Repository navigation
finding(components): element:text / button / image / number apply aria.ariaLabel as aria-arialabel, so the declared accessible name never reaches the DOM #11051
Description
Activity
Path: changing a running app without code — an accessible name reaches the DOM | 缺项 (
ariaAttrswritesaria-arialabel) | P2Triage: first grade —
bug·priority:p2·domain:ui·area:studio·pm:queue. Direction: map the spec'sAriaPropskeys through the shared aria readerTriage: lands in
packages/components/src/renderers/basic/elements.tsx(the file-localariaAttrs, spread at six call sites byelement:text/button/image/number) ⇒domain:ui.Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) ·
session_01AavokzJ5DndAwitDXvKy4U· 2026-09-29T06:53Z. ⛔ Not a claim, ⛔ not a dispatch.Why p2. A declared accessible-name channel is silently dropped. An author who sets
aria.ariaLabelgetsaria-arialabel, which no assistive technology reads, and no error. That is declared ≠ enforced (ADR-0049) on four core elements, for plain strings too. It is runs-but-wrong.Direction.
- Replace the file-local prefixing with the shared aria reader the other renderers use (
resolveAriaProps, named in PR objectui#10855). It mapsariaLabel→aria-label,ariaDescribedBy→aria-describedbyandrole→role, and resolves anI18nLabelagainst the UI language. ⛔ No second mapping table inelements.tsx. - The
[object Object]half for a per-locale map is objectui#10993's census row. It is fixed by the same reader, so note it there when this lands. - Pins: a plain string and a locale map on one element render
aria-label, with a lit control on a renderer that already applies the name.
- Replace the file-local prefixing with the shared aria reader the other renderers use (
Claim: PM loop round 19 · an opus-only card while the review tier is unavailable
Session:session_01DuWo5bdP9SdVebamn99GGk
Account:os-steve(the seat's linked user asGET /useranswers it; the card's assignee)
Branch:claude/issue-11051-element-aria-reader
Worktree:objectui-issue-11051
Domain:domain:ui
Seat:domain:ui#1
File surface (triage grade5885230119):packages/components/src/renderers/basic/elements.tsx: the file-localariaAttrsprefixing is replaced by the shared aria reader the other renderers use (resolveAriaProps, named in PR objectui#10855). It mapsariaLabel→aria-label,ariaDescribedBy→aria-describedbyandrole→role, and resolves anI18nLabelagainst the UI language. This covers all sixariaAttrs(call sites (element:text,button,image,number).- Pins: a plain string and a locale map on one element render
aria-label, with a lit control on a renderer that already uses the shared reader. - A changeset.
⛔ No second mapping table in
elements.tsx, and no change to the spec'sAriaPropsor to any schema.
Stop on breach; explain in the report.
Container & model:M,mode:subagent,model: opus(TIER_DEFAULT, objectstackscripts/pm/dispatch-gates.mjs:12692at81137630)
Clause-②: no
Why no: a renderer now honours a key its published schema already declares and accepts (ADR-0049 declared ⇒ enforced); no accept set, type or documented contract moves.
Thread-read: 5885230119
Dispatch note: under the maintainer's 「当前没有fable，你可以派 opus 卡」 (this seat's chat, 2026-09-29T01:00Z), this card needs no review atCONTRACT_REVIEW_TIER. The seat verifies the changeset prose itself at the ACCEPT.
Serial constraints cleared: read 2026-09-29T07:13Z at objectuiorigin/main. No open PR toucheselements.tsx. PR objectui#11050 (seat 2, objectui#10993 batch 1) isobject-form, not these elements.domain:uiseat #1 · claim · 2026-09-29T07:13Z
Generated by Claude Code
os-dev-report
{
"issue": 11051,
"status": "needs_decision",
"branch": "claude/issue-11051-element-aria-reader",
"pr": null,
"session": "session_01DuWo5bdP9SdVebamn99GGk — mode:subagent, so this is the dispatching seat's id",
"premise_still_valid": true,
"summary": "The defect is real. On objectui origin/main 37a19d4, through the real SchemaRenderer, under en and zh: properties.aria.ariaLabel 'Plain name' renders aria-arialabel="Plain name" and no aria-label on element:text, element:button, element:image and element:number. A locale map renders aria-arialabel="[object Object]". ariaDescribedBy renders aria-ariadescribedby, and role passes through. I wrote no code, because the Zone 1 route does not hold up and every route that works breaks the claim. The claim allows 'no second mapping table in elements.tsx' and a file surface of elements.tsx, the pins and a changeset, and it says 'Stop on breach'. The route fails on three points. (1) resolveAriaProps is module-private in packages/react/src/SchemaRenderer.tsx and is not exported. No renderer calls it: SchemaRenderer calls it once per node and injects the result as React props, and the element renderers take only schema, so they ignore those props. (2) It reads the FLAT node keys (BaseSchema.ariaLabel, ariaDescribedBy, role). It does not read the nested aria bag that the spec declares on these four elements. The installed @objectstack/spec 17.4.0 declares aria on ElementText, ElementButton, ElementImage and ElementNumber PropsSchema, and it is AriaPropsSchema. A flat ariaLabel on element:text renders nothing today. (3) It resolves ariaLabel with resolveKeyedI18nLabel, which reads the KEYED vocabulary (key, defaultValue). For an inline locale map it returns undefined (measured). In the same run, the lit control container with flat ariaLabel 'Flat' rendered aria-label="Flat", but with a flat locale map it rendered no aria-label in either en or zh. So the ruled route would turn [object Object] into no name at all, and the ruling's own locale-map pin would fail. The repo splits the two vocabularies on purpose: see the BaseSchema.ariaLabel docblock in packages/types/src/base.ts, the PageSchema.aria docblock in packages/types/src/layout.ts, ListView's ariaLabel comment ('Neither resolver accepts the other's shape') and objectui#4580 Q2-B. Two readers do map the spec's nested AriaProps and resolve a locale map against the UI language: ListView in plugin-list (inline) and useRecordAriaProps in plugin-detail (recordComponentAria.ts). Both use the spec's resolveI18nLabel against useDisplayLocale(). Neither can be imported from @object-ui/components, which sits below both plugins. So the choice between them is open question 1. Minor correction: the file has five ariaAttrs call sites plus the definition, not six call sites. The branch was pushed empty at 37a19d4 as the claim's marker, and the worktree has been removed.",
"tests": "No diff, so no gates were run: NOT MEASURED for every gate family, because there is nothing to gate. Zone 2 readings: (1) Reproduced with a throwaway probe through the real SchemaRenderer, the element:* registry and I18nProvider, 23 cases,pnpm exec vitest run packages/components/src/__probe11051__/probe.test.tsx: exit 0, 'Tests 23 passed (23)'. The probe wrote the attributes to a file; it is kept in scratch and was deleted from the tree. Readings, identical in en and zh: text/button/image/number with properties.aria.ariaLabel 'Plain name' give aria-arialabel="Plain name" and no aria-label anywhere in the tree. text with a locale map gives aria-arialabel="[object Object]". text with ariaDescribedBy 'd1' and role 'status' gives aria-ariadescribedby="d1" role="status". text with the flat ariaLabel gives nothing. text with a top-level aria bag gives nothing (readProps reads only props and properties). Lit control: container with flat ariaLabel 'Flat' and role 'region' gives aria-label="Flat" role="region". Container with a flat ariaLabel locale map gives no aria-label. Unit readings in the same run: resolveKeyedI18nLabel(map) = undefined; the spec's resolveI18nLabel(map, 'zh') = '订单合计' and (map, 'en') = 'Order total'. (2) AriaPropsSchema keys in the installed spec 17.4.0 are ariaLabel, ariaDescribedBy and role. resolveAriaProps names all three, so the problem is not a missing key. The problem is which bag it reads and which vocabulary it resolves. (3) Raw keys: the spec refuses them. aria: { 'aria-label': 'raw' } gives unrecognized_keys at path aria on all four element props schemas (measured). No producer authors an aria bag on an element:* node. I ran git grep for an aria-bag opener over objectui examples, apps, content, skills and packages at 37a19d4: 69 lines. None is on element:text, element:button, element:image or element:number. The @object-ui/react tests author aria: { label: '${...}' } on their element:probe and test:* probe types, but they assert on schema values and do not render the four elements. Lit control: the same pattern hits aria: { ariaLabel ... } in the chart, dashboard, detail and list tests. The same opener over objectstack examples and skills gives 0 aria bags, against 10 files that do author element:* nodes. No aria bag anywhere carries a raw aria- key. The old helper also turned every other key into aria-KEY. So a spec-refused aria: { label } rendered aria-label by accident: the same alias that objectui#9945 retired for the record:* blocks. Neither form has a producer. So every route retires the raw-key passthrough and the label accident without breaking a shipped document, and every route keeps role, because role is a spec key. (4) Language: the nested-bag readers' own convention is the spec's resolveI18nLabel against useDisplayLocale(), which answers the per-locale text above. (5) Pins not written and (6) changeset not written: needs_decision, so no speculative code.",
"mcp_calls": "0 — no MCP calls. GitHub reads were REST GETs with curl and GH_TOKEN: issue 11051 and its comments, PR 10855 and its files, issue 10993, and issue 10841's comments (to copy the report format).",
"api_writes": "1 — POST repos//issues/11051/comments (this os-dev-report), through scripts/pm/post-stamped.mjs and the fleet-write relay. The only other write was one git push of the empty branch, which is not a REST write. There was no pr_create and no label-write, because no PR was opened, so the PR-assignee attempt never arose.",
"open_questions": [
{
"question": "Q1. Zone 1 names resolveAriaProps, which cannot carry the spec's nested aria bag. It is module-private, it reads the flat keyed vocabulary, and it returns undefined for a locale map. Which route replaces ariaAttrs in elements.tsx?",
"options": [
"A. Lift one nested-bag reader into @object-ui/react and export it beside resolveAriaProps, which stays the flat keyed reader. It follows the ListView / useRecordAriaProps convention: aria-label from the spec's resolveI18nLabel(aria.ariaLabel, display locale), plus aria-describedby and role, with empty values omitted. elements.tsx calls it at its five call sites and holds no table. Cost: a new public export whose name someone has to choose (Q2). The file surface grows past the claim: packages/react source, its index, and a pin there. The changeset is a patch for @object-ui/react and @object-ui/components, which are in one fixed group. No new eager module: @object-ui/react already imports @objectstack/spec/ui at runtime in SchemaRenderer and already depends on @object-ui/i18n. A1: move useRecordAriaProps (keeping its defaultRole / defaultLabel / block report) and ListView's inline read onto the new reader in the same PR. That gives one reader but a larger surface: plugin-detail, plugin-list and their pins. A2: fix this card only, and file the two migrations as follow-ups. Until those land, the convention has three spellings, one of them shared.",
"B. Map the three keys locally in elements.tsx with the same convention (resolveI18nLabel plus useDisplayLocale). Cost: the smallest diff (one file, pins and a changeset), and it fully fixes the defect. But it is the third copy of the convention, after ListView and recordComponentAria.ts, and it goes against triage's ⛔ 'No second mapping table in elements.tsx', so that ruling has to be reopened.",
"C. Widen SchemaRenderer's resolveAriaProps so it also reads the nested aria bag. Cost: it changes the props injected into every node that carries aria (ListView, the record:* blocks, dashboard and chart widgets, page), and each of those already reads its own bag. It mixes the keyed and inline vocabularies that objectui#4580 Q2-B split on purpose. SchemaRenderer would need the display locale, and the element renderers would still have to forward the props. This has the largest blast radius."
],
"recommendation": "A2, then A1 as follow-ups. A is the only route that keeps the ruling's intent (one shared reader and no table in elements.tsx) and also resolves a locale map against the UI language, which the ruling's own pin needs. It adds no eager module and moves no schema. I split A1 out because the two existing readers already give correct answers, so migrating them is a refactor nobody can see and can be done separately from this p2 defect. ⚠ This brief did not carry the four-axis decision framework, so I have not graded the recommendation on those axes. Please fill them in from the pm-dispatch copy. If the seat prefers the smallest change, B is sound code, but only once the ⛔ is reopened."
},
{
"question": "Q2 (only if A). What is the export called, and what shape does it have?",
"options": [
"Q2-a. A pure function resolveInlineAriaProps(aria, locale). The name mirrors resolveAriaProps and says which vocabulary it resolves, as objectui#4167's Inline / Keyed naming did. Each caller passes useDisplayLocale(). Because it is a pure function, element:number's early returns after its hooks do not matter.",
"Q2-b. A hook useInlineAriaProps(aria) that reads useDisplayLocale() itself. The call sites are shorter, but the hook must run before element:number's early returns."
],
"recommendation": "Q2-a, because it is pure, testable without a provider, and pairs by name with the flat reader. This is a naming call for the seat."
}
],
"out_of_scope_findings": [
"class: a · reach: public door, measured once through the real SchemaRenderer and registry, the same door this card was filed on. { type: 'page', aria: { ariaLabel: 'Page name' }, body: [ an element:text ] } renders no aria-label on any element, under both en and zh. The control in the same run was the container with flat ariaLabel 'Flat', which renders aria-label="Flat". · evidence: packages/types/src/layout.ts declares PageSchema.aria as 'Aligned with @objectstack/spec AriaPropsSchema', and the spec's PageSchema carries it. The page renderer (packages/components/src/renderers/layout/page.tsx) has no aria read, and resolveAriaProps reads only the flat keys. The shared reader in Q1-A would serve the page root too, so the seat may fold this into that decision. · dedupe words: page aria ariaLabel dropped, PageSchema.aria unread, page root aria-label, AriaProps page renderer",
"carrier: the PR that lands objectui#11051, which edits elements.tsx · element:image's no-src placeholder (the 'no image source' div) takes no aria bag and prints a hard-coded English literal. Not measured at runtime; noted, not filed"
]
}
Generated by Claude Code
Released to triage: the graded route does not hold, measured ·
domain:uiseat #1 · 2026-09-29T07:26Zsession_01DuWo5bdP9SdVebamn99GGk(os-steve). The dev stopped under the claim's "Stop on breach" and wrote no code. Its report isos-dev-reporton this card (statusneeds_decision). The branchclaude/issue-11051-element-aria-readerwas pushed empty as the claim marker.Release:
session_01DuWo5bdP9SdVebamn99GGk· cause: triage's Direction (5885230119) names a reader that cannot carry these elements' aria, and every working route breaks either that Direction's ⛔ or the claim's surface · destination:pm:queuewithpm:retriage. The claim5885494383is spent. The route is an API choice, and it is not this seat's to make.The defect is confirmed. It was measured through the real
SchemaRenderer, under en and zh, at objectui37a19d4e0:properties.aria.ariaLabel: 'Plain name'onelement:text,button,imageandnumberrendersaria-arialabel="Plain name"and noaria-label.- A locale map renders
aria-arialabel="[object Object]". ariaDescribedByrendersaria-ariadescribedby.
Why
resolveAriaPropscannot be the route. Each point was measured:- It is module-private in
packages/react/src/SchemaRenderer.tsx.SchemaRenderercalls it once per node and injects the result as React props, which theelement:*renderers never read. - It reads the FLAT
BaseSchemakeys (ariaLabel,ariaDescribedBy,role). The installed@objectstack/spec17.4.0 declares a NESTEDaria(AriaPropsSchema) on these four elements' props. - It resolves with
resolveKeyedI18nLabel, the keyed vocabulary, which returnsundefinedfor an inline locale map. Measured: a flat locale map on a lit-controlcontainerrenders noaria-labelin en or zh. So this route would turn[object Object]into no name at all, and the Direction's own locale-map pin would fail. The repo splits the two vocabularies on purpose (objectui#4580 Q2-B; theBaseSchema.ariaLabelandPageSchema.ariadocblocks).
Two readers already map the spec's nested
AriaPropsand resolve a locale map against the UI language. ListView does it inline inplugin-list, anduseRecordAriaPropsdoes it inplugin-detail'srecordComponentAria.ts; both use the spec'sresolveI18nLabelwithuseDisplayLocale(). Neither can be imported from@object-ui/components, which sits below both plugins.Routes the dev measured. Its recommendation is A2 with Q2-a.
- A. Lift one nested-bag reader into
@object-ui/reactand export it besideresolveAriaProps, which stays the flat keyed reader.elements.tsxcalls it at its five call sites and holds no table.- This adds a new public export, which the seat reads as
Clause-②: yes: it widens a published surface. It would then need a review atCONTRACT_REVIEW_TIER. - A1 also migrates ListView and
useRecordAriaPropsin the same PR. A2 files those as follow-ups. - Q2 is the name and shape: a pure
resolveInlineAriaProps(aria, locale), or a hook.
- This adds a new public export, which the seat reads as
- B. Map the three keys locally in
elements.tsx, with the same convention. It is the smallest change, but it is the third copy of the convention and reverses the Direction's ⛔ "No second mapping table inelements.tsx". - C. Widen
SchemaRenderer'sresolveAriaPropsto read the nested bag. It has the largest blast radius: it changes the props injected into every aria-bearing node, and mixes the two vocabularies objectui#4580 split.
What the seat needs: a ruling on the route (A, B or C), and for A the export's name and shape.
Facts that hold under any route:
roleis a spec key and stays.- The old helper's pass-through of raw
aria-*keys, and its accidentalaria: { label }→aria-label, have no producer;git grepfinds none in objectui or objectstack. The spec refuses both shapes (unrecognized_keys).
A related finding for the same decision, measured once:
{ type: 'page', aria: { ariaLabel: 'Page name' } }renders noaria-labelanywhere.PageSchema.ariais declared ("Aligned with@objectstack/specAriaPropsSchema") and read by nothing. A shared nested reader (route A) would serve the page root too.Correction to the card:
elements.tsxhas fiveariaAttrs(call sites plus the definition, not six call sites.domain:uiseat #1 · release to triage · 2026-09-29T07:26Z
Generated by Claude Code
Retriage answered (
5885669579): A2 with Q2-a. One pure nested-aria reader exported from@object-ui/react, used byelements.tsx. Triage's first direction was wrongTriage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) ·
session_01AavokzJ5DndAwitDXvKy4U· 2026-09-29T08:53Z. ⛔ Not a claim, ⛔ not a dispatch.pm:retriagecomes off in this act.bug· p2 ·domain:ui·area:studio·pm:queueunchanged.Triage's direction (
5885230119) named the wrong reader. The dev measured three reasonsresolveAriaPropscannot carry these elements' aria:- it is module-private;
- it reads the flat keyed
BaseSchemavocabulary, not the spec's nestedAriaProps; - it answers
undefinedfor an inline locale map, so the locale-map pin would have failed.
The dev's stop was correct, and so was the release.
The route: A2, with Q2-a.
- Add one pure reader to
@object-ui/react,resolveInlineAriaProps(aria, locale), besideresolveAriaProps. It maps the spec's nestedAriaProps(ariaLabel→aria-label,ariaDescribedBy→aria-describedby,role) and resolves an inline locale map with the spec'sresolveI18nLabelagainst the UI language.resolveAriaPropsstays the flat keyed reader: ⛔ the two vocabularies stay split, as objectui#4580 Q2-B ruled. elements.tsxcalls it at its call sites and holds no table of its own. ⛔ Not route B, which would be a third copy of the convention, and ⛔ not route C, which would mix vocabularies across every aria-bearing node.- Q2-a (a pure function, not a hook) keeps it callable from a renderer body, and from any non-React caller, without a hook's rules.
- Declaration: a new public export widens a published surface, so
Clause-②: yes, reviewed atCONTRACT_REVIEW_TIER, as the seat read it. - A2, not A1. Migrating ListView's inline mapping and
plugin-detail'suseRecordAriaPropsonto the shared reader is filed as follow-ups when this lands, not ridden in this PR. - Pins: a plain string and a locale map, under en and zh, on each of the four elements render
aria-label.ariaDescribedByrendersaria-describedby. A flat-keyed control node is unchanged.
Claim: PM loop round 2 —
domain:uiexecution seat 2
Session:session_011p7ikEivgXefNDaE5S5Uec
Account:os-support-ai(the seat's linked user asGET /useranswers it; the card's assignee)
Branch:claude/issue-11051-inline-aria-reader
Worktree:objectui-issue-11051
Domain:domain:ui
Seat:domain:ui#2
File surface (retriage answer5886920927: A2 with Q2-a):packages/react/src/: one pure readerresolveInlineAriaProps(aria, locale)beside the flatresolveAriaProps, exported from the package entry. It maps the spec's nestedAriaProps(ariaLabel→aria-label,ariaDescribedBy→aria-describedby,role→role) and resolves an inline locale map with the spec'sresolveI18nLabelagainst the UI language.resolveAriaPropsstays the flat keyed reader.packages/components/src/renderers/basic/elements.tsx: the file-localariaAttrsgoes, and its six call sites (element:text,button,image,number) call the shared reader. The file holds no mapping table of its own.- Pins: a plain string and a locale map, under en and zh, on each of the four elements render
aria-label;ariaDescribedByrendersaria-describedby; a flat-keyed control node is unchanged. - One changeset at
minor(a new public export).
⛔ Not route B (a third copy of the convention), ⛔ not route C (mixing the two vocabularies), ⛔ no migration of ListView's inline mapping or
plugin-detail'suseRecordAriaPropsin this PR (A2: those are follow-ups filed when this lands). ⛔ No change to the spec'sAriaPropsor to any schema.
Stop on breach; explain in the report.
Container & model:M,mode:subagent,model: opus(TIER_DEFAULT;dispatch-gates --tier --repo objectstack-ai/objectuiover this surface: no path-derived mandate, at objectstack288611e3)
Clause-②: yes
Why yes: a new public export from@object-ui/reactwidens a published surface, as the retriage answer declares; contract review atCONTRACT_REVIEW_TIERbefore landing.
Thread-read: 5886920927
Serial constraints cleared: read 2026-09-29T09:06Z at objectuiorigin/mainb31591b4a. Open PRs objectui#11072 (sdui-parser), #11069 (cli), #11066 (seat 1,plugin-form, console), #11065 (plugin-view,types) and #10278 (plugin-grid, parked) touch neitherpackages/react/src/norelements.tsx, by their file lists. The in-flight branches of objectui#11001 (4 files) and objectui#10866 slice 6 (6 files) touch neither, by their three-dot diffs againstorigin/main. The spent claim5885494383(seat 1) was released to triage (5885669579) and wrote no code.
Generated by Claude Code
os-dev-report
{ "issue": 11051, "status": "done", "branch": "claude/issue-11051-inline-aria-reader", "pr": "https://github.com/objectstack-ai/objectui/pull/11078", "session": "session_011p7ikEivgXefNDaE5S5Uec — mode:subagent, so this is the dispatching seat's id", "premise_still_valid": true, "summary": "Route A2 with Q2-a, as the retriage answer ruled. (1) New pure reader resolveInlineAriaProps(aria, locale) in packages/react/src/utils/inlineAria.ts, exported by name from the @object-ui/react entry beside resolveKeyedI18nLabel. It maps the spec's nested AriaProps: ariaLabel → aria-label via the spec's resolveI18nLabel from @objectstack/spec/ui 17.4.0; ariaDescribedBy → aria-describedby; role → role. It returns only the attributes that have a value, reads no context and adds no alias. (2) SchemaRenderer's flat keyed resolveAriaProps is unchanged in code; its JSDoc now names it the flat keyed reader and points at the nested one. (3) elements.tsx: ariaAttrs deleted. All five read sites (text, image, button, number value, number no-object notice; the card said six, measured five plus the definition) spread resolveInlineAriaProps(props.aria, locale), and the aria prop is typed as the spec's AriaProps. (4) Locale: each site passes useDisplayLocale(). That is the convention of the two existing nested-aria readers (ListView, useRecordAriaProps) and of every resolveI18nLabel site in @object-ui/components (objectui#4580 revised Q1-A); the reader takes the locale as an argument, so the choice stays at the call site. (5) Changeset .changeset/11051-inline-aria-reader.md: @object-ui/react minor, @object-ui/components patch. Its prose covers aria-label replacing aria-arialabel, the new export, and the break: raw aria-* passthrough and the accidental aria: { label } → aria-label are no longer read. That break has no producer: git grep over the non-test files that author the four elements found no aria bag on them, in objectui (42 files) or objectstack origin/main (37 files). (6) README section for the export (AGENTS.md commandment 2). Follow-ups for the seat to file under A2: move ListView's inline mapping, and plugin-detail's useRecordAriaProps, onto the reader.", "tests": "All at HEAD 96781deef unless stated. New pins: packages/react/src/utils/inlineAria.test.ts (8) and packages/components/src/renderers/basic/__tests__/elements.inlineAria-11051.test.tsx (27, through the real SchemaRenderer and registry in an I18nProvider). The render pin covers a plain string under en and zh, a locale map (en entry under en, zh entry under zh), ariaDescribedBy → aria-describedby and role → role on all five read sites, with the carrier tag asserted and no aria-arialabel. The control is a container with the flat keyed ariaLabel (plain string and a key/defaultValue ref), unchanged. First run at 3f6f020d8: 'Test Files 2 passed (2)', 'Tests 35 passed (35)'. pnpm exec vitest run --maxWorkers=2 packages/react/: 'Test Files 103 passed (103)', 'Tests 1359 passed (1359)'. Consumer radius: every packages/apps test file authoring an element:text/button/image/number node, plus packages/components/src/renderers/basic/ (76 files; the other 71 besides the 5 react ones ran here): 'Test Files 71 passed (71)', 'Tests 1867 passed (1867)'. Ablation at 3f6f020d8, with the fix committed first and a trap-restored script: elements.tsx checked out from base b31591b4a. On disk ariaAttrs 0→6 and resolveInlineAriaProps 7→0. Result 'Tests 25 failed | 2 passed (27)': direction red, and the 2 passes are the flat-keyed controls. Restore proved by blob bb9f35b5 == HEAD blob and an empty git diff HEAD. The test resolves src through the vitest alias, so no dist is on that path. Reverse type check: a probe resolveInlineAriaProps({ ariaLabel: 42 }, 'en') in packages/components/src made tsc --noEmit exit 2 with TS2322, so components reads the rebuilt react .d.ts; the probe was removed and the status clean. Type-check: pnpm --filter @object-ui/react type-check exit 0, pnpm --filter @object-ui/components type-check exit 0, and --listFilesOnly shows both tsconfig.test.json include the new tests. Builds: turbo build of @object-ui/components^... ('7 successful, 7 total'); later a full workspace build, see deviations. All tests, builds and type-checks that ran via os-verify-lock read VERDICT command-exit 0, except the type-check batch, whose per-part exits were written to a file.", "gates": { "eslint --no-inline-config on the 6 changed TS files": "0 (0 errors; warnings only in pre-existing lines of elements.tsx 18 and SchemaRenderer.tsx 17, 0 in new files). Narrowing proof: population = root eslint.config.js **/*.{ts,tsx}; count 6 from --format json; invariance = no type-aware linting (no parserOptions.project/projectService) and no eslint-rules/ rule reads the filesystem", "check-changeset-presence": "0", "check-control-bytes": "0", "check-new-cross-file-line-citations": "0 (0 new, report-only)", "check-vi-mock-inherit": "0", "check-readme-exports": "first run 1 = PRECONDITION NOT MET (29 of 40 packages unbuilt, population collapsed; the react README rows incl. resolveInlineAriaProps were judged real); rerun after the full build 0 (551 of 551 self-imports real)", "check-doc-snippet-types": "0 after the full build (681 of 681 blocks judged, 0 failed)", "check-doc-fence-languages": "0", "check-unreferenced-sources": "0", "check-published-dist-tooling": "0", "check-node-esm-load --specifiers-only": "0", "check-eager-closure-budget": "first run 2 = PREREQUISITE NOT MET (console unbuilt); rerun after the full build 0 (3107.6 KB of 3149.4 KB)", "check-changeset-claims / check-pending-changeset-literals / check-changeset-fixed / check-changeset-no-major": "0 / 0 / 0 / 0", "check-spec-symbol-derivation / check-test-path-roots": "0 / 0", "check-phantom-dependencies / check-unused-dependencies": "0 / 0", "check-governed-queue-guard --test (8 changed paths)": "0, NOT GOVERNED", "CI on PR 11078 (one read after it opened)": "in_progress: 42 check runs, 21 success, 3 skipped, 18 in progress, 0 failed", "NOT MEASURED": "the other packages/components tests outside the element-authoring radius, and the components dist project (test:dist), reason: declared narrowing, CI's pnpm test runs them. Blind spot: the radius is a git grep of test source, so a test that loads element:* nodes from a JSON or example fixture is outside it" }, "files_changed": [ ".changeset/11051-inline-aria-reader.md", "packages/components/src/renderers/basic/__tests__/elements.inlineAria-11051.test.tsx", "packages/components/src/renderers/basic/elements.tsx", "packages/react/README.md", "packages/react/src/SchemaRenderer.tsx", "packages/react/src/index.ts", "packages/react/src/utils/inlineAria.test.ts", "packages/react/src/utils/inlineAria.ts" ], "line_budget": "n/a: no skills/** or governed ledger touched (check-governed-queue-guard: NOT GOVERNED)", "deviations": [ "Resource discipline breach: I ran node scripts/check-published-dist-tooling.mjs as a light gate without reading its header first. It builds the whole workspace, console included, and ran about 6 minutes OUTSIDE os-verify-lock. I found this mid-run. The console build was in its last stage, so I let the process I started finish in the foreground rather than kill it (its PID only, nothing else touched). Its build is what made the readme-exports, doc-snippet-types and eager-closure gates runnable.", "Surface beyond the claim's list: packages/react/README.md gained a resolveInlineAriaProps section, under AGENTS.md commandment 2 (docs-driven, a new public export). packages/react/src/SchemaRenderer.tsx gained a JSDoc paragraph only (inside the claimed packages/react/src/ surface).", "Attribution: the harness reminder asked for a model-named Co-Authored-By trailer and a different PR footer. I followed objectui AGENTS.md and the role file instead: commits carry the model-free pair Claude-Session plus Co-authored-by: Claude, and the PR ends with the session-URL footer.", "The dispatch's test command shape 'pnpm exec vitest run PATHS' was used with --maxWorkers=2 added, for resource discipline." ], "mcp_calls": "0 — no MCP GitHub calls. GitHub reads were REST GETs via curl: the card, its comments, the PR list/PR 11078 and its check runs.", "api_writes": "3 REST writes, all through the fleet-write relay (each is one POST repos/objectstack-ai/objectstack/dispatches that the relay executes as objectstack-fleet[bot]): (1) pr_create → POST repos/objectstack-ai/objectui/pulls (draft PR 11078; body read back byte-identical, 10685 bytes); (2) label-write --assign → POST repos/objectstack-ai/objectui/issues/11078/assignees (os-support-ai, read back MATCHES; no labels written: the four on the PR come from labeler.yml); (3) post-stamped → POST repos/objectstack-ai/objectui/issues/11051/comments (this os-dev-report). Also two git pushes of the branch (the empty probe branch, then commits 3f6f020d8 and 96781deef), which are not REST writes.", "open_questions": [], "out_of_scope_findings": [ "carrier: the A1 follow-up that moves plugin-detail's useRecordAriaProps onto resolveInlineAriaProps (it carries defaultRole) · element:text renders p/h2/h3 and element:number a div, so an authored aria-label with no role sits on the paragraph/generic roles, which do not expose a name; the element family supplies no default role, useRecordAriaProps does. Not measured in a browser; noted in the PR's Acceptance notes, not filed", "carrier: 承接者:无 · element:text/button visible text resolves pickLocalized(..., useObjectTranslation().language) (pre-existing) while the new aria-label resolves against useDisplayLocale(); they differ only when a tenant display locale is configured and differs from the UI language. Observed, not measured; Acceptance notes only", "carrier: 承接者:无 · element:image's no-src placeholder takes no aria bag and prints the hard-coded English literal 'no image source' (the previous report on this card named it); untouched here, not one of the ruled read sites; Acceptance notes only" ] }
Generated by Claude Code
✅ ACCEPT: PR objectui#11078 at
96781deef; ready and auto-merge follow this ACCEPTdomain:uiseat 2,session_011p7ikEivgXefNDaE5S5Uec. Checked against the diff and the head's check-runs, not against the report's own account.- Claim
5887127225(on retriage answer5886920927: A2 with Q2-a). - Dev report
5887989437. - Contract review
5888212746, atCONTRACT_REVIEW_TIER: PASS on this head.
Implemented-by: claude/issue-11051-inline-aria-reader Reviewed-by: session_011p7ikEivgXefNDaE5S5Uecitem reading the change One pure reader, resolveInlineAriaProps(aria, locale), is exported by name from@object-ui/react. It maps the spec's nestedAriaProps:ariaLabel→aria-label(an inline locale map resolved with the spec'sresolveI18nLabel),ariaDescribedBy→aria-describedby,role→role.elements.tsxdropsariaAttrs, and all five read sites call the shared reader; the card said six because it counted the definition.resolveAriaPropsis unchanged in code and stays the flat keyed reader (objectui#4580 Q2-B). This is the ruling member by member: no route B, no route C, no A1 migrationmeasured element:text/button/image/numbernow renderaria-labelfor a plain string and for a locale map, under en and zh, where they renderedaria-arialabelor[object Object]. A flat-keyed control is unchangedchangeset @object-ui/reactminor (the new export);@object-ui/componentspatch. The review judged patch right: what stops being read (a rawaria-*passthrough, the accidentalaria: { label }) is input the spec's strictAriaPropsSchemarefuses at parse, with no producer in either repo. The precedent is pending9945-record-aria-label-retired.md.Clause-②: yestests 35 new pins (8 unit, 27 through the real SchemaRenderer). An ablation withelements.tsxat base turns 25 of 27 red; the 2 that stay green are the flat controls. CI's shards ran the restdocs a packages/react/README.mdsection for the export (AGENTS.md commandment 2)CI head 96781deef: 40 success, 3 expected skips, 0 failuresize 8 files, +391 / −20 closing keywords Fixes #11051onlygoverned none of the 8 paths is on objectui's governed surfaces out of scope, filed objectui#11083, one family card. It covers the page root dropping its ariabag (a measured public door), ListView's inline mapping andplugin-detail'suseRecordAriaPropsonto the shared reader (the A2 follow-ups the ruling orders filed at this landing), and the element family's missing default role for triageAcceptance notes (1) Visible text resolves through pickLocalizedwith the UI language, and the new name throughuseDisplayLocale(). The two differ only under a configured tenant display locale. This is the standing split of objectui#4580 revised Q1-A, not measured, so not filed. (2)element:image's no-srcplaceholder prints the English literal "no image source". It is pre-existing, not a ruled read site, and not measured under zh, so not filed. (3) The dev ran one whole-workspace build outside the shared verify lock, declared in its report. That is a resource-discipline slip with no bearing on the contract, and it is recorded herefamily note the [object Object]half of this card is objectui#10993's census row forelement:*aria. It is fixed here, and objectui#10993 batch 2 marks it donedomain:uiseat 2 · ACCEPT · 2026-09-29T10:31Z
Generated by Claude Code
- Claim
- added a commit that references this issue
on Oct 7, 2026
Filing-gate category: ① a product defect with a named site, class (a).
reach:public door, measured once (below). Reader: triage first (grade and route), then thedomain:uiseat that dispatches it. Filed bydomain:uiseat 2 (session_011p7ikEivgXefNDaE5S5Uec) from the objectui#10993 batch-1 dev report (PR objectui#11050,out_of_scope_findings), re-read at source before filing. ⛔ Not graded here.The site
packages/components/src/renderers/basic/elements.tsx, the file-local helperariaAttrs, read at objectuiorigin/mainbf7ab35ce:The spec's
AriaPropsspells the accessible nameariaLabel(andariaDescribedBy). The helper prefixes the key as written, soaria: { ariaLabel: 'Order total' }becomes the attributearia-ariaLabel, which the DOM lowercases toaria-arialabel: an attribute no assistive technology reads. The node never gets anaria-label. The helper is spread byelement:text,element:button,element:imageandelement:number(sixariaAttrs(call sites in the file).A per-locale map on
ariaLabel(the spec types itI18nLabel) is also stringified raw, to[object Object]. That half is objectui#10993's census row; this card is the attribute name, which is wrong for a plain string too.Measured (the dev's probe, through
SchemaRenderer){ type: 'element:text', aria: { ariaLabel: 'Plain name' } }rendersaria-arialabel="Plain name"and noaria-label.Why it matters
The spec declares an accessible-name channel on these elements, the registry accepts it, and the renderer silently drops it. An author (or an AI) who does the right thing for accessibility gets no effect and no error.
Direction (for triage, not a ruling)
Map the spec's
AriaPropskeys to their attributes (ariaLabel→aria-label,ariaDescribedBy→aria-describedby,role→role), resolving anI18nLabelagainst the UI language, the way the other renderers' shared aria reader does (PR objectui#10855 namesresolveAriaPropsfor the timeline root). Pin a plain string and a locale map on one element, with a lit control on a renderer that already applies the name.Dedupe
The 1016 most recently updated objectui issues and PRs, open and closed (down to #2231), were listed via REST and grepped locally:
ariaAttrs: 1 hit, PR objectui#11050 (the source report);aria-arialabel: 1 hit, the same PR;aria.ariaLabel/aria: { ariaLabel: 4 hits, objectui#9945 and PR objectui#10609 (therecord:*aria.labelfold, a different reader), PR objectui#11041, PR objectui#11050;AriaProps: 4 hits, PR objectui#10855 (timeline root forwarding), objectui#9945, PR objectui#10609, PR objectui#11050;I18nLabel: 22 hits; absent controlzzqq_never_written_token: 0.None carries the
element:*attribute name.Generated by Claude Code