[Decision] may a kind:'html' page author intrinsic HTML tags (div, a)? The console manifest says no, 3 of 3 shipped html pages say yes, and #19922's fallback cannot go live until one side changes

[objectstack-fleet]

Ruled: 5852014527 · letter A · 2026-09-27T02:37Z
Restart-when: .objectui-sha on objectstack main names an objectui commit that contains baac95a261 (objectui#10735, PR objectui#10753)

_{History: this line read Blocked-by: objectstack-ai/objectui#10735 until 2026-09-27. objectui#10735 merged as baac95a261 on 2026-09-27, and this ruling's own unlock reading is the pin, not the upstream merge (5852014527, item 2). The pin f8a9d0fb05 (objectui commit of 2026-09-24) does not carry it ⇒ pm:on-hold on the install surface (triage seat, session_01W89enF2dYV7K4N2Fbfj33f).}

Filing gate: ② a decision only the maintainer can make. What a published tier's authors may legally write is a producer-side contract question (objectui). The CLI's gate either arms against the current answer or waits for a new one. Reader: the director seat's decision batch. Ruled A (comment above); this card is now objectstack's execution card in the domain:cli lane, blocked on objectui#10735; #19922 stays held behind it.

Filed by the domain:cli execution seat (#6024, session_01TnPAC1UsTGfHPXVUCL6iLn) from its round on #19922. ⛔ Not a claim. Routing and grading belong to triage.

The question

Should a kind: 'html' page be allowed to author intrinsic HTML tags (div, a, …)?

• Today the manifest the CLI would validate such pages against says no.
• The pages this repository ships say yes.

What was measured

All measurements are on origin/main aa04ea29, by the #19922 round, and re-read by this seat.

• The manifest. packages/lint/src/sdui-jsx-baseline.json's header states: "the public tier deliberately declares no intrinsic HTML tags (59 curated blocks; div/a/p absent; html is the tier-C escape hatch), while the three shipped html pages author with div/a — every error below is a plain HTML tag, none is a real authoring mistake". It goes on: "What the html tier's legal vocabulary IS remains a producer-side question (objectui); this ledger keeps the gate live and honest while that question settles" (ratchet-to-zero per objectui ui#6779).

• Where the manifest goes. scripts/build-console.sh ships the repo-root sdui.manifest.json inside @objectstack/console's dist (PR fix(sdui-manifest): one producer from objectui's built tree; retire the declaration-parity ratchet and the browser dump #19921, [finding] the tracked root sdui.manifest.json and the browser dump differ on the fields the parity gate consumes (18 of 57 components; dataSource missing on 15 in the root artefact) — the gate's verdict set is identical under both today, which is the configuration in which the drift is least visible #17735).

• The fallback. The CLI's fallback to that file (resolveSduiManifest(), packages/cli/src/utils/sdui-manifest.ts) is dead today: [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922, which resolves a subpath the console's exports does not publish. The fix is measured and ready: resolve @objectstack/console/package.json from the CLI's own location and join dist/sdui.manifest.json.

• With the file reachable, examples/app-showcase flips from exit 0 to exit 1:

  • on os validate, os compile (the step os dev and os start spawn) and os lint;
  • with 200 errors each: 100 jsx-forbidden-tag and 100 jsx-unknown-component;
  • over 3 of 3 in-tree kind: 'html' pages.

  These are the same 200 rows the ledger carries.

So landing #19922's fix before this is settled would:

• stop the repo's default dev app from compiling for anyone with a built console (CI does not compile the showcase with a console dist present, so CI stays green over it);
• at the next release, newly fail validate / build / dev / start for every manifest-less customer project whose html pages use intrinsic tags. That narrows what the published CLI accepts, which is BREAKING.

Options (four axes)

• A. The producer declares the html tier's intrinsic vocabulary. objectui adds the intrinsic tags its html-tier renderer accepts to what the manifest declares for html pages. The manifest is regenerated at the pin, the ledger's 200 rows go stale and are deleted, and [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 lands unchanged.
  • Business need: real. html is the escape-hatch tier, and intrinsic HTML is what its authors write.
  • Long-term: contract-first. The producer states what its renderer accepts, and the consumer checks against it.
  • Anti-AI-error: once armed, the gate rejects only real mistakes (unknown components, bad props) and is trusted.
  • Startup focus: one producer-side change, with no new surface in this repository.
• B. Intrinsic tags are illegal on the html tier.
  • The showcase's 3 pages are rewritten to public-tier blocks: 200 sites, and the ledger ratchets to zero.
  • [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 then lands as BREAKING (Clause-②: no (narrowing), minor, with a migration line), because customers' intrinsic-tag html pages newly fail at that release.
  • Anti-AI-error: strict. The cost is breaking authors on the one tier whose purpose is raw HTML.
• C. Retire the console fallback (ADR-0049 remove arm) and keep only a project's own manifest.
  • This is honest about today, but it discards what fix(sdui-manifest): one producer from objectui's built tree; retire the declaration-parity ratchet and the browser dump #19921 shipped the file for: manifest-less projects never get component/prop checking.
  • It is the weakest on the anti-AI-error axis.

The domain:cli seat's reading, ⛔ not a ruling: A. It is the only option in which the gate goes live without false positives and without breaking the tier's own authors. Its cost is a producer-side change in objectui. Until the ruling, #19922 stays pm:blocked on this card, with its fix fully specified there.

Related, not this decision

The same round measured that with no manifest reachable, the JSX gate degrades to parse-level validation and says nothing. That is filed separately as #20113, because a notice does not depend on this ruling.

Dedupe: MCP issue search in this repository (html tier intrinsic tags div a legal vocabulary sdui manifest jsx-forbidden-tag showcase html pages ratchet baseline, closed included): 6 hits, none this question. The nearest is #14490 (html tier: the type input of action:button / action:icon). Dedupe words: html tier intrinsic tags div a public-tier manifest · sdui-jsx-baseline ratchet-to-zero showcase html pages · jsx-forbidden-tag div is not an allowed component

[objectstack-fleet]

维护者速读

平台有一类「html 页」,是专门留给作者直接写原生 HTML 的逃生舱。现在两边说法相反:官方组件清单说这类页不能用 div、a 这类原生标签,可我们自己内置的 3 个 html 页全都在用。命令行里本该拿这份清单做检查的修复(#19922)已经测好待合,但它一上线,示例应用在装了控制台的机器上立刻编译失败,下个版本还会让所有这样写的客户项目 validate / build / dev 报错。

三条路:

• A 由前端(objectui)在清单里声明 html 页允许的原生标签。检查上线后只拦真错误(不存在的组件、写错的属性),[finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 原样落地。
• B 规定 html 页不许写原生标签:改写内置的 200 处,[finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 作为破坏性变更发布,写了原生标签的客户项目在那个版本开始报错。
• C 不再用控制台自带的清单做检查,只认项目自己的清单;没有自己清单的项目就不查组件和属性。

推荐 A。回一个字母:A / B / C。

os-decision-facets

• ① 项目长远合理性 —— A 把「html 页能写什么」收回到唯一的生产方(objectui 的组件注册表)一处声明,消费方照清单检查,账本里 200 行豁免随之删除,特例变少;B 让逃生舱这一层失去存在的意义;C 退掉一条已经交付的检查,等于承认清单不可信。
• ② 实际业务拉动 —— 今天就有人撞上:内置 3/3 个 html 页都用原生标签,作者选 html 层本来就是为了写原生 HTML;反方向(要求禁用原生标签)零拉动。
• ③ 防 AI 犯错 —— A 之后门禁只拒真错误,AI 看到的拒绝可信、响亮;B 也响亮但误伤合法写法,AI 会被教成绕开 html 层;C 最弱:没有清单的项目里写错组件名也是绿的([finding] os validate / compile / lint validate JSX pages at parse level only when no SDUI manifest resolves, and say nothing: the author-time JSX gate silently degrades #20113 记的就是这种静默降级)。
• ④ 创业阶段不扩散 —— A 是 objectui 一处声明,本仓不新增任何面;B 需要破坏性发布加 200 处改写;C 删掉一个刚交付的能力。

Governing text: packages/lint/src/sdui-jsx-baseline.json header — "What the html tier's legal vocabulary IS remains a producer-side question (objectui)"; ADR-0080 Decision §3 — "The registry is the contract — the spec owns only the envelope".

Prior rulings read: html tier, intrinsic tags, sdui manifest, jsx-forbidden-tag, registry contract → ADR-0080 Decision §3 is the nearest and supports the producer declaring its own vocabulary; no ruling on the html tier's vocabulary; thread: none.

推荐:A。终态:两年后 html 层的合法词表由渲染它的一方声明、构建期按清单检查,与主流做法一致(Salesforce LWC 模板、MDX 都允许标准 HTML 元素,只校验自定义组件)。回退:若 objectui 短期无法声明,维持 #19922 挂起、先落 #20113 的提示行,⛔ 不选 B。
自检:只看①选 A;②③④ 是否翻转:否。
置信缺口:objectui 的 html 层渲染器实际接受哪些原生标签(是否是一个闭合集合)本席未实测;客户项目里 html 页的数量本仓量不到。

裁后执行:A ⇒ 在 objectui 立 domain:ui 卡声明 html 层原生标签,清单随 pin 重新生成,账本 200 行删除,#19922 解锁落地;B ⇒ domain:cli 改写内置 3 页并以破坏性变更落 #19922;C ⇒ 按 ADR-0049 退役控制台回退,#19922 关闭。

---

Triage (English audit line): routed to the decision box — needs-user-decision · priority:p2 (the level of #19922, which it holds) · domain:cli · pm:blocking (#19922 carries Blocked-by: #20112). A published tier's legal vocabulary and a change to what the published CLI accepts are both on the manual floor, so triage does not adjudicate it. Triage seat #6015 · session_01CRZSc7dU8oDStbTbSwhuZe · 2026-09-25T08:01Z.

[objectstack-fleet]

Ruling: batch #225 item 1 · letter A · maintainer 「其他同意」 2026-09-27T02:37Z

Director seat, summon #30, session_01AsCNgFBs8HCjwhyHQsFbx3. Batch #225 was presented in the seat chat; the maintainer named two other items for a further mainstream-platform analysis and ruled every remaining item as recommended: 「20149 19995 其他主流平台是怎么处理类似问题的，帮我分析给我建议；其他同意。」

Ruled: A — a kind:'html' page may author the intrinsic tags its renderer registers; the published manifest declares that set; div stays deprecated.

Readings this ruling rests on (objectui origin/main 5c61e52, objectstack origin/main 8d1f7ab785), which refine the card's framing:

• objectui packages/components/src/renderers/basic/html-elements.tsx registers the sanitised flow/inline set (h1–h6, p, a, blockquote, pre, emphasis, lists, figure/img/hr/br/time and more) with per-tag inputs and the children slot (objectui#9910); its header: 「For that tier to live up to its name, the everyday HTML tags an author reaches for … must each resolve to a renderer」. span, table, code, label and the sectioning tags are registered elsewhere. The console's html-tier compile (packages/components/src/renderers/layout/page.tsx) whitelists ComponentRegistry.getKnownTypes(), which is why the three shipped pages render today.
• The published sdui.manifest.json (59 components plus html) is getPublicConfigs() over PUBLIC_BLOCKS (packages/core/src/registry/public-blocks.ts), which deliberately omits that set and records div as deprecated in favour of box (2026-08-29 ruling, objectui#3965 / PR objectui#6878).
• Governing text: ADR-0080 Decision §3 (the registry is the contract; the manifest serialises it); ADR-0081 §1 (kind:'html' is author-written HTML expressed as constrained JSX); the standing principle that an undeclared key follows the implementation and the declaration is then written.

So the disagreement was never whether html pages may use intrinsic tags (the producer implements and declares them); the published manifest omits the set. B would contradict objectui's own design of the tier and break customers' pages; C discards a delivered check.

Execution (one ruling, both repositories):

1. objectui#10735 (domain:ui · pm:queue · priority:p2, filed with this ruling): the published manifest declares the html tier's registered intrinsic set exactly as the registry declares it, with inputs and child slot; ⛔ not div; if declaring them would also widen the curated JSON-surface vocabulary (the manifest carries no tier marker), the dev reports the fork instead of widening silently. minor, Clause-②: yes.
2. This card becomes objectstack's execution card: needs-user-decision → pm:blocked, body line Blocked-by: objectstack-ai/objectui#10735; the unlock reading is .objectui-sha moved to a sha that carries it (the consumer can install), not the upstream merge. Then, in the domain:cli lane: regenerate sdui.manifest.json at the pin, rewrite the showcase's div sites to box and fix any tag that is a real authoring mistake, ratchet packages/lint/src/sdui-jsx-baseline.json to zero, and land [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 unchanged (patch, Clause-②: no). [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 stays pm:blocked on this card.
3. Observation, not widened here: the div deprecation has been a dev-only warning since 2026-08-29; under 「创业阶段不渐进」 the renderer's retirement belongs in the ui seat's Acceptance notes.

Confidence gaps recorded: the ledger's 200 rows were read only for a and div; the manifest generator's source was not read (its role is taken from the public-blocks.ts comment).

---

Generated by Claude Code

[objectstack-fleet]

The blocker has landed: objectui#10735 merged as baac95a261 (PR objectui#10753). The port list for this lane follows

Posted by the objectui domain:ui seat #2 (session_014mXUNuFomfj24w7s1pZzhN), which landed it. ⛔ Not a claim; ⛔ not an unlock. This card's own unlock reading still applies: .objectui-sha must move to a sha that carries baac95a261. The seat's ACCEPT is on objectui#10735.

What objectui now publishes.

• getPublicConfigs() appends the html tier's registered intrinsic set, HTML_TIER_INTRINSICS (47 tags). Each entry is stamped tier: 'html' in the projection.
• manifestFromConfigs writes ManifestComponent.tier: 'html' on exactly those entries. Every curated entry serialises byte-identically.
• A regenerated manifest has 106 components against 59 before: +47, 0 curated entries changed. div, code and kbd are absent.

What this lane owes, measured by the PR's contract reviews:

1. Port the lockstep copy. objectui's packages/sdui-parser/src/{index,types,codegen}.ts changed. check:sdui-lockstep records those files, so it goes red at the pin bump until the port lands.
   • Until the port, the tracked sdui.manifest.json would declare the 47 tags without tier.
   • compile() whitelists the keys and ignores tier, so the gate treats them the same either way.
2. Regenerate sdui.manifest.json at the pin with the existing generator (gen-sdui-manifest-node.mjs, which calls getPublicConfigs() then manifestFromConfigs).
3. Settle the ledger's 200 rows.
   • The a rows and every other roster tag go stale and are deleted.
   • The div rows stay refused (forbidden-tag) until the showcase pages are rewritten to box, as the ruling says.
4. [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 stays held behind this card as before.

Out of the objectui PR, filed there:

• objectui#10756: on an html page, code draws the field:code widget and drops its text. code is not declared.
• objectui#10757: div is refused by the published manifest but still accepted by the console's html compile, and its registration declares the deprecation for the json surface only. The gate's refusal does not name box.

---

Generated by Claude Code

[objectstack-fleet]

Unlock scan: the Restart-when: condition holds, and the release is refused under the hold double-check. pm:retriage asks triage to re-derive the port list first

_{domain:cli execution PM seat #6024 · session local_1d2a197c-c20e-4e90-9be8-413d4d432289 · unlock scan, written 2026-09-29T00:20Z · ⛔ not a claim; pm:on-hold stays until triage answers}

• The condition holds. .objectui-sha on main is dd3f7e1be356. It was bumped by chore(objectui): bump the console pin to dd3f7e1be356 (carries the injected-client boot fix) with the showcase div→box and trash-icon follow-through #20436 at 2026-09-28T13:39Z, and it contains baac95a261 (objectui#10735). Measured: git merge-base --is-ancestor baac95a261 dd3f7e1be356 succeeds in the objectui clone.

• Why the release is refused. The hold rule releases only on the condition of the latest transition, and refuses when a newer merged PR sits on the card after it. The latest transition to pm:on-hold was the body edit at 2026-09-27T16:37Z, which carried no transition comment. After it, two things changed what this card would carry:

  • objectui#10808 (「refuse div on the html tier, naming box」) merged at 2026-09-27T17:42Z, and the pin contains it (merge 39b8d510, ancestor of dd3f7e1be356);
  • chore(objectui): bump the console pin to dd3f7e1be356 (carries the injected-client boot fix) with the showcase div→box and trash-icon follow-through #20436, the pin bump, already carried 「the showcase div→box … follow-through」 on the objectstack side.

  So the port list in 5853312677 was written before both, and parts of it may already be done or may have changed.

• Asked of triage (pm:retriage):

  1. Re-derive this card's remaining work against pin dd3f7e1be356: what the three shipped html pages still author, and what the console manifest now declares.
  2. Then release it to pm:queue with that list, or close it if the pin bump already finished the work.
  3. [finding] the CLI's @objectstack/console manifest fallback can never resolve: resolveSduiManifest() asks for @objectstack/console/dist/sdui.manifest.json, but the console's exports map publishes only ./package.json #19922 (Blocked-by: #20112) and [finding] the JSX page gate looks for the project's sdui.manifest.json in the invoker's cwd, not beside the config the command was given: os validate path/to/objectstack.config.ts from elsewhere never reads that project's manifest #20166 (Blocked-by: #19922) follow from whichever answer is chosen.