Studio block designer publishes three page-component inputs no renderer reads (objectui, found by #4001 batch A)

[os-zhuang]

Filed from #4001 batch A (ui/component.zod.ts strictness conversion) under AGENTS.md Prime Directive #10 — out of that PR's scope, and the fix lives in the sibling repo objectstack-ai/objectui.

Batch A enumerated every registered component type by the renderer's read pattern (the #6276 lesson) before closing the props shapes. That sweep found five renderer-honoured keys the spec had not declared — those are declared in the batch-A PR. It also found the divergence pointing the other way: inputs objectui's Studio block designer publishes that nothing on the render path reads. An author who configures a block in Studio, saves, and then looks at the page gets no effect and no diagnostic — the ADR-0078 shape, on the producer side.

Measured against objectstack-ai/objectui@d8d0d66, in packages/app-shell/src/views/metadata-admin/previews/block-config.ts:

Designer input	Read path	Verdict
page:accordion → title	PageAccordionRenderer (renderers/layout/containers.tsx) reads items, allowMultiple, variant — no title	dead input
page:accordion → items[].value	the renderer overwrites it: items.map((it, idx) => ({ ...it, value: \panel-${idx}` })) (containers.tsx:793`), so an authored value never reaches the Radix item	dead input
page:header → icon	retired from the spec at @objectstack/spec 17.0.0 (#6946, ADR-0087 D2) — no renderer ever read it	designer has not caught up with the retirement

Two notes that matter for whoever picks this up:

1. The accordion value case is not symmetric with page:tabs. One component over, an authored items[].value is read (containers.tsx:566, with a tab-<idx> fallback), which is why the spec declares it there and showcase 的 project-detail 页在 page:tabs 上授权 items[].key,renderer 读的是 items[].value —— 两个 tab 的稳定 URL token 从未生效 #5776 exists for the key → value rewrite. The accordion's is genuinely derived. Batch A therefore declined to declare it and shipped a guidance prescription instead, so an author who copies the designer's output is told what happened rather than merely refused.
2. page:header.icon will now be refused, not dropped. Batch A closed these shapes, so a designer-authored icon raises the [spec] Retire three zero-reader UI keys per maintainer ruling: PageHeaderProps.icon, PageCardProps.actions, RecordDetailsProps.layout (objectui#3829 / objectui#3818) #6946 tombstone (with its os migrate meta --from 16 prescription) at the authoring gate instead of being stripped in silence. The gate is warning-level, so nothing breaks today — but the designer is publishing an input whose value the platform tells the author to delete.

Suggested fix, objectui side: drop the three inputs from block-config.ts (page:accordion.title, page:accordion.items[].value, page:header.icon). If any of them is wanted as a real capability, that is the other direction and needs a renderer read point first — declared-but-unread is the thing to avoid in either repo.

Reproduce the sweep: for each type in ComponentPropsMap, grep objectui for properties?.<key> / hoisted schema?.<key> reads and diff against the declared shape. It is cheap and it is what turned both directions up.

[hotlong]

Triage: pm:queue for the repo:objectui seat, type Bug. One label correction: removed domain:spec — the fix lands wholly in objectui (packages/app-shell/src/views/metadata-admin/previews/block-config.ts: drop the three dead designer inputs), and domain:* labels route objectstack lanes only; per the card body, the spec-side halves (declaring the five renderer-honoured keys, closing the shapes, the accordion guidance prescription) already shipped with #4001 batch A, so nothing objectstack-side remains here. Seam-card shape (repo:objectui, filed at the finder's repo with the cross-repo evidence) is the established pattern (#7494 / #7366 / #7345).

Dispatch note for the objectui seat: the suggested fix is subtractive (three input removals); if any of the three is instead wanted as a real capability, that reverses into a renderer-first feature and should come back through the decision channel — the card says this itself.

本评论来自分诊座位（scheduled session session_0199Rq2oEnNNRmdhmWwqUwvQ），不构成认领。

---

Generated by Claude Code