finding(plugin-detail): record:alert's CTA action.label is the third narrow slot in the same local props type — renderer resolves an inline locale map, the type says string

发现于 #4970 的实施(PR 见下)。#4970 的范围裁定点名的是 RecordAlertProps 的 title / body 两处,这是同一个 interface 里第三处同形的槽位,按不扩围单独记档。

机制

packages/plugin-detail/src/renderers/record-alert.tsx 的未导出局部 RecordAlertProps 里,CTA 那一支两处都写作:

action?: { actionName: string; label?: string; variant?: string };

(properties.* 一份 + flat 兼容一份,与 title / body 完全并列。)

而同一个文件的读点是:

const ctaLabel = pickLocalized(props.action?.label, language);

即渲染器对 CTA 文案做的是内联 locale 映射解析,和它上面两行对 title / body 做的事逐字相同 —— 只有类型声明说这里只能是字符串。#4970 已经把 title / body 放宽成 string | I18nLabel(绑定 spec 符号),这一处留在原样。

今天不咬人(observation-class 的依据)

interface 没有导出(只做 React.FC< RecordAlertProps > 的参数),所以没有任何消费者读到这个窄声明;
pickLocalized 收 unknown,所以 TS 不报错、运行时也照样解析映射;
作者面(manifest / sdui-intrinsics.d.ts)根本没有声明这个成员的类型:plugin-detail/src/index.tsx:569 是 { name: 'action', type: 'object', … },成员形状只写在 description 的散文里({ actionName, label?, variant? })。

这也是它与 title / body 的判据差别,值得单独记而不是并进 #4970:那两个键的第二条臂有已落地的发布面声明(['string', 'object'],#3832 / PR #4975)可以对齐,所以「声明比自己的发布面窄」是个可指认的矛盾;action.label 的发布面对成员类型什么都没说,窄声明的唯一对照物是渲染器本身。要把这一处说成契约不一致,得先有成员形状可声明 —— 那是 ComponentInput 的成员形状缺口(PR #3795 的 open question),不是这一行能解决的。

修复面(尺度最小的一档)

两行:label?: string | I18nLabel(I18nLabel 已在同文件 import,#4970 引入)。如果同时想让发布面能表达这个事实,则要等成员形状那张卡。

关联:#4970(title / body 两处 + elements.tsx 两个标本)、#3832 / PR #4975(机制 + 五标本)、#4581(声明窄于渲染器的同族,已关闭,范围是 BaseSchema)、PR #3795(ComponentInput 无成员形状,本卡发布面那一半的前提)。

docs(plugin-detail): name the CTA label slot this pass deliberately l…

Triage first-touch — concentrated round #2 (triage seat, session session_019ZyCHcjXeTFPiS8RWRk4nN, 2026-08-19): promoted finding → pm:queue · Task. Two-line fix (label?: string | I18nLabel in both copies of the local RecordAlertProps CTA slot), aligning the declaration with what the renderer already does (pickLocalized) — the same move #4970 landed for title/body in the same interface. The publish-surface half (declaring the member shape in the manifest) stays out of scope pending the ComponentInput member-shape question (#3795), exactly as the body reasons. Size/model suggestion: S-mechanical, sonnet.

Generated by Claude Code

Claim: PM loop round 3 (batch raised to 5 at the maintainer's instruction — 「任务很多，并发保持5」)
Session: session_012u2pRjcqAYtoEjgr3wwhnK
Branch: claude/issue-4998-record-alert-label-i18n
Worktree: objectui-issue-4998
Domain: domain:ui
File surface: packages/plugin-detail/src/renderers/record-alert.tsx and the second copy of the local RecordAlertProps CTA slot wherever it lives in packages/plugin-detail/** (+ tests, + a changeset) (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: sonnet — the floor, and this is the card that earns it: triage measured it as S-mechanical with a landed precedent (#4970 made the identical move for title/body in the same interface), the change is a type widening the renderer's runtime behaviour already implements, and correctness is decided by type-check rather than by judgment. The one search task — finding both copies, not one — is bounded and gate-checked.
Clause-②: no — widening a local props type to match what the renderer already does aligns a declaration with existing behaviour; it neither changes the accepted set nor widens the published surface. ⚠️ The publish-surface half (declaring the member shape in the manifest) is deliberately out of scope, parked on the ComponentInput member-shape question in #3795 — do not pull it in.
Serial constraints cleared: none — packages/plugin-detail/** has no in-flight claim in this lane and no PR this session touched it. Disjoint from round 3's other four (apps/console/**, packages/app-shell/src/views/studio-design/**, packages/components/src/renderers/layout/grid.tsx, packages/core/**).

Located on origin/main this round: packages/plugin-detail/src/renderers/record-alert.tsx with __tests__/record-alert.test.tsx and __tests__/record-alert.visibleWhen.evidence.test.tsx beside it.

Generated by Claude Code

fix(plugin-detail): record:alert CTA action.label accepts the inline …

os-dev-report
{
"issue": 4998,
"status": "done",
"branch": "claude/issue-4998-record-alert-label-i18n",
"pr": "#5572",
"premise_still_valid": true,
"summary": "Widened the local, unexported RecordAlertProps CTA slot's action.label from string to string | I18nLabel in both copies in packages/plugin-detail/src/renderers/record-alert.tsx (schema.properties.action.label line 98, flat-compat schema.action.label line 108), matching the pickLocalized(props.action?.label, language) resolution the renderer already performs at runtime. This is the same move #4970 landed for title/body in the same interface. Added a test asserting the case the type previously refused (a locale-map action.label rendering the resolved string via a real I18nProvider); the publish-surface manifest half stays out of scope per the ruling, parked on #3795.",
"tests": "All at commit dc14b64. pnpm --filter @object-ui/plugin-detail type-check: exit 0 (tsc --noEmit && tsc -p tsconfig.test.json, 0 errors). pnpm exec vitest run packages/plugin-detail/src/renderers/tests/record-alert.test.tsx: 21/21 passed (1 test file). pnpm exec vitest run packages/plugin-detail/: 90 test files / 848 tests passed. pnpm --filter @object-ui/plugin-detail lint: exit 0, 0 errors (809 pre-existing no-explicit-any/react-hooks warnings, none on touched lines, confirmed by grep against changed files). node scripts/check-control-bytes.mjs: OK, 4629 tracked text files scanned. node scripts/check-changeset-presence.mjs: 2 source files changed under src/, 1 changeset declared -> pass. node scripts/check-changeset-no-major.mjs: pass, no major bump. node scripts/check-changeset-fixed.mjs: pass, all workspace packages in fixed group. Reverse verification: reverted only the type declaration in record-alert.tsx to the pre-fix label?: string (source-only, confirmed on disk via grep), reran type-check -> exit 2, tsc reported TS2322 at exactly the new test's locale-map literal (record-alert.test.tsx:219:17, "Type '{ en: string; zh-CN: string; }' is not assignable to type 'string'"); the pre-existing plain-string CTA test stayed unaffected. Restored the fix via git checkout from the committed branch state (git status clean afterward), reran type-check -> exit 0 again. This confirms the new locale-map test -- not the pre-existing plain-string CTA test -- is the case the narrow declaration refused.",
"open_questions": [],
"out_of_scope_findings": []
}

Generated by Claude Code

PM · review of PR #5572 — the ablation is the right one for a type-level fix

Most reverse verifications this session mutated behaviour and watched tests go red. That would have proved nothing here, because the renderer already resolved locale maps at runtime — pickLocalized takes unknown, so every vitest assertion is green with the narrow declaration too. You ablated the thing that actually carries the fix:

reverted only the type declaration to label?: string (source-only, confirmed on disk via grep), reran type-check → exit 2, TS2322 at exactly the new test's locale-map literal (record-alert.test.tsx:219:17, "Type '{ en: string; zh-CN: string; }' is not assignable to type 'string'"); the pre-existing plain-string CTA test stayed unaffected. Restored, re-ran → exit 0.

That establishes two separate things, and the second is the one reviewers usually skip: not just that the pin fails without the fix, but which test carries the discrimination — the new locale-map case, not the plain-string case sitting three lines above it. I had flagged exactly this doubt on my own read before your report arrived; you had already closed it.

Checked independently, not taken from the report

Both copies widened — schema.properties.action.label (:98) and the flat compat mirror schema.action.label (:108). Finding one and shipping was this card's failure mode; triage said "both copies" for that reason.
The stale docblock was rewritten, not left to rot. It previously read "deliberately NOT widened here (objectui#4998)" — a sentence that would have become false the moment this merged. It now states the widening, why (pickLocalized parity with title/body), and that it is type-only.
The out-of-scope half stayed out. The published surface still declares action as a bare object with the member shape in prose, so there is no manifest arm to align; parked on fix(plugin-detail): record:highlights 的 fields 声明补上 readonly,让 manifest 能被作者读到 (#3407) #3795 and said so in both the code comment and the changeset, which is where a future reader will actually look.
Unprompted care worth naming: wrapping the new case in a real I18nProvider pinned to en so it does not ride whatever language another test file left on the global i18next singleton. Nothing asked for that; a locale test that silently depends on file execution order is a flake waiting for a busy CI runner.

This was dispatched at the sonnet floor — triage measured it S-mechanical with a landed precedent (#4970) and correctness decided by type-check rather than judgment. That call holds: the floor was the right tier, and the work came back with a better ablation than the tier required.

Status

Path surface (get_files): 3 files, all packages/plugin-detail/** plus the changeset. No governed surface → ordinary ready → merge-queue path.

CI on #5572: Build & E2E, Build Docs and all three changeset gates green; the four Test shards and Type Check still in_progress, nothing red. Type Check is the gate that decides this card, so no verdict and no enqueue until it reads completed: success by name.

Generated by Claude Code

ACCEPT — PR #5572

CI converged. Read by name: Type Check (the gate that decides this card), Test (shard 1/4) (2/4) (3/4) (4/4), Lint, Build & E2E, Bundle Analysis, Control Byte Scan, all three changeset gates, both doc type checks, Internal Docs Link Check, Skill Guide Path Check, Build Docs, Live E2E → every one completed: success. 19 success / 3 skipped / 0 failures / 0 in progress.

Path surface (get_files): 3 files, all packages/plugin-detail/** plus the changeset. No governed surface → ordinary ready → merge-queue path.

Substantive review is in the comment above and does not depend on this run.

Incidental corroboration of the "type-only" claim: plugin-detail reads 60.89 KB gzipped in Bundle Analysis — byte-identical to the pre-change figure, which is exactly what a type widening should produce.

Driving to landing.

Generated by Claude Code

fix(components,plugin-detail): element:text.content / element:button.…

fix(plugin-detail): record:alert CTA action.label accepts the inline …

fix(plugin-detail): record:alert resolves its CTA through resolveDecl…

Navigation Menu

Repository navigation

finding(plugin-detail): record:alert's CTA `action.label` is the third narrow slot in the same local props type — renderer resolves an inline locale map, the type says `string` #4998

机制

今天不咬人(observation-class 的依据)

修复面(尺度最小的一档)

PM · review of PR #5572 — the ablation is the right one for a type-level fix

Checked independently, not taken from the report

Status

ACCEPT — PR #5572

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(plugin-detail): record:alert's CTA action.label is the third narrow slot in the same local props type — renderer resolves an inline locale map, the type says string #4998

Description

机制

今天不咬人(observation-class 的依据)

修复面(尺度最小的一档)

Activity

os-support-ai commented on Aug 19, 2026

os-sales commented on Aug 21, 2026

os-sales commented on Aug 21, 2026

os-sales commented on Aug 21, 2026

PM · review of PR #5572 — the ablation is the right one for a type-level fix

Checked independently, not taken from the report

Status

os-sales commented on Aug 21, 2026

ACCEPT — PR #5572

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(plugin-detail): record:alert's CTA `action.label` is the third narrow slot in the same local props type — renderer resolves an inline locale map, the type says `string` #4998