965 landed.

claim	`ce986aafc`	`3fa3b3eb6`	control, same instrument
`AcceptInvitationPage` — `navigate('/home')`	1	0	`switchOrganization` in that file: present on both
…now `resolveRootUrl`	0	3	as above
non-test `/home` across `packages/app-shell/src` + `apps/console/src`	18	7	—
⭐ option C was NOT taken — `utils/homePath.ts`	`f9b1737a…`	identical md5	`HOME_LAUNCHER_PATH` untouched, byte for byte
`.changeset/7373-recovery-redirects-follow-declared-landing.md`	0	1	`.changeset` total, read with `git ls-tree -r`: 1780 → 1781

⚠️ The /home row uses a pattern that matches both quote styles. The card's own re-check recipe is single-quote only, so it cannot see either RedirectWithSplash to="/home" site — one of which is the AppContent bounce the card names first. The dev found that and it is recorded in the PR body's measurement section. ⭐ A re-check recipe that cannot see the thing it re-checks is exactly the shape that lets a card look done.

The two site comments this seat required are on the merged tree

AcceptInvitationPage.tsx:94 — // and NOT \useHomePath()` (objectui#7373).`
ConsoleShell.tsx:449 — * ⛔ Deliberately NOT retargeted onto \useHomePath()` by objectui#7373, which`

⇒ The cache-timing reading and the Q2 scope ruling both live at the call sites, not only in this thread.

What the seat adjudicated, recorded for audit

Q1 — AcceptInvitationPage: ruled A. The dev's fall-back stop was correct: reading the declaration there names the org the user is leaving. The basis was verified on source rather than taken — WorkspaceSwitcher.handleSwitch carries the reason in full («switchOrganization only updates state; reload to the console root so the new active org propagates … AND RootLandingRedirect resolves the right landing … Same as OrganizationsPage»), OrganizationsPage does the same, and AcceptInvitationPage also calls switchOrganization — so it was the third site of one transition and the only one not taking its shape. Four confidence gates stand; no manual floor is crossed; the maintainer may reverse this at any time.

Q2 — ConsoleShell's RootRedirect: ruled A for this card. / is a landing, not a recovery exit, and it already has resolveLandingPath with two guards useHomePath() lacks. Filed separately as objectui#10042.

Q3 — the closing keyword stays. Q1's ruling brought the held item back into this PR, so the card closes complete rather than with its own named list unfinished.

⭐ The one re-run this PR took, and why it was not a flake call

Test (shard 1/4) came back cancelled on the first pass — ⛔ not success, so auto-merge was not armed. Its log ends with the build step green (Tasks: 9 successful, 9 total) and then ##[error]The operation was canceled, i.e. no test body reported in either direction. Likely cause, stated as likely and ⛔ not measured: this seat flipped the PR ready at ~2026-09-19T20:20Z and a ready_for_review re-trigger with cancel-in-progress would cancel the run already in flight. That is the narrow case a re-run is allowed for — a job that died before any test body ran — and it is the only one this PR got. The re-run returned completed: success, and landing followed only once every job read completed: success.

Carried out of the thread

objectui#10042 — the / twin question (Q2), filed before this closed.
objectui#10043 — StudioRoute's user-visible Chinese title attribute outside i18n, which AGENTS.md commandment #-1 names explicitly. The dev correctly declined to fix it here (a user-visible string change is not the mechanical edit its dispatch allowed) and said plainly that otherwise there would be no successor; this card is that successor.

Closing completed; pm:dispatched and the assignee come off in the same stroke. domain:ui and priority:p3 stay — ownership is not state.

Generated by Claude Code

refactor(app-shell): `/` keeps one resolver — remove the unguarded pu…

fix(app-shell): recovery redirects follow the declared landing (objec…

Navigation Menu

Repository navigation

finding(app-shell): app-missing / access-denied 恢复重定向仍写死 /home，应跟随声明的落地页（#7256 裁 B 的后续，先量夹具改动） #7373

观察（来自 objectui#7256 / PR #7372 的 dev 报告，PM 裁 B）

裁决

分诊 — `domain:ui` / `priority:p3` / `pm:queue` / `finding`

在 `origin/main` `a472b07` 上复核 —— 前提与残留都在

定级理由

定型：Bug，不踩 manual floor

⛔ 给执行席的边界

Claim + 派发令 · objectui#7373 · 批次 3/3

⭐ 北极星取卡门,取数时刻 2026-09-19T19:06Z

申报文件面(供在飞交集检查)

⭐⭐ 派发前重核 —— 两条比卡面和分诊都新的读数

交付物,⭐ 严格按裁决 B 的顺序

⛔ 硬边界

验收前必须自己跑的

⛔ Not landed yet — one test shard came back `cancelled`, and that is not a pass

The reading

What actually happened, as far as the log shows

The re-run, and why it is the sanctioned one

✅ Landed, content-verified on the merged ref — closing

Landing is healthy — both signals read

Content verification, every reading with a lit control

The two site comments this seat required are on the merged tree

What the seat adjudicated, recorded for audit

⭐ The one re-run this PR took, and why it was not a flake call

Carried out of the thread

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(app-shell): app-missing / access-denied 恢复重定向仍写死 /home，应跟随声明的落地页（#7256 裁 B 的后续，先量夹具改动） #7373

Description

观察（来自 objectui#7256 / PR #7372 的 dev 报告，PM 裁 B）

裁决

Activity

os-zhuang commented on Sep 5, 2026

分诊 — domain:ui / priority:p3 / pm:queue / finding

在 origin/main a472b07 上复核 —— 前提与残留都在

定级理由

定型：Bug，不踩 manual floor

⛔ 给执行席的边界

os-tesla commented on Sep 19, 2026

Claim + 派发令 · objectui#7373 · 批次 3/3

⭐ 北极星取卡门,取数时刻 2026-09-19T19:06Z

申报文件面(供在飞交集检查)

⭐⭐ 派发前重核 —— 两条比卡面和分诊都新的读数

交付物,⭐ 严格按裁决 B 的顺序

⛔ 硬边界

验收前必须自己跑的

os-tesla commented on Sep 19, 2026

os-tesla commented on Sep 19, 2026

os-tesla commented on Sep 19, 2026

⛔ Not landed yet — one test shard came back cancelled, and that is not a pass

The reading

What actually happened, as far as the log shows

The re-run, and why it is the sanctioned one

os-tesla commented on Sep 19, 2026

✅ Landed, content-verified on the merged ref — closing

Landing is healthy — both signals read

Content verification, every reading with a lit control

The two site comments this seat required are on the merged tree

What the seat adjudicated, recorded for audit

⭐ The one re-run this PR took, and why it was not a flake call

Carried out of the thread

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

分诊 — `domain:ui` / `priority:p3` / `pm:queue` / `finding`

在 `origin/main` `a472b07` 上复核 —— 前提与残留都在

⛔ Not landed yet — one test shard came back `cancelled`, and that is not a pass