refactor(types,components,sdui-parser): retire the body child-list dialect — children is the one spelling (objectui#6771) (#9895)

Fixes #6771

> ## ⚠️ CORRECTION BLOCK — what this body gets wrong at head
`b37fefa017cb7ff09456fc5de19ae45d93ebfc18`
>
> Re-stamped by the `domain:ui` seat at 2026-09-19T00:18Z, replacing the
block written at 2026-09-18T19:49Z for head `25e0130eb…`. The head has
moved **three times** since (`059a41c65…` → `f50fccc2f…` → this one), so
that block's own corrected figures had themselves gone stale — which is
the failure mode it existed to prevent, now measured on itself.
>
> ⛔ **This block is NOT claimed to be exhaustive.** The block it
replaces said it was 「exhaustive by construction」 and a review falsified
that with one unlisted sentence. A list of what a reader has checked is
honest; a guarantee that nothing else is wrong is ⛔ not one this seat
can make about a 106-path diff.
>
> **Contract-review records on this PR: five**, each fetched by id with
its `issue_url` read at 2026-09-19T00:18Z — `5732059417` (head
`9547e08ee`), `5733750066` (`f4b109034`), `5734518119` (`2e729bc6f`),
`5735358922` (`25e0130eb`), `5736698949` (`059a41c65`). All five are
**FAIL**. ⚠️ The fifth's **verdict is VOIDED** — `5736705386` records
why: the reviewer ran below `CONTRACT_REVIEW_TIER` because the seat
spawned it without passing the tier. ⭐ Its **measurements were not
withdrawn**; only its authority to issue a verdict was. A sixth review,
at tier and verified at spawn, is running against this head.
>
> **Figures in this body that are stale or false at this head.** ⚠️
Where a number moved, this block names the **instrument** rather than
printing a replacement — a count copied into prose is what `AGENTS.md`
#9 forbids, and re-copying one here would only restart the clock on the
next head:
> - Verification's `221 / 5009` and `282 / 2743` — bound to earlier
heads. Run the affected-package suites under the verify lock; ⛔ do not
read these.
> - Verification's 「81 paths」, and the previous block's replacement
figure 「**102**」 — **both stale**. The real path surface, read by this
seat from `GET /pulls/9895/files` across all pages at 2026-09-19T00:18Z,
is **106**, ⛔ not governed (0 of `docs/adr/**`, `.claude/**`,
`skills/**`, `AGENTS.md`, `CLAUDE.md`), with a present-path control
firing and a fabricated-path control reading absent.
> - 「159 nodes carrying `body` … before, 38 after」 — bound to an earlier
head. `pnpm census:body-dialect` is the instrument and it ships in this
PR; ⛔ read it, not this sentence.
> - The previous block's 「the instrument reads **27** on this head」 —
that was a reading of `25e0130eb…`, ⛔ not of this head, and it is not
re-taken here.
> - 「About twenty per-component docblocks」 — measured at **137 across 12
files**, which is why it has its own card, objectui#9933.
> - 「Twelve registrations read `body` and nothing else」 — thirteen,
agreeing with the changeset and with this body's own next paragraph.
>
> ⚠️ **Most of these are this seat's own**, not the dev's: the block
this replaces was the seat's, the 「26」/「27」 census figure was the
seat's, and the voided review was the seat's error in spawning. The dev
of round 7 ⛔ did not touch this body — the standing clause says the
implementer writes it once, and this round's dispatch did not name it as
an owed item, so carrying it is the seat's job. ⚠️ Which of the two
rules governs when a dispatch *does* name it is with the maintainer as
objectui#9976.
>
> **Not corrected here, and named so it is not lost:** the shipped
changeset says the `page:card` retirement 「is objectui's」 without naming
**objectui#9961**, the card this seat filed for it after this head was
cut. ⛔ Not a falsified sentence — an uncited carrier, and the seat's
call whether the CHANGELOG paragraph should carry it.
>
> ⛔ Nothing below this block was edited in place.
Clause-②: yes

⭐ **Resolved by the seat on 2026-09-18, exactly the way the note below
asks:** the declaration of record is this card's claim comment, which
re-declares `yes`, and `needs:contract-review` is hung on both carriers.
The `no` that stood on this line was the **dispatch's** score — ⛔ the
note below is unchanged and was right to refuse to resolve it silently.

⚠️ **The ruling this executes says otherwise for one of its five steps,
and I am not choosing between them silently.** The dispatch scored this
card `no` on the ground that it NARROWS an accepted authoring set. The
2026-09-01 ruling scores its own **step 3** — retiring `body` from the
published type — as **`Clause-② yes at dispatch`**, and triage's
2026-09-17 answer bound the `CONTRACT_REVIEW_TIER` quota to that step
alone. Today's release record (comment `5730313581`) argued only the
producer-side condition, which binds steps 2 / 4 / 5. ⇒ **step 3's
recorded condition was not addressed by the release that unblocked this
card**, and whether it is live today is a seat reading, not mine. I have
drafted the whole ruling; the `needs:contract-review` label is the
seat's to attach and I have neither attached nor waited for it.

---

## What this is

The maintainer ruling of 2026-09-01 (「同意。」, option B) retires the `body`
child-list dialect: one concept, one spelling, and the spelling is
`children`. Step 1 (the liveness census) landed as #7184. **This is
steps 2 through 5.**

`BaseSchema` declared two spellings and left the choice per component.
Twelve registrations read `body` and nothing else, so it was their only
door; the wider set read `children || body` and took either. The
authoring tier only ever knew `children` — so an author writing the one
spelling that resolved got `unknown-prop`, the same warning a typo
draws, on the tier built for AI authors where the diagnostic **is** the
contract.

## The five steps, and where each one landed

| step | state |
|---|---|
| 1 — census | landed earlier (#7184); re-run here as the verification
instrument |
| 2 — converge the `body`-only registrations; fallback readers drop the
arm in the same change | this PR |
| 3 — retire `body` from the published type | this PR ⚠️ see the
Clause-② note above |
| 4 — the tier teaches `children` only | this PR |
| 5 — corpus and teaching migrate in the same commit | this PR |

**Step 2.** `alert`, `badge`, `tooltip` and the ten `sidebar-*` reads
converge on `children`. Every `children || body` fallback drops its
`body` arm: `basic/div.tsx`, `layout/card.tsx`, `layout/semantic.tsx`,
`layout/aspect-ratio.tsx`, `form/button.tsx` (named in the ruling's own
cross-card boundary paragraph), `layout/page.tsx`'s three reads,
`@object-ui/core`'s recursive `validateSchema`, the two container
walkers that enumerated child channels, and both VS Code extension
readers — whose own comments deferred the arm to this card by number.

**Step 3.** `BaseSchema.body` is `never` on the TypeScript face and an
alias refusal naming `children` on the Zod mirror, as are the four
per-component redeclarations. **Refused by name, not deleted** —
`BaseSchema` carries `[key: string]: any` and the mirror ends
`.passthrough()`, so a deleted member is accepted silently and rendered
by nothing, which is the exact silence this retirement ends.

**Step 4.** A new `sdui-parser/src/body-dialect.ts` answers an authored
`body` with the replacement NAMED, and it REPLACES the diagnostic the
prop walk would emit rather than stacking on it — the shape
`checkKanbanQuickAdd` next door is written against. A `body` child list
under a non-container draws the same containment code the `children`
spelling draws, which is the ruling's own sentence.

**Step 5.** 114 authored nodes moved across the docs site, the schema
catalog, five package READMEs and the extension's snippets / JSON schema
/ syntax map / hovers / completions, plus the prose that taught the key.

## ⚠️ A contrary fact, measured while implementing step 2 — pinned, not
worked around

Converging the twelve gives them a `children` read, so #6808's runtime
predicate calls them containers. #6804's ruling anticipated this and
wrote the answer in advance, verbatim: a census-pin exception **with its
ground**, ⛔ not `isContainer`. That is what landed — ten reasoned
entries in `container-declaration-baseline.json`'s `excluded`, and its
exact-equality pin moved with them.

**But the exception is no longer free, and the cost is this card's own
defect one key over:**

| | renders | draws `not-a-container` |
|---|---|---|
| before | `body` | `children` |
| after | `children` | `children` |

`validateTree`'s containment branch reads `node.children?.length`, so
the false warning that used to sit on a key these registrations did not
read now sits on the only key they do. Per the standing clause I
executed the ruling literally, made no promote and no revert in this PR,
and **inverted the pin into a refusal** that states the cost
(`container-declaration-ratchet.test.tsx`, the converged-readers block).

⇒ **The question for whoever reads this:** #6804's ground — `children`
here is a *label / content fallback*, not layout containment — transfers
cleanly to `badge`, `alert`, `button` and `tooltip`. It does not
obviously transfer to the `sidebar-*` family, whose child lists ARE
layout containment and which are NOT in the public tier, so the
react-page cost that justified `button`'s exclusion does not apply to
them either. That ruling was made before the convergence existed. ⛔ Not
decided here.

## Two readers deliberately left alone, each with its ground

- **`page:card`** still reads `schema?.body ?? schema?.children`. That
key is `PageCardProps.body` under a **different** retirement
(objectstack#5775, ADR-0087 D2), and its own comment records a
sequencing precondition — the load-time `body` to `children` conversion
— that deleting the read before would blank stored documents silently.
- **`items[]`-level `body`** on `list` (and the same shape on `tabs`
items and `dashboard` widgets) is objectui#9590's card, fenced off this
one. ⛔ Untouched, in either direction.

## Verification

- `pnpm census:body-dialect`, the instrument step 1 committed for
exactly this re-reading: **159 nodes carrying `body` on the retirement's
keys before, 38 after** — and all 38 are test fixtures, the scanner's
own planted fixtures, or two pre-existing changesets this branch does
not own.
- `packages/types` + `packages/sdui-parser`: **221 files, 5009 tests,
all passing.**
- `packages/components`: **282 files, 2743 tests, all passing.**
- Changeset gates green: presence (1 declared), no-major, no-overwrite.
Control bytes green. Governed-surface guard: NOT GOVERNED, 81 paths
checked.

## Files outside the dispatched surface, declared rather than widened
silently

The dispatch named `packages/components/src/renderers/`,
`packages/sdui-parser/src/` and `packages/types/src/base.ts`. The
ruling's steps 2, 3 and 5 reach further and each of these is named for
that reason: the rest of `packages/types/src` (the four redeclarations,
their mirrors, the pins that move with them),
`packages/core/src/validation/schema-validator.ts` (a generic `children
|| body` reader), `packages/vscode-extension/**` (two readers that
deferred to this card, plus the teaching surfaces),
`scripts/container-declaration-baseline.json` (#6804's ordered
exception), `content/docs/**`, `examples/schema-catalog/**` and five
READMEs (step 5).

## Acceptance notes

Noted while measuring, ⛔ not filed and ⛔ not repaired here:

- `dashboard` widgets author a per-widget `body`
(`content/docs/api/schema-reference.md`, `plugin-dashboard/README.md`'s
vocabulary note) that no renderer reads — same item-level family as
objectui#9590.
- About twenty per-component `body?: never` docblocks still justify
themselves with "`body` is inherited-and-optional from `BaseSchema`,
whose own docblock admits …". The refusals are correct and stay
word-for-word; only that one premise sentence is now stale.
- Two pre-existing changesets carry `body` migration recipes that will
teach the retired spelling when they reach a CHANGELOG. Not edited —
they belong to other cards and `check-changeset-overwrite` reports a
branch that modifies them.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01Xm4WFhEe5mwcgyqHjxR2hn

---
_Generated by [Claude Code](https://claude.ai/code)_

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: os-sales

.changeset/3965-mint-box-neutral-container.md

@@ -19,11 +19,12 @@ gap with a three-clause contract, pinned in
 2. authored `className` passes through **verbatim**;
 3. **zero** injected classes.
 
-Deliberately unlike `div`, `box` reads `children` only — never `schema.body`.
-The `div` renderer's `children || body` fallback is what made a mechanical
-`div`→X swap silently drop content on `body`-authoring nodes while the element
-count stayed unchanged; content moves into `children` at migration time (the
-objectui#6771 B-ruling direction).
+Deliberately unlike `div` AT THE TIME, `box` reads `children` only — never
+`schema.body`. The `div` renderer's `children || body` fallback is what made a
+mechanical `div`→X swap silently drop content on `body`-authoring nodes while the
+element count stayed unchanged. ⚠️ objectui#6771 has since executed that B-ruling and
+retired `body` across the protocol, so `div` reads `children` only too and `box` is no
+longer the exception this paragraph mints it as.
 
 Landed on both contract faces per the zod-mirror-parity pairing (objectui#6424
 family form): `BoxSchema` interface in `@object-ui/types`, its zod mirror in

.changeset/5632-sidebar-trigger-dom-passthrough.md

@@ -9,8 +9,9 @@ The renderer forwarded its whole prop bag to `SidebarTrigger`, which spreads its
 own rest onto the `<button>` it renders — so every authored SDUI key on the node
 became an attribute. Fourteen of them, one more than the shape this target
 derives from, because this registration also never destructured `schema`: every
-other registration in `renderers/navigation/sidebar.tsx` names it (they need
-`schema.body`), while this one renders no children and named only `className`,
+other registration in `renderers/navigation/sidebar.tsx` names it (they need the
+node's child list — spelled `schema.body` when this change landed, `schema.children`
+since objectui#6771 retired that spelling), while this one renders no children and named only `className`,
 so the node `SchemaRenderer` injects on every render rode the spread and landed
 as `schema="[object Object]"`.
 

.changeset/6764-container-declaration-census.md

@@ -7,9 +7,10 @@ they render (objectui#6764).
 
 `renderers/layout/semantic.tsx` registers `aside`, `main`, `header`, `nav`,
 `footer`, `section` and `article`, and `renderers/layout/aspect-ratio.tsx`
-registers one more; all eight call
-`renderChildren(schema.children || schema.body)` and none declared
-`isContainer`. Nothing on the render path reads that flag, so children always
+registers one more; all eight called
+`renderChildren(schema.children || schema.body)` when this change landed — the `body`
+arm has since been retired by objectui#6771 and they call
+`renderChildren(schema.children)` — and none declared `isContainer`. Nothing on the render path reads that flag, so children always
 rendered — what the omission did was make `validateTree` warn `not-a-container`
 on a child list it then rendered, on the tier built to accept AI-authored pages.
 A warning that lies trains authors to discount the true ones.

.changeset/6771-retire-body-child-list-dialect.md

@@ -0,0 +1,128 @@
+---
+'@object-ui/types': minor
+'@object-ui/components': minor
+'@object-ui/core': minor
+'@object-ui/sdui-parser': minor
+'@object-ui/cli': minor
+'@object-ui/app-shell': minor
+'object-ui': minor
+---
+
+**BREAKING (authoring surface): `body` is no longer a child-list key. Author `children`.**
+
+`BaseSchema` declared two spellings of one concept — `body` and `children` — and left the
+choice per component. **Thirteen** registrations read `body` and nothing else, so it was
+their ONLY door — the twelve the ruling enumerated plus `tooltip`, which it did not, and
+which is the count the committed instrument carries (`BODY_ONLY` plus
+`BODY_ONLY_UNRULED` in `scripts/body-dialect-census.mjs`). `div`, `card`, `page`,
+`button`, `aspect-ratio`, the seven sectioning tags and the safe-HTML tag factory read
+`children || body` and took either. The authoring tier only ever knew `children`, so
+an author writing the one spelling that resolved got `unknown-prop` — the same warning a
+typo draws, on the tier built to accept AI-authored pages, where the diagnostic **is** the
+contract (objectui#6771).
+
+Ruled 2026-09-01: one concept, one spelling, and the spelling is `children`.
+
+## What changed
+
+- **Renderers.** `alert`, `badge`, `tooltip` and the `sidebar-*` family read `children`.
+  Every `children || body` fallback drops its `body` arm **except the four `page:*`
+  reads named below** — `div`, `card`, `button`, `aspect-ratio`, the sectioning tags,
+  the safe-HTML tag factory behind ~36 tags, `page`'s flat content list, and
+  `@object-ui/core`'s recursive `validateSchema`.
+- **Item-level `body` is a DIFFERENT key and is untouched.** `list` draws each entry
+  as `item.content || renderChildren(item.body)` and `tabs` as
+  `item.content || item.body`, both filed under the ITEM type rather than the node,
+  and `tabs` still ships `body` inside its own `defaultProps`. ⛔ Neither is this
+  spelling and neither is refused here: the node-level retirement does not reach a
+  member of a declared `items` array. Retiring the item-level dialect is
+  objectui#9590's card, and the two named above are recorded on it.
+- **What still reads `body` at NODE level, and why.** Four renderer reads, all `page:*`: `page:card`
+  (renderer and Studio canvas) and the three thin `page:section` / `page:footer` /
+  `page:sidebar` containers. ⛔ Authoring the key is refused on them as it is
+  everywhere else; only the READ survives, for documents already STORED under it.
+  `@objectstack/spec` states the ground on `PageContainerProps` itself: «The renderers
+  keep reading `body` as a back-compat fallback for stored documents; that fallback is
+  objectui's to retire on its own schedule, and it is not a second authorable
+  spelling.» ⚠️ What separates the two halves is a CONVERSION, not a ground: the spec's
+  registry carries `page-card-body-to-children` (`toMajor: 17`, surface
+  `page.component.page:card.body`, shipped `retiredFromLoadPath: true` in spec 17.0.0)
+  and carries none for the other three — so a stored row under their key has no
+  migration path at all. Nothing in this repository authors any of them, and the
+  committed census over those keys is how to see that rather than take it on trust:
+  `node scripts/body-dialect-census.mjs --keys page:section,page:footer,page:sidebar,page:card,card`.
+  ⚠️ This sentence used to restate that table's answer instead of pointing at it, and
+  the very commit that edited the sentence moved the figure — a count copied into prose
+  is what AGENTS.md #9 forbids, so the instrument replaces the number. Read the table
+  for three things: `card` + `body` is a control that FIRES on the head that ships (so
+  a zero elsewhere is a reading, not a blind spot), every node still carrying `body`
+  under these keys falls in the `test` bucket, and `page:section` is a LIT zero — 6
+  nodes resolved, 0 `body`, 4 `children` — rather than a key nobody asked about.
+  Retiring the THREE THIN reads is objectui#9916,
+  which asks for the stored-corpus reading they turn on; `page:card`'s read is not
+  that card's either — it is objectui's own, grounded by objectstack#5775 /
+  ADR-0087 D2 rather than owned by it, and it ends when the stored rows have been
+  replayed through the conversion that already shipped for it.
+- **The published type.** `BaseSchema.body` is `never` on the TypeScript face and an
+  alias refusal naming `children` on the Zod mirror, as are the four per-component
+  redeclarations (`CardSchema`, `AspectRatioSchema`, `PageNodeSchema`, `TooltipSchema`).
+  ⚠️ Refused **by name**, not deleted: `BaseSchema` carries an index signature and the
+  mirror ends `.passthrough()`, so a deleted member would be accepted silently and
+  rendered by nothing — the exact silence this retirement ends.
+- **The VS Code extension (`object-ui`) changes behaviour, not just teaching.** Its
+  preview no longer draws a `body`-spelled document — the node renders empty, the way
+  the runtime renders it — and its validator now emits a warning naming `children` at
+  the key's own position. That warning is skipped for node types that declare their
+  own `body` input (`record:alert`), so a declared translation-map `body` is not
+  reported as the retired spelling.
+- **Two new `@object-ui/sdui-parser` exports.** `RETIRED_CHILD_LIST_KEY` (the retired
+  spelling, so a consumer names it once) and `checkRetiredBodyDialect` (the diagnostic
+  the tier substitutes for `unknown-prop` on that key).
+- **The authoring tier.** `sdui-parser` answers an authored `body` with the replacement
+  named, instead of the bare "has no prop" every typo gets — and a `body` child list
+  under a non-container draws the same containment code the `children` spelling draws.
+- **Corpus and teaching, in the same change.** 114 authored nodes across the docs site,
+  the schema catalog, five package READMEs and the VS Code extension's snippets, JSON
+  schema, syntax map, hovers and completions. The platform does not refuse a spelling it
+  still ships.
+
+## Migrating
+
+Rename the key. `{ "type": "card", "body": [...] }` becomes
+`{ "type": "card", "children": [...] }`; both faces now name `children` in the refusal,
+and `pnpm census:body-dialect` reports where the dialect still lives in a tree.
+
+⚠️ **Four renderer reads are deliberately untouched, all in the `page:*` namespace** —
+`page:card` and the three thin `page:section` / `page:footer` / `page:sidebar`
+containers. They are read-only back-compat paths for STORED documents, not a second
+authoring face: `@objectstack/spec`'s `PageContainerProps` says so itself — `children`
+is canonical, `body` is deliberately not declared, and «the renderers keep reading
+`body` as a back-compat fallback for stored documents; that fallback is objectui's to
+retire on its own schedule».
+
+⛔ **Authoring `body` on them is still refused**, on both published faces and at the
+authoring tier, exactly as everywhere else. What survives is the READ.
+
+⚠️ The two halves are not in the same state. `page:card` has a migration path — the
+spec's conversions registry carries `page-card-body-to-children` (`toMajor: 17`,
+surface `page.component.page:card.body`), which shipped with `retiredFromLoadPath: true`
+in spec 17.0.0. The other three have **no conversion at all**, so a stored row under
+their key has nowhere to be migrated to. ⇒ the two halves are tracked separately, and
+the pointer is not one card: retiring the three thin reads is objectui#9916, whose
+question is the stored corpus; retiring `page:card`'s read was **grounded by**
+objectstack#5775 / ADR-0087 D2, whose conversion has already shipped and whose
+remaining condition is replayed rows. ⚠️ Grounded by, not owned by — the distinction
+is the point. objectstack#5775 is CLOSED and its body is a props-declaration audit
+that carries no such step, so a reader sent there for the action finds none. The
+action is objectui's, on the spec's own terms quoted above: «that fallback is
+objectui's to retire on its own schedule.»
+
+**Non-rendering readers keep their arm on the same rule**, and none of them renders
+anything: while a renderer still reaches stored `body` content, a reader that must see
+the SAME content keeps its arm, or the renderer draws what the reader cannot find.
+Those are the two tab-subtree walkers in `renderers/layout/containers.tsx`,
+`app-shell`'s `pageSchemaIntrospect` (`CONTAINER_KEYS`) and `PageBlockInspector`
+(`STRUCTURAL_PROP_KEYS`, the inspector half of a stored `properties.body`), and the
+CLI's `OBJECTUI_STRUCTURAL_KEYS` — a file-IDENTIFICATION marker, where keeping `body` is
+what lets an old file still be recognised as an ObjectUI node and therefore refused,
+instead of silently not judged.

.changeset/6939-overlay-trigger-mirror.md

@@ -10,10 +10,12 @@ groups on that card, dispatched as its own PR per the ruling).
 Both members demanded `children` and omitted keys the renderer reads first, so
 `safeValidateSchema` refused two catalog entries that draw correctly:
 
-- **`tooltip`** now declares `trigger`, and `content` / `body` as the two halves of
-  one read (`renderers/overlay/tooltip.tsx:28,31` — `renderChildren(schema.trigger)`
-  and `schema.content || renderChildren(schema.body)`). The registration's own
-  `inputs` list `trigger` / `content` / `body` and never `children`. `trigger`
+- **`tooltip`** now declares `trigger`, and `content` plus the rich-content slot as
+  the two halves of one read (`renderers/overlay/tooltip.tsx` —
+  `renderChildren(schema.trigger)` and `schema.content || renderChildren(…)`). The
+  registration's own `inputs` list `trigger` / `content` / that slot. ⚠️ The slot was
+  spelled `body` when this change landed; objectui#6771 retired that spelling and it
+  is `children` now, on both the read and the published `inputs`. `trigger`
   follows `HoverCardSchema` two entries below, which is the settled in-repo shape
   for this slot.
 - **`context-menu`** now declares `triggerClassName`, `contentClassName` and

.changeset/7926-page-node-refuses-actions.md

@@ -13,7 +13,7 @@ parse GREEN and render nothing. `PageNodeSchema` never declared the key and
 through `BaseSchema`'s `.passthrough()`. Measured through the real
 `SchemaRenderer`: a `page` node with `actions: [{type:'button',label:'Add
 Product'}, …]` drew **0** buttons and the label appeared nowhere in the DOM,
-while the SAME two buttons in `body` drew **2**. Until objectui#7933 the array
+while the SAME two buttons in the page's child list drew **2**. Until objectui#7933 the array
 also reached the wrapper element as `actions="[object Object],[object Object]"`.
 
 `PageNodeSchema` now declares `actions` as an ADR-0049 refusal arm, so the same
@@ -26,14 +26,16 @@ the alert-dialog fixtures), and the authorable action FORM was already ruled on
 2026-08-25 for objectui#6497 / #6182: the declarative action object. Growing a
 reader here would have minted a fourth `actions` shape.
 
-**Migration.** Put the buttons in `body` as nodes — a `button`, or an
-`action:button` with a declared `actionType`:
+**Migration.** Put the buttons in the page's child list as nodes — a `button`, or an
+`action:button` with a declared `actionType`. ⚠️ That list was spelled `body` when this
+change landed; objectui#6771 retired the spelling, and the recipe below is written in
+the one that is left so a reader copying it gets a document that parses:
 
 ```json
 {
   "type": "page",
   "title": "Products",
-  "body": [
+  "children": [
     { "type": "flex", "justify": "end", "gap": 2, "children": [
       { "type": "button", "label": "Add Product", "variant": "default" }
     ] }

.changeset/8284-content-channel-per-component.md

@@ -25,7 +25,7 @@ that names the channel to write instead:
 | the renderer reads | components | now refused |
 |---|---|---|
 | `children` | `box`, `span`, `container`, `flex`, `stack`, `grid`, `scroll-area`, `form`, `toggle` | `body` |
-| `body` | `alert`, `badge`, `tooltip` (which reads `content` first, `body` as its fallback) | `children` |
+| `body` (at the time of this change; objectui#6771 has since retired that spelling and these three read `children`) | `alert`, `badge`, `tooltip` (which reads `content` first, the child list as its fallback) | the other channel |
 
 Which channel each renderer reads was measured with the TypeScript type checker over
 every `ComponentRegistry.register(...)` call in `packages/components` — a read site is a
@@ -43,9 +43,12 @@ this change. If your own metadata authors the refused channel on one of these tw
 types, the component was already drawing nothing there; rename the key to the one in the
 table.
 
-**Not narrowed here, and why.** Components whose renderer reads BOTH channels through a
+**Not narrowed here, and why.** Components whose renderer read BOTH channels through a
 live `children || body` fallback (`div`, `card`, `button`, `aspect-ratio`, the `page`
 family), components that read neither, and components with no dedicated declaration
-(`sidebar-*`, the `any`-typed registrations) keep both channels. Each is named on
+(`sidebar-*`, the `any`-typed registrations) kept both channels at the time of this
+change. ⚠️ objectui#6771 has since retired the `body` spelling outright, so those
+fallbacks now read `children` alone — this paragraph records what THIS change did and
+did not do, not the state of the tree at release. Each is named on
 objectui#8284 with the specific measurement it still needs; acting on any of them from the
 `packages/components`-only sweep would have been a guess.

.changeset/8344-node-recursion-point-redirect.md

@@ -6,7 +6,8 @@ Redirect the node recursion point from `BaseSchemaCore` to `AnyComponentSchema`
 (objectui#8344) — a nested node is now judged by its OWN component schema.
 
 **Behaviour change, deliberately, at every depth below the root.** Every child slot
-(`body`, `children`, and every per-component redeclaration of them) is
+(`body`, `children`, and every per-component redeclaration of them — objectui#6771 has
+since retired `body`, leaving `children` and its redeclarations) is
 `z.union([SchemaNodeSchema, z.array(SchemaNodeSchema)])`, and `SchemaNodeSchema`'s
 component arm was `BaseSchemaCore` — the ~21 base keys and nothing type-specific. So
 per-type enforcement was ROOT-ONLY, for every component type: objectui#7869 measured
@@ -56,8 +57,8 @@ a child by its own schema would therefore have ADMITTED, at every child slot, a
 the base arm refused. ⛔ That widening is eliminated rather than declared: the arm the
 recursion point installs carries a check that a nested `chatbot` node's `body` still fits
 the node slot. Measured, corpus-valid chatbot seed plus `body: { model, temperature }`:
-accepted at the root before and after; inside `card.body[]` and `div.children[]` refused
-before and refused now. ⇒ the redirect narrows at all 109 redeclarations and widens at
+accepted at the root before and after; inside the card's and the div's child lists
+refused before and refused now. ⇒ the redirect narrows at all 109 redeclarations and widens at
 none. The published `ChatbotSchema` is untouched — whether its own `body` should carry the
 chat API's params is a separate question, recorded on objectui#8572 and deliberately not
 decided here.
@@ -100,7 +101,7 @@ the `./zod` barrel under Node or vitest, `@object-ui/cli`'s `check` / `validate`
 Measured on the published `dist/zod` face of this head (Vite 8.2.1 lib build, `es`,
 esbuild-minified, `zod` 4.4.3 and `@objectstack/spec` external, so the figures are this
 package's own bytes; nested off-spec node = `{ type: 'icon', icon: 'check', size: 'huge' }`
-inside `card.body[]`, parsed through `CardSchema`):
+inside the card's child list, parsed through `CardSchema`):
 
 | entry | nested off-spec node | bundle (raw / gzip) | fill in output |
 | --- | --- | --: | --- |

.changeset/8572-chatbot-body-retired.md

@@ -49,7 +49,7 @@ objectui validate <root chatbot with body: { model, temperature }>
   exit 1 · Path: body · Code: invalid_type · message names `requestBody`   <- SUBJECT
 objectui validate <the same document spelled requestBody>
   exit 0 · "Schema is valid!"                                              <- LIT CONTROL
-objectui validate <the same chatbot inside card.body[]>
+objectui validate <the same chatbot inside the card's child list>
   exit 1 · refused through the node union                                  <- DEPTH
 ```
 
@@ -62,7 +62,7 @@ keys and recursion and no per-component key at all. Measured on the same documen
 chatbot + body: { model: 'gpt-4' }   console.warn "schema.children.type: type is required"
 chatbot + requestBody                (no warning)                          <- LIT CONTROL
 chatbot, neither key                 (no warning)                          <- LIT CONTROL
-card + body: [{ label: 'x' }]        console.warn "schema.children[0].type: type is required"
+card + <child list> [{ label: 'x' }]  console.warn "schema.children[0].type: type is required"
                                                                            <- FIRING CONTROL
 ```