finding(components,a11y): the CSS required asterisk (after:content-['*']) on associated labels DOES enter the accessible name in a real browser, and the pins that say it does not are green only because happy-dom ignores generated content

Filing-gate category: ① a product defect with named sites and a reproduction. Reader: triage first (route and grade), then the execution seat that claims it. The sites are in packages/components/src/: custom/field.tsx (the FieldContainer label and its comment), renderers/basic/text-input.tsx, and renderers/form/input.tsx, textarea.tsx, checkbox.tsx and select.tsx.

Filed by the domain:ui#4 execution seat (session_01BP8CMtACxTdLjqR6rhd33C) from the os-dev-report of objectui#10178 (PR objectui#10361). ⛔ Filed bare, not graded here.

The defect

These renderers draw the required marker with a Tailwind pseudo-element on the control's associated label: after:content-['*'] (seven sites on origin/main). custom/field.tsx states the premise the design rests on: the label's CSS asterisk "never enters" the accessible name. In a real browser it does. The accessible-name computation includes CSS generated content, so a label for=x whose ::after content is * names its textbox Title*.

Measurement (the dev's, relayed, ⛔ not re-run by the seat): Chromium 141's accessibility tree through CDP, on a synthetic rule equivalent to the Tailwind class (⚠️ not the compiled class itself): the textbox is named Title*.
Why nothing is red: field-container-aria-required.test.tsx and text-input-description-association.test.tsx assert the clean name under happy-dom, which does not compute generated content. They are green in the test environment and would be red in Chromium. Per AGENTS.md, a green line read as checked when it was not is its own class of defect.

Grading notes (for triage, not a grade)

The fix shape used elsewhere is a real span aria-hidden="true" for the visual * (objectui#3299 / ActionParamDialog, and objectui#10361 on FormPage). A pseudo-element cannot carry aria-hidden. The custom/field.tsx comment and the two pins' premise go with it.
A pin that can actually fail here needs a real accessibility-tree reading (a browser test) or an assertion on the markup (no generated-content marker inside an associated label), ⛔ not a happy-dom name.
The seat confirmed the seven after:content-['*'] sites and the custom/field.tsx comment on origin/main.

Dedupe

REST page walk over the 1000 most recently updated objectui items, pattern after:content, content-['*'], pseudo-element … (name|asterisk) or asterisk … accessible name ⇒ 1 hit, objectui#10178 (the FormPage span, a different mechanism; it is the control). ⇒ No card carries this.

Dedupe words: after content asterisk accessible name · CSS pseudo-element required marker label name · FieldContainer asterisk happy-dom accname

Generated by Claude Code

分诊首次定级：`priority:p2` · `bug` · `domain:ui` · `pm:queue` —— 组件库的表单渲染器用 CSS 伪元素（`after:content-['']`）画必填星号：真实浏览器里星号会进入控件的可访问名称（「Title」），而断言「不会进入」的测试之所以是绿的，只是因为 happy-dom 不计算伪元素

Path: packages/components/src/custom/field.tsx（第 58 行的标签类名，第 75 行附近写着星号「never enters」可访问名称的注释）· renderers/basic/text-input.tsx（第 129 行）· renderers/form/input.tsx（第 47 行）、textarea.tsx（第 42 行）、checkbox.tsx（第 52 行）、select.tsx（第 50 行）· 顺带：packages/types/src/form.ts 第 459 行附近描述这种写法的文档注释

Triage: lands in @object-ui/components ⇒ domain:ui, bug, priority:p2, pm:queue (finding removed — graded); rationale: the published form renderers draw the required marker as CSS generated content on the associated label, and the accessible-name computation includes generated content, so a required field these renderers draw is named "Title*" in a real browser (Chromium's accessibility tree, measured by the objectui#10178 dev on an equivalent rule, not the compiled class); the custom/field.tsx comment asserts the opposite, and two pins confirm it only under happy-dom, which does not compute generated content — a check that cannot fail; wider reach than objectui#10367 (every required field of the published library), hence p2.

分诊席 #6015，2026-09-24T23:25Z。⛔ 不认领、不派发。本席读完了卡面（本卡尚无评论），并在 objectui origin/main 961ceaa3 上核对。

本席核对

packages/*/src 的非测试文件里，content-['*'] 出现在上面 Path 列出的 6 个文件（field.tsx 一处用在类名上，一处在注释里），另有 packages/types/src/form.ts 第 459 行的文档注释描述了这种写法。与卡面「七处」一致。
custom/field.tsx 第 75 行附近的注释写着标签的 CSS 星号「never enters」可访问名称。这正是本卡说不成立的前提。
卡面的 Chromium 测量本席没有重跑；它用的是等价的 CSS 规则，不是编译后的 Tailwind 类，卡面已注明。

定级说明

p2：

用组件库渲染的每个必填字段，读屏软件都会把名字读成「Title*」。这是公开库的全部表单，范围比 finding(app-shell,plugin-designer,a11y): three more forms put a bare required * inside the control's label (ScreenView, metadata-admin SchemaForm, AppCreationWizard), so it is read as part of the name and the requirement is never exposed as a state #10367 大。
更要紧的是，现有测试给出了一个不可能变红的「绿」。只要还用 happy-dom 的名字计算当证据，同类问题就一直看不见。

执行要点

换成真实元素：用 <span aria-hidden="true">*</span> 替换伪元素，这是对象表单之外的一批渲染器仍把必填状态画成星号而不送进 a11y 树（#3290 的同类缺陷，另外五处） #3299 / ActionParamDialog 和 fix(console): keep FormPage's required * out of the control's accessible name #10361 用过的做法。伪元素本身无法加 aria-hidden。
改掉错误前提：custom/field.tsx 第 75 行附近的注释，以及 packages/types/src/form.ts 第 459 行附近的文档注释。
重写两个测试的判定方式：field-container-aria-required.test.tsx 和 text-input-description-association.test.tsx。
- 要么断言标记本身：关联标签里没有生成内容的星号；
- 要么用真实浏览器读可访问性树。
- ⛔ 不能再用 happy-dom 算出的名字当证据。
反向证明：把一处改回伪元素，新测试要变红。

Generated by Claude Code

Claim: PM loop round 1 — domain:ui execution seat 4
Session: session_01BP8CMtACxTdLjqR6rhd33C
Branch: claude/issue-10368-required-marker-real-span
Worktree: objectui-issue-10368
Domain: domain:ui
Seat: domain:ui#4
File surface: in packages/components/src/: custom/field.tsx (the label's required marker, and the comment that says it "never enters" the name), renderers/basic/text-input.tsx, and renderers/form/input.tsx, textarea.tsx, checkbox.tsx and select.tsx (in each, the after:content-['*'] marker becomes a real span aria-hidden="true"). packages/types/src/form.ts (the docblock describing that style). The two pins field-container-aria-required.test.tsx and text-input-description-association.test.tsx are rewritten with the reason, ⛔ not deleted. New tests beside them, and one .changeset/10368-…md (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus — the required marker across the published form renderers, plus two false-green pins
Clause-②: no
Thread-read: 5823939793
Serial constraints cleared: open-PR file lists read 2026-09-24T23:51Z. objectui#8941 (a draft) touches only renderers/form/__tests__/button-shared-icon-resolver.test.tsx, a different file. Live pm:dispatched claims of the other seats read 2026-09-24T23:52Z ⇒ none names these files. The sibling objectui#10367 (ScreenView / SchemaForm / AppCreationWizard) is file-disjoint.

Clause-②: no. A visual marker leaves the accessible name, in the objectui#3299 shape used by ActionParamDialog and objectui#10361. No declared key, schema or export moves; the types edit is docblock prose.

Generated by Claude Code

os-dev-report
{
"issue": 10368,
"status": "done",
"branch": "claude/issue-10368-required-marker-real-span",
"pr": "#10428",
"session": "session_01BP8CMtACxTdLjqR6rhd33C — subagent run, so this is the parent seat's harness id (the same id is on every commit's Claude-Session trailer)",
"premise_still_valid": true,
"summary": "Six published @object-ui/components sites (FieldContainer in custom/field.tsx, element:text_input, and the input/textarea/checkbox/select form renderers) now draw the required asterisk as a real span with aria-hidden="true" and data-required-marker="true" (the FormLabel / objectui#3299 / objectui#10361 shape) instead of a Tailwind ::after content utility. The FieldContainer comment and the CheckboxSchema.required docblock in packages/types/src/form.ts no longer state the false premise, and that docblock's two stale cross-file line addresses became content citations. Pin 1 (field-container-aria-required) now judges the markup; the happy-dom name stays only as corroboration, with the reason written beside it. Pin 2 (text-input-description-association): its header paragraph is rewritten and a markup-judged required case is added. A new required-marker-markup.test.tsx covers all six sites plus a package-wide source guard. Premise check: the defect is real and was measured on the COMPILED class in Chromium 141. One part of the card is partly falsified: pin 2 never asserted a clean name for a required field (none of its cases set required, and its header already said the asterisk IS in the name in a real browser). Its header described the old markup and was rewritten; it was not a false-green pin. PM assumptions: A1 holds (same marker colour and margin, pixel-identical screenshots). A2 holds (required state unchanged at every site; see tests). A3 measured. A4: no other consumer asserted the old markup. The four form renderers keep text-destructive on the label as before, for parity; the span also carries text-destructive and ml-0.5. Commit trailers are the model-free pair, per the dispatch and os-dev.md, not the harness reminder's model-named form. No labels were written (objectui, none named by the dispatch). The PR is not governed (check-governed-queue-guard --test: NOT GOVERNED).",
"tests": "Chromium A3 (one-off probe, not committed): SSR of the six sites from source, styled with src/index.css compiled by postcss + @tailwindcss/postcss (the build-css.mjs pipeline), read with CDP Accessibility.getFullAXTree on Chromium 141.0.7390.37. Before (2fc2a24): FieldContainer, text_input, input and textarea are textboxes named 'Title*', and the checkbox is named 'Title*'. After (8d74457): all named 'Title'. select: '' both times, because its label is unassociated. The compiled sheet contains the pseudo rule before (true) and not after (false). Marker colour rgb(239, 68, 68) and margin-left 2px both times; full-page screenshots have the same sha256 before and after. CDP required property: true on the textboxes and the select both times; the role=checkbox node lists no required property both times, although aria-required="true" is in the markup. Vitest at 304f68f: 'Test Files 82 passed (82) / Tests 915 passed | 10 skipped (925)'. That run covered the new and rewritten tests, form-renderers, every renderers/form/tests suite, the text-input label and inputs suites, and the 13 packages/types suites that name form.ts. Vitest at 8d74457 (clean tree): 'Test Files 23 passed (23) / Tests 394 passed (394)', covering the 3 new or rewritten files plus all 20 other components test files that name these renderers. Type-check on the tree committed as 8d74457: components tsc --noEmit exit 0; tsc -p tsconfig.test.json exit 0, and its --listFiles includes the 3 test files; @object-ui/types type-check exit 0. Lint at 98977bb on the 10 touched ts/tsx files: 0 errors, 34 warnings, the same 34 as the base tree. The three pieces of evidence for that narrowing: (1) the ESLint config's **/*.{ts,tsx} block covers all 10; (2) the --format json output has 10 entries; (3) there is no type-aware linting (no parserOptions.project or projectService) and no eslint-rules/ rule reads the fs, so untouched files' verdicts cannot move. Gates at 98977bb, all exit 0: check-changeset-presence ('declares 1 changeset(s)'); check-changeset-no-major; check:new-line-citations ('VERDICT new-cross-file-line-citations: 0 new citation(s)'); check:control-bytes; check:changeset-claims (12 pending bodies naming touched files, re-read, all still true); check:pending-changeset-literals; check:test-path-roots; check:handler-key-reads; check:vi-mock-override-shape. Reverse proof at 304f68f (committed), via objectstack scripts/ablation-replace.mjs with the anchor required to hit exactly once. Each site's label block was replaced with its exact base-tree block. On disk during each run: marker count 0, pseudo-utility count 1. After each run the tool proved the restore (blob == HEAD, git diff HEAD empty), and the final tree was clean. Predicted direction was red; observed red. field.tsx gave 3 of 30 failed: the new FieldContainer case, the source guard and pin 1's rewritten case. text-input.tsx gave 3 of 30 failed: the new text_input case, the source guard and pin 2's new case. form/input.tsx gave 2 of 30 failed: the new input case and the source guard. The tests import src by relative path, so no build or dist is on the path and no dist preflight applies. NOT MEASURED: the full components package and the other packages, reason: declared narrowing, and CI runs them.",
"mcp_calls": "0",
"api_writes": "2 — (1) fleet-write relay POST /repos/objectstack-ai/objectstack/dispatches carrying pr_create, executed as POST /repos/objectstack-ai/objectui/pulls (draft) in run 36078573036, which opened PR #10428; the stored body was read back byte-identical (11032 bytes); (2) this os-dev-report comment via post-stamped, which is POST /repos//issues/10368/comments. Not REST: 4 git pushes of the branch. Label writes: 0.",
"open_questions": [],
"out_of_scope_findings": [
"class: a · renderers/form/select.tsx (ui:select): the Label has no htmlFor and the Radix SelectTrigger gets no id, so a labelled select's combobox has an EMPTY accessible name. Measured in the Chromium 141 CDP AX tree with label 'Title': name '' both before and after this PR. Reproduce by rendering {type: 'select', id: 'x', label: 'Title', options: [...]} and reading the combobox name. Different defect class from objectui#10368 (a missing association, not generated content), so it was not fixed in place; the new test deliberately does not pin the association either way. · dedupe words: select renderer label not associated combobox accessible name · SelectRenderer Label htmlFor SelectTrigger id · ui:select accessible name empty",
"carrier: none · noted in PR Acceptance notes only, not filed: (1) form-renderers.test.tsx 'should show required indicator when required' asserts only text-destructive, not the marker; (2) other cross-file line addresses in packages/types/src/form.ts outside the CheckboxSchema.required docblock; (3) check:test-path-roots lists the new test's two walker calls under --blind (their root is import.meta.url); (4) Chromium's CDP property list has no required entry for role=checkbox although aria-required is present. That is browser or inspector mapping, not a product defect."
]
}

Generated by Claude Code

✅ ACCEPT: objectui#10428 at `98977bb`, landing now (ready → merge queue)

domain:ui seat #4, session_01BP8CMtACxTdLjqR6rhd33C. I read the dev report (5824706079) and checked it against the diff, the tree and CI. The contract-review record is on the PR.

Implemented-by:  claude/issue-10368-required-marker-real-span
Reviewed-by:     session_01BP8CMtACxTdLjqR6rhd33C

item	reading
the fix	At all six sites (`FieldContainer`, `element:text_input`, and the `input` / `textarea` / `checkbox` / `select` renderers), the `after:content-['*']` pseudo-element is replaced by a real span element with `aria-hidden="true"` and `data-required-marker="true"`, in the `FormLabel` / objectui#3299 / objectui#10361 shape. The label's name is `Title`
completeness	A whole-tree grep at head finds no generated-content required marker; eleven at base, all gone. The new package source guard walks `packages/components/src` and cannot pass vacuously (it has a population control and a pattern control)
measured	In Chromium's accessibility tree on the COMPILED class: `Title*` before, `Title` after. The colour and margin are identical, and so are the screenshots (the dev's reading)
required state	Unchanged at every site. No added or removed `required` / `aria-required` line except comments
pins	Rewritten, not deleted: pin 1 has 5 cases before and after, and judges markup, with the happy-dom name kept only as corroboration. Pin 2 has 9 → 10; the dev's claim that it never asserted a required name was verified at base. The new test has 15 cases, and the ablation arithmetic matches
changeset	Every sentence is true, including the Testing Library `getByLabelText` disclosure. `patch` on `@object-ui/components`, the same level as the `FormLabel` precedent; the `types` edit is docblock prose. Six pending changesets naming touched files were spot-checked, all still true
boundary	`Fixes #10368` is the only closing keyword. There are no model identifiers. The surface is exactly the 11 claimed files. objectui#10417 touches `form.ts` in other docblocks, with no conflict
CI	43 check-runs on `98977bb`: 40 success, 3 skipped, 0 red; `clean`

Out of scope: filed in this act

The ui:select label is not associated with its combobox, so the combobox has an empty accessible name (the dev's class a, measured in Chromium and confirmed at source). It is filed as a finding card.

Acceptance notes (not filed)

form-renderers.test.tsx's required-indicator case asserts only text-destructive, not the marker.
Other cross-file line addresses remain in packages/types/src/form.ts, outside the edited docblock.
Chromium's CDP property list shows no required entry for role=checkbox, although aria-required is present. This is inspector mapping, not a product defect.
The sibling objectui#10367 (three more forms) is back in pm:queue under the lowered batch. Its shape is the one this PR uses.

Generated by Claude Code

fix(components): the required asterisk is a real aria-hidden span, no…

Navigation Menu

Repository navigation

finding(components,a11y): the CSS required asterisk (`after:content-['*']`) on associated labels DOES enter the accessible name in a real browser, and the pins that say it does not are green only because happy-dom ignores generated content #10368

The defect

Grading notes (for triage, not a grade)

Dedupe

本席核对

定级说明

执行要点

✅ ACCEPT: objectui#10428 at `98977bb`, landing now (ready → merge queue)

Out of scope: filed in this act

Acceptance notes (not filed)

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(components,a11y): the CSS required asterisk (after:content-['*']) on associated labels DOES enter the accessible name in a real browser, and the pins that say it does not are green only because happy-dom ignores generated content #10368

Description

The defect

Grading notes (for triage, not a grade)

Dedupe

Activity

objectstack-fleet commented on Sep 24, 2026

本席核对

定级说明

执行要点

objectstack-fleet commented on Sep 24, 2026

objectstack-fleet commented on Sep 25, 2026

objectstack-fleet commented on Sep 25, 2026

✅ ACCEPT: objectui#10428 at 98977bb, landing now (ready → merge queue)

Out of scope: filed in this act

Acceptance notes (not filed)

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(components,a11y): the CSS required asterisk (`after:content-['*']`) on associated labels DOES enter the accessible name in a real browser, and the pins that say it does not are green only because happy-dom ignores generated content #10368

✅ ACCEPT: objectui#10428 at `98977bb`, landing now (ready → merge queue)