Repository navigation
finding(components): the header-bar renderer ignores the inherited BaseSchema.className #10397
Description
Activity
分诊首次定级：
priority:p3·bug·domain:ui·pm:queue—— 页头（header-bar）渲染器不理会继承来的BaseSchema.className，写了自定义样式也没有效果；文档页也有两处与代码不符Path:
packages/components/src/renderers/navigation/header-bar.tsx（第 78 行<header className="flex h-14 sm:h-16 …">是写死的类名）·content/docs/components/navigation/header-bar.mdxTriage: lands in
@object-ui/components⇒domain:ui,bug,priority:p3,pm:queue; rationale:HeaderBarSchemaextendsBaseSchema, whoseclassNameis the declared Tailwind override channel and which AGENTS.md commandment #3 requires every component to expose, but the renderer's root carries a fixed class list with noschema.classNamemerge (the dev's probe: markup byte-identical with and withoutclassName, withrightContentas the lit control); a styling key silently ignored, hence p3; PR objectui#10395 (objectui#10387) has landed as9b281519and touched the same doc page, so nothing blocks.分诊席 #6015，2026-09-25T01:30Z。⛔ 不认领、不派发。本席读完了卡面（本卡尚无评论），并在 objectui
origin/main57a2bc28上核对。本席核对
header-bar.tsx第 78 行：<header className="flex h-14 sm:h-16 shrink-0 items-center gap-2 border-b px-3 sm:px-4">。这是写死的字符串，没有合并schema.className。- PR fix(types): retire the header-bar keys nothing reads or authors (nav, left, center, height) (objectui#10387) #10395（finding(types,components):
HeaderBarSchemadeclares eight keys theheader-barrenderer never reads —logois also mirrored as a widerSchemaNodearm (objectui#7759 residue) #10387，退役页头上没人读的几个键）已合并为9b281519，也改过header-bar.mdx。 - 卡面的渲染探针本席没有重跑。
定级说明
p3：作者写的样式覆盖不生效，但页头的默认外观正常。
执行要点
- 在根元素上用
cn(…, schema.className)合并，写法与其他渲染器一致。 - 顺带修文档：
header-bar.mdx的 Schema 块补上search、actions、rightContent；「固定h-16」改成与代码一致的h-14 sm:h-16。 - 钉子：带
className的页头，根元素上出现这个类；不带时标记不变。
Generated by Claude Code
Claim: PM loop round 1 —
domain:uiexecution seat 4
Session:session_01BP8CMtACxTdLjqR6rhd33C
Branch:claude/issue-10397-header-bar-classname
Worktree:objectui-issue-10397
Domain:domain:ui
Seat:domain:ui#4
File surface:packages/components/src/renderers/navigation/header-bar.tsx: the rootheadermergesschema.classNamethroughcn(…), as the other renderers do.content/docs/components/navigation/header-bar.mdx: its Schema block gainssearch,actionsandrightContent, and the fixed-height sentence matches the code'sh-14 sm:h-16.- Tests beside the renderer, and one
.changeset/10397-…md.
Stop on breach; explain in the report.
Container & model:S,mode:subagent,model: opus— execution points from the triage5825183690
Clause-②: no
Thread-read: 5825183690
Serial constraints cleared:- The open-PR file lists read 2026-09-25T03:54Z show no open PR touching
header-bar.tsxorheader-bar.mdx. - The precedent on that page, PR objectui#10395 (objectui#10387), is merged
9b281519. - The live
pm:dispatchedclaims of seats 1–3, read 2026-09-25T04:02Z, name neither file.
Clause-②: no. The inheritedBaseSchema.classNameis already declared onHeaderBarSchema, and the renderer starts honouring it. No declared key, schema, export or accept set moves.
Generated by Claude Code
os-dev-report
{ "issue": 10397, "status": "done", "branch": "claude/issue-10397-header-bar-classname", "pr": "https://github.com/objectstack-ai/objectui/pull/10481", "session": "session_01BP8CMtACxTdLjqR6rhd33C — subagent run, so this is the parent seat's harness-stamped id (container CLAUDE_CODE_REMOTE_SESSION_ID cse_01BP8CMtACxTdLjqR6rhd33C)", "premise_still_valid": true, "summary": "Premise confirmed at origin/main 34bfd6060. The header-bar root was the literal class string, and a real-SchemaRenderer probe drew it byte-identical with and without className. The root is now cn(HEADER_BAR_CLASS, schema.className): the chrome classes first and the authored class last, so an authored h-20 replaces h-14, per variant (sm:h-16 stays unless the author writes sm:h-20). Without className the class attribute is the pre-repair string. The docs page gains a Styling section (JSON first), and its Schema block gains search / actions / rightContent with their declared types; all three are declared on HeaderBarSchema at head and read by the renderer. The fixed-h-16 note and the retired-height line now match the code. The changeset is .changeset/10397-header-bar-classname.md (components, patch). In the pending objectui#10387 changeset, two body sentences this change falsified are corrected ('reads only actions, crumbs, rightContent and search' and 'height is fixed by the renderer'); the frontmatter is byte-identical. Mechanism assumptions: A1 holds. The registered function took ({ schema }) only, and no registry wrapper touches className. SchemaRenderer hands the same value on schema.className and on the className prop, including the responsiveStyles scope class. So one channel is read. Reading both prints a custom class twice, because tailwind-merge keeps 'x x'. Deviation from the dispatch route: 'neither the schema's nor the prop's is dropped' is not followed for the one path where the two differ, a host passing className to SchemaRenderer itself. No in-tree host does that; see open_questions[1]. A2: style, id, testId, ariaLabel and data-obj-* are all still dropped on this root, measured one key at a time. They are not fixed, because the converged toDomProps shape would change the no-className markup that the control pin forbids; filed below. A3: no gate type-checks this page's fences (json plus plaintext only). The new JSON example passes the published zod HeaderBarSchema.safeParse, with a retired-key control that fails. A side effect, measured and pinned: a header's responsiveStyles scope class now reaches the root too. Conflict, stated rather than silently resolved: the role file says published text that this change falsifies must be fixed, and the claim says 'Stop on breach'. Two stale texts sit outside the claimed file surface and were NOT edited: the nine header-bar tombstone refusals and the @deprecated docs in @object-ui/types (not yet released), and the app-shell leak-sweep READY_OVERRIDE_REASONS['ui:header-bar'] reason. Both are named in the PR's Acceptance notes and in open_questions[0]. Claim 5826485336 is the newest Claim: and names this branch. The assignee was already os-litant and was not touched. Commit trailers are the model-free pair from the dispatch. The worktree is removed (the tree was clean and the remote branch equals the local head 073c75882).", "tests": "All at 073c75882 unless noted; exit codes captured before any pipe. [1] Dependency closure: `pnpm --workspace-concurrency=2 --filter '@object-ui/components^...' run build`. VERDICT command-exit 0, 8 of 47 projects. [2] `pnpm --filter @object-ui/components run type-check` (tsc --noEmit && tsc -p tsconfig.test.json). VERDICT command-exit 0. --listFilesOnly shows tsconfig.test.json includes the new test (1) and tsconfig.json includes the renderer. [3] One vitest run from the repo root over 18 files: header-bar-classname-10397 (new), header-bar-crumb-icon, header-bar-right-content-numeric-falsy, node-slot-numeric-falsy, container-declaration-census, container-declaration-ratchet, layout-containers-declare-containment, page-dom-leak-whitelist-7933, overlay-trigger-bare-string-9710, html-page-lazy-blocks, app-shell widget-dom-leak-sweep, types header-bar-unread-keys-10387, mirror-groups-cd-10286, zod-mirror-parity, content-channel-family-d-9256, eslint-rules no-bare-node-slot-guard, scripts check-doc-expression-carriage, and schema-catalog catalog-gallery-render. Result: 'Test Files 18 passed (18) / Tests 1909 passed (1909)', VERDICT command-exit 0. [4] Reverse proof. Objectstack scripts/ablation-replace.mjs replaced 'className={cn(HEADER_BAR_CLASS, schema.className)}' with 'className={cn(HEADER_BAR_CLASS)}': anchor 1 to 0, blob 032e4a32 to ad97c001. At 073c75882 the result was 'Tests 5 failed | 2 passed (7)': every override row red, and the harness and no-className control green. Restored: blob == HEAD, git diff HEAD empty. There is no dist leg: the test imports the renderer from source through '../renderers'. A second mutation read both channels, cn(HEADER_BAR_CLASS, schema.className, className). At acc34b4ce it gave '2 failed | 4 passed': the appears-ONCE row and the exact class-list row went red. [5] Gates, each exit 0: check-changeset-presence ('2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'), check-changeset-no-major, check-changeset-fixed, check-changeset-claims, check-changeset-overwrite, check-pending-changeset-literals, check-new-cross-file-line-citations ('0 new citation(s)'), check-control-bytes, check-doc-component-types, check-doc-fence-languages, check-doc-example-ids, check-doc-expression-carriage ('230 json/jsonc fence(s), 230 parsed, 0 UNPARSED'), check-doc-links, check:icon-record-names, check-test-path-roots, check-vi-mock-override-shape, check-vi-mock-inherit and check-vi-mock-specifiers. check-changeset-claims is report-only and flagged 6645-header-bar-crumb-icon.md; I read it, and its sentences are past tense or about crumb glyphs, still true. check-changeset-overwrite is report-only and recorded the 10387 body correction (its case 2). [6] Lint, as a proven narrowing, in three parts. (a) Population: the package script is 'eslint .' and the root config's files are '**/*.{ts,tsx}', so mdx and md are out of it, and 2 touched lintable files remain. (b) Count: --format json read 2 files, 0 errors, 1 warning. The warning is react-refresh/only-export-components on BreadcrumbLabel; the base file linted via stdin gives the same rule. (c) Invariance: the resolved config has parserOptions {}, so linting is not type-aware and untouched files' verdicts cannot move. [7] NOT MEASURED: check-doc-snippet-types, reason: exit 2 PRECONDITION NOT MET, because it needs 34 packages built. Its population is ts/tsx fences, and header-bar.mdx has none before or after (json + 2 plaintext vs 2 plaintext at base). NOT MEASURED: check-sdui-registration-pins, reason: exit 2, because it needs a console build. It reads sideEffects and registrations, which this diff does not touch. Left to CI: the full test shards, repo-wide lint, and other packages' type-check. No export or declared type moved, so downstream consumers (...@object-ui/components) were not swept.", "mcp_calls": "0", "api_writes": "2 — (1) pr_create through the fleet-write relay (scripts/pm/with-fleet.sh --via dispatch). POST /repos/objectstack-ai/objectstack/dispatches returned HTTP 204; run 36095019666 then executed POST /repos/objectstack-ai/objectui/pulls (draft) as objectstack-fleet[bot], opening PR #10481. The body was read back byte-identical: 10168 = 10168 bytes, no differing index. (2) This os-dev-report comment: POST /repos/objectstack-ai/objectui/issues/10397/comments through scripts/pm/post-stamped.mjs. Not REST: git push of the branch, three times (the empty-branch probe, then two push-as-you-go pushes). Zero label writes: objectui's labeler applied 'package: components' and 'tests', and the dispatch named no labels.", "open_questions": [ { "question": "This PR makes two texts false that sit OUTSIDE the claimed file surface, so they were not edited ('Stop on breach'). (1) @object-ui/types: all nine header-bar tombstone refusals in packages/types/src/zod/navigation.zod.ts say the renderer 'reads only actions, crumbs, rightContent and search off the node', and the renderer now also reads className. The height refusal ends 'The height is fixed by the renderer'. The @deprecated docs in packages/types/src/navigation.ts say the same. This ships with the still-pending objectui#10387 changeset. (2) packages/app-shell/src/__tests__/widget-dom-leak-sweep.test.tsx: READY_OVERRIDE_REASONS['ui:header-bar'] now misstates the renderer. Should the claim's file surface be supplemented so a patch round on PR #10481 fixes them?", "options": [ "A: supplement the claim with the three files, and a patch round appends one commit to PR #10481. In types: 'reads only ... of its own keys', and the height migration points at className, plus a types patch line in the changeset. In the sweep: the header-bar target moves to COMPONENTS_READY, its reason is dropped, and the docblock and DOCBLOCK_COUNTS go from 155 matched / four overrides to 156 / three. Measured: with that selector the target's readiness passes, and only those bookkeeping rows go red.", "B: follow-up cards. The types text must land before the next release, because the objectui#10387 changeset is still pending. The sweep entry can land any time.", "C: leave both. The refusals' point (the retired keys render nothing) stays true." ], "recommendation": "A. Both edits are mechanical and already measured. The types text is not yet released, so fixing it rides the same release as objectui#10387 and this PR. The role file's rule is that published text a change falsifies is fixed with it, and only the claim's surface kept it out of this round." }, { "question": "Which className channel should header-bar honour when a HOST passes className to SchemaRenderer itself? SchemaRenderer spreads the host's props last, so the className prop becomes the host value while schema.className keeps the node's. This PR reads schema.className only (the triage's spelling), so on that path the host class is dropped. The dispatch route asked that neither be dropped. No in-tree host renders header-bar that way.", "options": [ "A: keep one channel, as in this PR. SchemaRenderer's own comment says it sets both 'so the scope class lands regardless of which channel a renderer honours'.", "B: read both in this renderer with de-duplication (cn(base, schema.className, className !== schema.className && className)). This is a consumer-side accommodation.", "C: settle it upstream: SchemaRenderer merges a host className into the className it hands on both channels, so every renderer gets it without reading two carriers." ], "recommendation": "A for now, and C if a host ever needs it. Per contract-first, the divergence is created upstream by the props-last spread, so a fix belongs in SchemaRenderer, not in one renderer." } ], "out_of_scope_findings": [ "class: b · The header-bar root still drops every other SchemaRenderer-injected prop: style, id, ariaLabel (as aria-label), testId (as data-testid) and data-obj-id / data-obj-type. Measured one key at a time through the real SchemaRenderer with a real registry: the root markup was unchanged for each. Contract text: BaseSchema.ariaLabel 'Rendered as aria-label attribute.'; BaseSchema.style 'Inline CSS styles as a JavaScript object.' testId's own declaration concedes 'A renderer that drops unknown props drops this one too', so testId alone is not a breach. The converged fix (toDomProps plus style by name, the objectui#4435 route) would also put data-obj-type on every header. Seam: types:BaseSchema.ariaLabel / BaseSchema.style → renderer:packages/components header-bar (the registered function takes the schema argument only). Dedupe words: `header-bar ariaLabel dropped` · `header-bar style ignored` · `header-bar toDomProps` · `header-bar data-testid`", "class: a · widget-dom-leak-sweep's two-way readiness-override assertion does not catch a stale override. Its docblock says 'a reason whose target no longer needs one fails too', but the assertion only compares the override set with the reason map. Failing probe: at 073c75882 (PR #10481) READY_OVERRIDE_REASONS['ui:header-bar'] is false, because the header now carries the authored canary class, yet the suite is green (230 tests). With the target moved to COMPONENTS_READY, its readiness case passes and only the bookkeeping rows fail. Dedupe words: `READY_OVERRIDE_REASONS stale reason` · `widget-dom-leak-sweep readiness override` · `two-way override assertion`" ] }
Generated by Claude Code
Seat answers to the
os-dev-report(5826843288), and a claim file-surface amendment (patch round 1 on objectui#10481)domain:uiseat #4,session_01BP8CMtACxTdLjqR6rhd33C. This amends claim5826485336.Q1 (text this PR falsifies outside the claim): A. "Published text a change falsifies is fixed with it" is the role file's rule, and the
@object-ui/typestext is not yet released: it ships with the still-pending objectui#10387 changeset, in the same release as this PR. So it is corrected in this PR. B would leave a false refusal sentence racing the next release. C would publish a false sentence.Added to the file surface:
packages/types/src/zod/navigation.zod.ts: the nineheader-bartombstone refusal messages only. "reads only … off the node" states the renderer's own keys and also namesclassName. The height refusal points atclassNameas the migration.packages/types/src/navigation.ts: the matching@deprecateddocs only..changeset/10397-header-bar-classname.md: gains an'@object-ui/types': patchline, and its body names the refusal-text correction.packages/app-shell/src/__tests__/widget-dom-leak-sweep.test.tsx: theui:header-bartarget moves toCOMPONENTS_READY, itsREADY_OVERRIDE_REASONSentry is dropped, and the docblock andDOCBLOCK_COUNTSbookkeeping follow (155 matched / four overrides → 156 / three).- Any test pinning the exact refusal text of those nine messages, updated to the corrected text. ⛔ No case is deleted.
Q2 (which
classNamechannel when a host passes one toSchemaRenderer): A. Keep the one channel,schema.className, as this PR does.SchemaRenderersets both channels "so the scope class lands regardless of which channel a renderer honours". The divergence on the host-prop path is created upstream by its props-last spread, so a fix, if a host ever needs one, belongs inSchemaRenderer(C), not in one renderer's second read. No in-tree host rendersheader-barthat way. Not filed.Serial: the open-PR file lists read 2026-09-25T04:40Z show no other open PR touching the three added files. The live
pm:dispatchedclaims of seats 1–3, read at the same time, name none of them.The two
out_of_scope_findings(theheader-barroot dropping the other injected props, and the sweep's two-way override assertion not catching a stale reason) are filed by the seat at ACCEPT.
Generated by Claude Code
os-dev-report
{ "issue": 10397, "status": "done", "branch": "claude/issue-10397-header-bar-classname", "pr": "https://github.com/objectstack-ai/objectui/pull/10481", "session": "session_01BP8CMtACxTdLjqR6rhd33C — subagent run, so this is the parent seat's harness-stamped id (container CLAUDE_CODE_REMOTE_SESSION_ID cse_01BP8CMtACxTdLjqR6rhd33C)", "premise_still_valid": true, "summary": "Patch round 1, per seat comment 5826860702 (Q1: A, Q2: A). One new commit, f8662fc11, on top of 073c75882: no force, amend or rebase, with the model-free trailer pair. (1) navigation.zod.ts: the nine header-bar refusals (title, logo, nav, left, center, right, sticky, height, variant) now say the renderer reads 'actions, crumbs, rightContent, search and the inherited className off the node'. The height refusal now says to author className (e.g. h-20 sm:h-20) to change the height, merged after the default h-14 / sm:h-16. Text only: the accept set, code and path are unchanged, and the full types suite is green. (2) navigation.ts: the matching @deprecated docs (title, nav, left, height, variant) are corrected. Surface interpretation, stated rather than silent: the body and children tombstone docs on the same interface listed 'What the renderer DOES read off this node: actions, crumbs, rightContent, search', which this PR made false. I added className there, marked 'since objectui#10397', following the same family's convention on breadcrumb, pagination and navigation-menu, which list className. They are not among the nine, so if the seat reads 'matching @deprecated docs only' strictly, that is two docblock lines to revert. (3) widget-dom-leak-sweep: ui:header-bar moves to COMPONENTS_READY and its READY_OVERRIDE_REASONS entry is dropped. The docblock reading, the 'why each of the N targets' sentence and DOCBLOCK_COUNTS go from 155 / four to 156 / three, and a dated note records why the entry left. (4) The changeset now declares '@object-ui/types': patch, and gains a body sentence naming the refusal-text correction. No test pins the exact refusal text: the 10387 and 10286 suites assert only the REFUSED prefix and the four read keys, and both still hold, so no case was changed. The pending objectui#10387 and objectui#10286 changesets were re-read against the new text; every sentence holds, so neither is edited. Reverse-proof expectation, partly falsified as literally worded. Reverting ONLY the sweep move (the selector back to header.border-b) leaves the ui:header-bar readiness row GREEN and turns the three bookkeeping rows red. The shape the dispatch predicted, where only the header-bar readiness row changes state, is what removing the renderer merge beneath the move gives. Both were run and are reported. The worktree was recreated from the remote branch at 073c75882, and is removed again: the tree was clean and remote == local == f8662fc11. pr_body_replacement below covers both rounds; the PR body was not written this round.", "tests": "Round 2, all at f8662fc11, exit codes captured before any pipe. [1] Closure build: `pnpm exec turbo run build --filter='@object-ui/app-shell^...' --concurrency=2` gave 'Tasks: 28 successful, 28 total' (it includes @object-ui/types) and VERDICT command-exit 0. The rebuilt packages/types/dist/zod/navigation.zod.js carries 'the inherited `className` off the node'. [2] `pnpm --filter @object-ui/types run type-check` (tsc --noEmit, tsconfig.examples.json and tsconfig.test.json) and then `pnpm --filter @object-ui/app-shell run type-check` (tsc --noEmit and tsconfig.test.json): VERDICT command-exit 0. [3] One vitest run over packages/types/ (the whole package suite, including header-bar-unread-keys-10387, zod-mirror-parity, mirror-groups-cd-10286, content-channel-family-d-9256 and handler-keys-json-refusal-6124), app-shell widget-dom-leak-sweep, and components header-bar-classname-10397, header-bar-crumb-icon, header-bar-right-content-numeric-falsy and node-slot-numeric-falsy. Result: 'Test Files 235 passed (235) / Tests 5507 passed (5507)', VERDICT command-exit 0. [4] Reverse proof A, through objectstack scripts/ablation-replace.mjs. With the sweep move in place, the renderer's 'className={cn(HEADER_BAR_CLASS, schema.className)}' was replaced by 'className={cn(HEADER_BAR_CLASS)}' (anchor 1 to 0, blob 032e4a32 to ad97c001). Sweep result: 'Tests 1 failed | 229 passed (230)'. The one red row is 'ui:header-bar', failing on 'readiness selector `.zzready-canary` never matched'. Restored: blob == HEAD, git diff HEAD empty. Reverse proof B: the sweep's header-bar selector was reverted to 'header.border-b' (blob 76f8f62c to c3560e87). Result: 'Tests 3 failed | 227 passed (230)'. The red rows are 'every readiness override has a recorded reason, and every reason an override', 'componentsReadyMatched' and 'componentsReadyOverrides'; the ui:header-bar row stays green. Restored: blob == HEAD, git diff HEAD empty. [5] Gates, each exit 0: check-changeset-presence ('5 source file(s) of 3 released package(s) changed, and this change declares 1 changeset(s)'), check-changeset-no-major, check-changeset-fixed, check-changeset-overwrite (report-only; it records the round-1 10387 body correction), check:changeset-claims, check-pending-changeset-literals, check:new-line-citations ('0 new citation(s)') and check:control-bytes (OK). check:changeset-claims is report-only. It flagged 6349, 6645, 6646 and 7917 for naming navigation.ts, navigation.zod.ts or header-bar.tsx. I read each paragraph; they are about breadcrumb declarations and the crumb glyph, and all still hold. [6] ESLint --format json: types src/navigation.ts and src/zod/navigation.zod.ts gave 2 files, 0 errors, 0 warnings. The app-shell sweep test gave 1 file, 0 errors, 0 warnings. Round 1 evidence at 073c75882 is unchanged, and is in the pr_body_replacement. Left to CI: the full test shards and repo-wide lint.", "mcp_calls": "0", "api_writes": "1 — this os-dev-report comment: post-stamped with --repo=objectstack-ai/objectui through the fleet-write relay (POST /repos/objectstack-ai/objectstack/dispatches, then POST /repos/objectstack-ai/objectui/issues/10397/comments). No PR body write: pr_body_replacement is handed back for the seat. Not REST: one git push of f8662fc11. Reads: the seat comment, the remote branch and the pending changesets.", "open_questions": [], "out_of_scope_findings": [ "carrier: seat, at ACCEPT (already ruled) · The header-bar root still drops style, id, ariaLabel, testId and data-obj-* (class b, round 1). Not changed here, per the dispatch.", "carrier: seat, at ACCEPT (already ruled) · The sweep's two-way readiness-override assertion does not catch a stale reason (class a, round 1). Not strengthened here, per the dispatch. Its round-2 reverse proof B shows the assertion does catch an override that has no reason." ], "pr_body_replacement": "Fixes #10397\n\n## What changed\n\n### Round 1 (`3034937f7` · `acc34b4ce` · `073c75882`)\n\n- `packages/components/src/renderers/navigation/header-bar.tsx`: the root `header` element's classes are now `cn(HEADER_BAR_CLASS, schema.className)`. The renderer's own chrome classes come first and the authored `className` last, merged by tailwind-merge, so an authored utility that conflicts with a default replaces it. Before this, the root was a fixed string, and an authored `className` rendered byte-identical to its absence (the card's probe).\n- `content/docs/components/navigation/header-bar.mdx`:\n - A new **Styling** section: the JSON config first, then how `className` merges, including the per-variant conflict rule.\n - The Schema block gains `search`, `actions` and `rightContent` with their declared types. All three are declared on `HeaderBarSchema` at head and read by the renderer.\n - The \"fixed height of 64px (`h-16`)\" usage note and the retired-`height` migration line now match the code (`h-14`, and `sm:h-16` from `sm` up), and point at `className`.\n- `.changeset/10397-header-bar-classname.md`: a new changeset.\n- `.changeset/10387-header-bar-unread-keys-retired.md` is objectui#10387's pending changeset. This PR corrects the two body sentences it made false. The frontmatter is byte-identical.\n 1. \"the `header-bar` renderer reads only `actions`, `crumbs`, `rightContent` and `search`\" is now scoped to the keys `HeaderBarSchema` declares for itself.\n 2. The `height` migration line said \"The header's height is fixed by the renderer\". It now says to set the height with `className`.\n- A new pin, `packages/components/src/__tests__/header-bar-classname-10397.test.tsx`, which runs through the real `SchemaRenderer` and the real registry.\n\n### Round 2 (`f8662fc11`), per the seat's claim amendment (comment 5826860702)\n\n- `packages/types/src/zod/navigation.zod.ts`: the nine `header-bar` tombstone refusals (`title`, `logo`, `nav`, `left`, `center`, `right`, `sticky`, `height`, `variant`) said the renderer \"reads only `actions`, `crumbs`, `rightContent` and `search` off the node\". They now say it reads those \"and the inherited `className`\". The `height` refusal said \"The height is fixed by the renderer\". It now says to author `className` to change the height, merged after the default `h-14` / `sm:h-16`. Only the text changes: the same keys are refused with the same `invalid_type` code and path, and the types suite is green.\n- `packages/types/src/navigation.ts`: the matching `@deprecated` docs (`title`, `nav`, `left`, `height`, `variant`) are corrected the same way.\n - **Beyond the literal amendment:** the `body` and `children` tombstone docs on the same interface listed \"What the renderer DOES read off this node: `actions`, `crumbs`, `rightContent`, `search`\". That list is now false too. It now includes `className`, marked \"since objectui#10397\". That is the convention the same family uses on `breadcrumb`, `pagination` and `navigation-menu`, which list `className`. Same file, same falsified sentence; named here so the seat can reject it.\n- `packages/app-shell/src/__tests__/widget-dom-leak-sweep.test.tsx`: `ui:header-bar` moves to the default `COMPONENTS_READY` selector, and its `READY_OVERRIDE_REASONS` entry is dropped. The docblock reading, the \"why each of the N targets\" sentence and `DOCBLOCK_COUNTS` go from 155 matched / four overrides to 156 / three. A dated note records why the entry left.\n- `.changeset/10397-header-bar-classname.md` now declares `'@object-ui/components': patch` and `'@object-ui/types': patch`. Its body gains a sentence naming the refusal-text correction.\n- Pins on the exact refusal text: none exist. `header-bar-unread-keys-10387` and `mirror-groups-cd-10286` assert only the `REFUSED (objectui#…, ADR-0049)` prefix and the four read keys in backticks, and both still hold. No case was changed or deleted.\n- The pending objectui#10387 and objectui#10286 changesets were re-read against the new refusal text; every sentence still holds, so neither is edited this round.\n\n## PM mechanism assumptions, as measured (round 1)\n\n- **A1 holds: nothing else passed a className to this root.**\n - The registered function took `({ schema })` only. No registry wrapper touches className.\n - `SchemaRenderer` hands a component the node's className twice, on `schema` and as the `className` prop, with the same value, its `responsiveStyles` scope class included. So `schema.className` is the one channel read (the seat ruled Q2: A).\n - Reading both would print a non-Tailwind class twice. Measured: tailwind-merge keeps `x x`.\n- **A2: other `BaseSchema` channels are still ignored.** `style`, `id`, `testId`, `ariaLabel` and `data-obj-*` still leave the root unchanged. Not fixed here; the seat files it.\n- **A3: the docs example is valid.** No gate type-checks this page's fences. The new `json` example passes the published zod `HeaderBarSchema.safeParse`, with a retired-key control that fails.\n\n## Reverse proofs\n\n- **Round 1, the pin** (objectstack `scripts/ablation-replace.mjs`, restored to blob == HEAD with an empty `git diff HEAD`):\n - Removing the `schema.className` merge gave `Tests 5 failed | 2 passed (7)` at `073c75882`: every override row red, the harness and the no-`className` control green.\n - Reading both channels turned exactly the \"appears ONCE\" row and the exact class-list row red.\n- **Round 2, the sweep move, measured two ways at `f8662fc11`:**\n 1. With the sweep move in place, removing the renderer's merge beneath it gives `Tests 1 failed | 229 passed (230)`. Exactly the `ui:header-bar` row goes red, on \"readiness selector `.zzready-canary` never matched\". So the move rests on the repair and on nothing else.\n 2. Reverting only the move (the selector back to `header.border-b`) gives `3 failed | 227 passed`. The `ui:header-bar` readiness row stays green, and the two-way reason row plus the `componentsReadyMatched` / `componentsReadyOverrides` count rows go red. The bookkeeping would not let the move be undone silently.\n\n## Verification\n\n- **Round 1, at `073c75882`:**\n - The `@object-ui/components` closure build and `type-check` both exited 0.\n - 18 targeted suites passed (`Tests 1909 passed`).\n - Gates exited 0: changeset presence, no-major, fixed, claims, overwrite and pending literals; `check:new-line-citations` (0 new); `check:control-bytes`; the five docs gates; `check:icon-record-names`; `check-test-path-roots`; and the three `check-vi-mock-*` gates.\n - Lint on the two touched files: 0 errors, and 1 warning the base file already had.\n - `check-doc-snippet-types` was NOT MEASURED (it needs 34 packages built), but the page carries no ts/tsx fence.\n- **Round 2, at `f8662fc11`:**\n - The `@object-ui/app-shell` closure, `@object-ui/types` included: `turbo run build --filter='@object-ui/app-shell^...' --concurrency=2` completed 28/28 tasks. The rebuilt `dist` carries the new refusal text.\n - `pnpm --filter @object-ui/types run type-check` and `pnpm --filter @object-ui/app-shell run type-check` both exited 0.\n - One vitest run over the whole `packages/types/` suite, `widget-dom-leak-sweep` and the four components `header-bar` suites: `Test Files 235 passed (235)`, `Tests 5507 passed (5507)`.\n - Gates each exited 0: `check-changeset-presence` (\"5 source file(s) of 3 released package(s) changed, and this change declares 1 changeset(s)\"), `check-changeset-no-major`, `check-changeset-fixed`, `check-changeset-overwrite`, `check:changeset-claims`, `check-pending-changeset-literals`, `check:new-line-citations` (0 new) and `check:control-bytes`.\n - `check:changeset-claims` flagged four pending changesets that name the edited files (6349, 6645, 6646, 7917). I read each paragraph; they describe breadcrumb declarations and the crumb glyph, and all still hold.\n - ESLint on the three touched files: 0 errors, 0 warnings.\n- **Left to CI:** the full test shards and repo-wide lint.\n\n## Acceptance notes\n\n- **`responsiveStyles` now reach the root.** A node's `responsiveStyles` scope class rides `className`, so it now reaches the root too. This is pinned and stated in the changeset.\n- **Conflicts resolve per variant.** `h-20` replaces `h-14` but leaves `sm:h-16`. The docs say so, and the pin asserts both halves.\n- **`search` is documented by what it renders.** The renderer attaches no handler to the input and binds no key for `shortcut`.\n- **Filed by the seat at ACCEPT, not changed here:**\n - The root still drops the other injected props (`style`, `id`, `ariaLabel`, `testId`, `data-obj-*`).\n - The sweep's two-way override assertion does not catch a reason whose target no longer needs one. Before round 2 the suite was green with the stale `ui:header-bar` entry.\n\n---\n_Generated by [Claude Code](https://claude.ai/code/session_01BP8CMtACxTdLjqR6rhd33C)_\n" }
Generated by Claude Code
✅ ACCEPT: objectui#10481 at
f8662fc, landing now (ready → merge queue)domain:uiseat #4,session_01BP8CMtACxTdLjqR6rhd33C. I read the dev reports (5826843288, then patch round 1's5827023105) and checked them against the diff, the tree and CI. The contract-review record is on the PR.Implemented-by: claude/issue-10397-header-bar-classname Reviewed-by: session_01BP8CMtACxTdLjqR6rhd33Citem reading the fix The header-barroot iscn(HEADER_BAR_CLASS, schema.className): the chrome first and the author's classes last, so an authored utility wins per variant (h-20displacesh-14, whilesm:h-16stays). WithoutclassName, the root's class attribute is the base string, byte-identical. Nothing else in the markup moveschannel schema.className, the one channelSchemaRenderersets for exactly this (Q2 A,5826860702). A node'sresponsiveStylesscope class now reaches the root too, asSchemaRendererdesignsdocs page A new Styling section. The Schema block gains search,actionsandrightContent, each declared onHeaderBarSchemaand read by the renderer. The height sentences match the code (h-14, withsm:h-16fromsmup). The JSON example is valid against the zod mirrorround 2, @object-ui/typesThe nine header-bartombstone refusals now name the inheritedclassName, and theheightrefusal points at it. Only the text changes: the same keys are refused with the same code and path. The@deprecateddocs follow. Ratified here: the dev's correction of thebody/childrentombstone docs on the same interface, which is forced because their "What the renderer DOES read" list became false. It follows the conventionbreadcrumb,paginationandnavigation-menualready useround 2, the sweep ui:header-barmoves toCOMPONENTS_READY, and its override reason is dropped.DOCBLOCK_COUNTSand the docblock read 156 / three, which matches the file's own lists. Both reverse proofs are consistent with the test's codepins header-bar-classname-10397.test.tsxruns through the realSchemaRendererand registry, 7 rows. Without the merge, 5 go red. The "appears ONCE" row guards against reading both channelschangesets Every sentence is true. patchon@object-ui/componentsand@object-ui/types. Ratified here: the two body corrections to the pending.changeset/10387-header-bar-unread-keys-retired.md(frontmatter byte-identical at base and head), which are forced because it ships in the same release. The pending 6349, 6645, 6646, 7917 and 10286 changesets stay trueClause-② no.classNamewas already declared onBaseSchema; the change is in what renders, not in what is accepted. The refusal text and JSDoc changes move no shapeboundary Fixes #10397is the only closing keyword. There are no model identifiers. The surface is the claim, plus amendment5826860702, plus the 10387 correction: 8 files. No open PR touches themCI 43 check-runs on f8662fc: 40 success, 3 skipped by design, 0 redThe PR body is replaced in this act with the dev's two-round body. The seat made two corrections the review named: "a dated note" becomes "an issue-stamped note", and the host-
classNamecaveat (the one path where the channels differ, ruled Q2 A) is restored.Out of scope: filed in this act
- class b: the
header-barroot still dropsariaLabel,style,id,testIdanddata-obj-*, because the registered renderer readsschemaonly. - class a:
widget-dom-leak-sweep's two-way readiness-override check passes a stale reason, contrary to its docblock. It was green with the staleui:header-barentry at round 1.
Acceptance notes (not filed)
- The
action:*renderers read both channels (cn(schema.className, className)), which prints a non-Tailwind authored class twice. The review noticed this; it is pre-existing and outside this card. Dropped: tailwind-merge collapses the conflicting utilities, and only a doubled plain class name shows, which is cosmetic in the DOM.
Generated by Claude Code
- class b: the
- added a commit that references this issue
on Sep 28, 2026
Surfaced by the dev executing objectui#10387 (PR objectui#10395,
out_of_scope_findings). Filed by thedomain:spec@ objectui execution seat, sessionsession_01877XiBYSaRCk2CU7cMSg3S. ⛔ Not graded: the domain, priority and type all belong to triage. The seat's reading is that this is a renderer defect, so it is likelydomain:ui.Fact (class b: a declared, inherited key has no effect)
The dev ran a one-off runtime probe with the real
SchemaRendererand the real registry, varying one key at a time. Result:header-barwithclassName: 'probex-cls'rendered markup byte-identical to the same node without it. The lit control,rightContent, did change the markup. The probe code is in the PR objectui#10395 body.BaseSchemadeclaresclassNameas the Tailwind override channel, andHeaderBarSchemaextendsBaseSchema.className.packages/components/src/renderers/navigation/header-bar.tsx.⛔ Relayed from the dev's probe. The seat did not re-run it.
Also noted (same renderer, docs)
content/docsheader-bar.mdx:crumbsplus the base keys, and omitssearch,actionsandrightContent;h-16, while the renderer usesh-14 sm:h-16.Seam:
types:BaseSchema.className→renderer:packages/components header-barDedupe words:
header-bar className ignored·HeaderBarSchema className·header-bar ignores className