release(v18): objectui's first 18.x release — the sync major, Changesets pre mode on next, and the two gate edits (objectui#12334) (#12401)

Fixes #12334

Clause-②: no

objectui's first 18.x release, prepared in one PR: the sync `major` that
follows `@objectstack/*` from 17 to 18, Changesets pre mode on the
`next` tag, and the two gate edits. Nothing is published by this PR.
Publishing stays the maintainer's act (merging the version PR);
`changeset version` and `publish-plan` ran only on a scratch clone and
nothing they wrote is committed.

## What changes

1. **The sync major.** `.changeset/12334-objectstack-18-sync-major.md`
declares `major` for all 40 members of the `fixed` group, so every
package's CHANGELOG says why it crossed to 18. Its body is published
text; for what a consumer must do it points at the `@objectstack/*` 18
entry of objectui#12030.
2. **Pre mode on `next`.** `.changeset/pre.json`, written by `pnpm
changeset pre enter next`: `{ "mode": "pre", "tag": "next" }`.
3. **`scripts/check-changeset-no-major.mjs`, under the maintainer's
ruling on objectui#12030 (comment 6149296149).** A `major` entry passes
only when its package is in a `fixed` group AND that group's current
major (the `version` its members declare, one major for all) plus one
equals the major of the `@objectstack/spec` range its members declare in
`dependencies` / `peerDependencies` / `optionalDependencies` (one major
for all). Every other `major` is refused, with the reason printed per
entry. The range reader is narrow (`^`, `~`, `=` or a bare `X.Y.Z`,
prerelease allowed): a union, an open lower bound, an x-range or a
protocol spelling refuses the `major` instead of guessing one, and so do
members that disagree, a missing manifest, and a group none of whose
members declares the spec. Manifests are read only when a `major` is
pending. `OBJECTUI_ALLOW_MAJOR` is retired: the script reads no
environment, pinned by a test on its comment-masked source. No workflow
file changes.
4. **`scripts/check-spec-range-floors.mjs` reads a prerelease floor.** A
lower bound may name a prerelease, and that exact prerelease is the
floor: fetched from the registry and judged by its own export surface,
like a release floor. No ordering is added: two different prerelease
floors in one manifest, a `||` union with a prerelease branch, and the
replacement-floor search (now `replacementFloorLine`, extracted from
`main` so it is testable) still refuse or name nothing, and an
unrecognised spelling still throws.
5. **Prose.** In `AGENTS.md` and `content/docs/guide/ci-cd-pipeline.md`,
only the sentence that named `OBJECTUI_ALLOW_MAJOR` changes; each now
states the computed admission and points at the script header as the
authoritative statement. Six pending changesets get a one-clause
correction (below).

## The `.changeset/pre/` question, measured

Changesets 3 (installed `@changesets/cli` 3.0.1) keeps no consumed ids
in `pre.json`. `enterPre` (`@changesets/pre`) writes `mode` and `tag`
only. In pre mode, `changeset version` MOVES each consumed changeset to
`.changeset/pre/` (`@changesets/apply-release-plan`). The planner leaves
`pre/` out while `mode` is `pre` (`getRelevantChangesets` in
`@changesets/assemble-release-plan`) and reads it again only on `pre
exit`.

So the bump gate reads top-level `.changeset/*.md` only. It always did;
this PR states why in the header and pins it:

- **It does not trip on a consumed `major`.** After the first prerelease
the group is on `18.0.0-next.0`, whose major is already the spec's, so
reading `pre/` would refuse the consumed sync major.
- **It does not miss a pending one.** A new `major` lands at the top
level and is refused, because 18 + 1 is not 18. That is "every other
`major` is refused".
- **On `pre exit`** the planner reads `pre/` again, but it increments
the group's prerelease. A pre-major `X.0.0-TAG.N` increments to `X.0.0`
for every bump type, so a consumed `major` cannot move the group off 18.

Measured on a scratch clone of `bd8892e70`:

- `pnpm changeset:version` printed `Version: 17.7.0 → 18.0.0-next.0`.
All 40 group manifests went to `18.0.0-next.0`, 471 of the 471 pending
changesets moved to `.changeset/pre/` (0 left at the top level), and
`pre.json` was unchanged.
- On that tree the bump gate exited 0 ("No changeset declares a `major`
bump."). With one probe `major` added at the top level it exited 1:
"carries the group from 18 to 19, but its @objectstack/spec range names
18".
- `changeset publish-plan --output`: 39 entries, every one `publish` at
`18.0.0-next.0` on tag `next`. The 40th member, `object-ui`, is private.
- The release lane's post-version validation on that tree:
`quick-reference:check` exit 0 and `check:control-bytes` exit 0. `vitest
run scripts/__tests__`: 184 files passed and 3 failed. Two of the
failures, `check-lint-rule-coverage` and `turbo-lint-inputs`, come from
the scratch clone itself: its `node_modules` is a symlink, and `git
check-ignore` refuses a path "beyond a symbolic link". Both pass in the
real worktree. The third, `markdown-test-inputs` `stale-test`, was this
branch's own: the rewritten bump-gate test read `.changeset/` only
through the gate, which the scanner cannot see. `3cbf4dbb6` fixes it,
and the ledger needed no edit.
- Not exercised: the `Changeset Release` workflow itself. Its publish
lane on a pre-mode tree is read from its clear step, which already
removes `.changeset/pre/*.md` and keeps `pre.json`. It is not measured.

## The dispatch's mechanism assumptions, measured on `main` at `28fb2a0`

- **A1 holds.** Every `@objectstack/spec` range in the workspace is
`^18.0.0-next.1`. The fixed group is 40 members, all at `17.7.0`, and 22
of them declare the spec in a consumer-facing field.
- **A2: both readings are right, at different layers.** `minVersionOf`
throws on every range. `effectiveFloor` turns each throw into an
`unreadable-range` finding, and no package counts as floored. `main()`
then stops at the collapse alarm before printing any finding. `pnpm
check:spec-floors` exited 1 with "The scan collapsed: 0 package(s)
declare @objectstack/spec in a consumer-facing field, fewer than the 8
this repository has."
- **A3 holds.** On `main` there is no `.changeset/pre.json` and no
`.changeset/pre/`. `node scripts/check-changeset-no-major.mjs` exited 0
with "No changeset declares a `major` bump."
- **A4 holds.** `git grep OBJECTUI_ALLOW_MAJOR` finds it in the script,
`AGENTS.md` and the CI/CD guide, and nowhere under `.github/`. Control:
the same grep for `check-changeset-no-major` finds 4 lines in
`.github/workflows/changeset-guard.yml`.
- **A5: carried by objectui#11824** (PR #12355, `cfb2947`). That PR
rewrote the `{record.FIELD}` spelling in
`.changeset/11788-notify-recipients-picker.md` as the CEL value
envelope. Re-read at this head, "These are the values the messaging
service already resolves" holds: the notify node evaluates an envelope
recipient per run (`resolveRecipients` in objectstack `notify-node.ts`)
and hands the messaging service the resolved values. Left unchanged.

## Pending changesets re-read (6159211733)

The six corrections below each add one clause. Every frontmatter is
byte-identical: `check-changeset-overwrite` reports "declared at base"
equal to "declares now" for all six.

- **`.changeset/11880-grid-filter-handoff.md`, the page-1 bullet.** This
is the item the review named. "It now returns to page 1" now says that
in 18 this holds for a named view's filter only, because `ObjectView` no
longer reads `table.defaultFilters` (objectui#12030).
- **The general check found four more.** The grep was by symbol and by
the words that spell it, not by file path. Three retirements had landed
after objectui#12030's sweep, and one more was a flow-template spelling:
- `.changeset/11838-rename-expression-refs.md`: "a record field reads
`{x.field}`". In 18 that value is a CEL value envelope, because spec 18
refuses the single-brace token there (objectui#12101). This is the
flow-template class the review found.
- `.changeset/11880-repeater-datasource.md` and
`.changeset/12085-element-binding-inputs.md`: both still said the
repeater reads its flat `properties.object` / `filter` / `sort` /
`limit` as a fallback. objectui#12345 (PR #12366) removed that read
today and left these two unchanged.
- `.changeset/11865-access-panels-shared-select.md` and
`.changeset/11865-object-settings-shared-select.md`: both still named
the external sharing-model picker. objectui#12353 removed it today.
These two spell it "external sharing model" in words, so a grep for
`externalSharingModel` alone misses them.
- **Read and left as written:**
- `.changeset/12032-approval-requests-data-source.md`: "a record page
over a `provider: 'api'` source". This is historical wording about
`ApiDataSource`, the same reading objectui#12364's PR recorded.
- The `defaultFilters`, `manage_org_presentation`, `*_OBJECT_REQUIRED`,
`element:record_picker`, `actionType: 'navigation'` and `onSuccess`
hits: each is already corrected, is its own retirement entry, or
describes something that still holds.
- **This list is valid at this head only.** Changesets that land before
this PR goes ready need the same re-read.

## Acceptance, mapped

| Acceptance | Result |
|:--|:--|
| This PR's changeset passes the bump gate | exit 0,
"`.changeset/12334-objectstack-18-sync-major.md`: 17 → 18, 40 `major`
entries" |
| `pnpm check:spec-floors` reads every floor at `18.0.0-next.N`, exit 0
on a built workspace | exit 0; 22 of 39 published packages floored; 11
entry points of `@objectstack/spec@18.0.0-next.1`; `--cross-check`
agrees |
| `changeset version` + `publish-plan` on a scratch copy plan every
package at `18.0.0-next.N` on `next` | 39 of 39 `publish`,
`18.0.0-next.0`, tag `next` |
| A `major` when the group already equals the spec's major is refused |
fixture, and live on the post-version tree (18 to 19) |
| A `major` two above the group's major is refused | fixture (group 16,
spec 18) |
| An unreadable range still fails the floor gate | fixtures, plus a live
leg below |
| `.changeset/pre/` measured, the gate neither misses a pending `major`
nor trips on a consumed one | section above; pinned by two fixture tests
and a live-tree enumeration test |
| One ablation leg | below |
| Nightly `Spec Range Floors` alarm green on `main` after landing | not
measurable before merge |

## Verification at `3cbf4dbb6`

- **Tests.** `pnpm exec vitest run` on `check-changeset-no-major`,
`check-spec-range-floors`, `markdown-test-inputs`, `ci-cd-pipeline-doc`,
`check-pre-install-import-graph` and `check-changeset-presence`: 6
files, 268 tests passed, exit 0. `pnpm type-check:scripts` exit 0, and
`tsc --listFilesOnly` lists both test files. `eslint` on the four
changed scripts: exit 0.
- **Gates, run locally, all exit 0:**
- `check-changeset-no-major`, `check-changeset-overwrite` (report-only,
six "declared at base = declares now"), `check-changeset-fixed`,
`check-changeset-presence` ("no changeset is owed") and
`check-changeset-claims`;
- `check-pending-changeset-literals`, `check-control-bytes`, and
`check-pre-install-import-graph` (plus its `--self-test`);
- `check-new-cross-file-line-citations` (0 new), `check-doc-links`,
`check-doc-fence-languages`, `check-shell-escape-residue` and
`check-entry-guard`.
- **`pnpm check:spec-floors -- --cross-check`** exited 0 after `turbo
run build` of the 22 floored packages and their dependencies (33 tasks).
- **Ablation, three legs, each restored by a trap and checked by blob
hash.** `git diff HEAD` was empty after all three.
1. The base `check-changeset-no-major.mjs` restored over the head
(on-disk blob = the base blob). The repository test "declares no pending
`major` the gate refuses — run as CI runs it" goes red on its assertion
with the base gate's own message: "A changeset declares a `major` bump:
@object-ui/types (.changeset/12334-objectstack-18-sync-major.md:2) …".
The other red tests are the new exports the base lacks.
2. `minVersionOf`'s lower-bound reader put back to plain `X.Y.Z`: 5
tests red, all prerelease-floor pins.
3. Live negative control: `packages/core/package.json` given a union
with a prerelease branch. The gate exits 1 with `@object-ui/core
[unreadable-range] … cannot order "18.0.0-next.1" against "17.0.0"`.

## Acceptance notes (observations, not filed)

- `.github/workflows/changeset-release.yml`, the comment on "Clear
pending changesets for the publish branch", still describes Changesets
2's `pre.json.changesets`. It says this repository has no `pre.json` and
no `.changeset/pre/`, which is false once this lands. The step's code
already removes `.changeset/pre/*.md`, so behaviour is unaffected. The
file is governed and outside this PR's surface. Carrier: none.

## Landing holds

- `AGENTS.md` is a governed surface (`node
scripts/check-governed-queue-guard.mjs --test` names it), so this PR
stays draft until a `GOVERNED_APPROVERS` approval.
- Every row of the pre-release list on objectui#12334 (`6164681402` and
its updates) must be closed, or released by the maintainer, before this
PR goes ready. That check is the seat's.

## 维护者速读(草稿)

**改了什么**:为 objectui 第一次发 18.x 做好准备,但本 PR 本身不发版。四件事:① 一个 changeset 把整个
fixed 组(40 个包)从 17 升到 18,对齐 `@objectstack` 18;② 进入 Changesets 的 pre
模式,标签 `next`,所以发出去的是 `18.0.0-next.N`,npm 的 `latest` 不动;③ 「禁止 major」门禁按您在
objectui#12030 的裁决收窄:只有正好把 objectui 带到 `@objectstack/spec` 所依赖的那个大版本的
major 才放行,其余照旧拦,原来那个靠手动设置的环境变量开关同时退役;④ spec 版本下限门禁学会读 `18.0.0-next.1`
这种预发布下限,否则发版命令会在发布前自己中止。另外顺手改正了 6 条待发 changeset 里在 18 下已不成立的半句话。

**为什么改**:您的指示是「objectui 的下一个版本,发 v18」。依赖已经在 objectui#12030、#12370 两步升到
`@objectstack` 18 的预发布线,但版本号没跟上;两道门禁在 18 预发布线上要么误拦(major
门禁),要么直接把发版命令卡死(下限门禁)。

**风险与代价(含回滚)**:合并本身不发包;真正的发布仍是您合并版本 PR 那一步。pre 模式下每次版本 PR 都会发 `next`
标签,直到 `@objectstack` 18 正式版后再执行 `changeset pre exit` 发
`18.0.0`。回滚办法:revert 本 PR,也就是删掉 `pre.json` 和那条 major
changeset,门禁恢复原样。已知未验证的一处:发布流水线本身在 pre 模式下的实际运行,目前只按其清理步骤的写法推断能正常走。

**席位意见**:

**你要做的**:批准本 PR,它改了 AGENTS.md,属于受管面。版本 PR 是否合并、何时发 `18.0.0-next.0`,由您决定。

---
_Generated by [Claude
Code](https://claude.ai/code/session_01JZnppgHRp4sREcouhkQYDB)_

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: os-zhuang <jack@objectstack.ai>

Author: 3 people

.changeset/11838-rename-expression-refs.md

@@ -4,7 +4,7 @@
 
 Renaming a node in the flow designer now also updates the expressions that read its outputs, removing a node that is still referenced is refused, and the Problems panel reports a reference to a node that does not exist as an error (objectui#11838).
 
-Later nodes read a node's outputs by its id. A decision branch reads `x.decision == 'approve'`, and a record field reads `{x.field}`. Renaming `x` used to update its edges and boundary event, but these expressions still named `x`. The saved draft then read a node that no longer existed, and the Problems panel showed only a scope warning.
+Later nodes read a node's outputs by its id. A decision branch reads `x.decision == 'approve'`, and a record field reads `{x.field}` (in 18 a CEL value envelope, `{ dialect: 'cel', source: 'x.field' }`, since `@objectstack/spec` 18 refuses the single-brace token there, see objectui#12101). Renaming `x` used to update its edges and boundary event, but these expressions still named `x`. The saved draft then read a node that no longer existed, and the Problems panel showed only a scope warning.
 
 - **The rename updates every expression that reads the node, in the same change.** This covers decision branches, edge conditions, value expressions, and `{x.field}` or `{{x.field}}` holes in text, including those inside a loop, parallel or try/catch region. Each expression is read with the expression parser. Only the node's name changes. A different identifier that contains the old id (`xy`), a field named like it (`y.x`), text inside quotes, and a variable the expression itself declares with the same name (`rows.exists(x, x.ok)`) are not changed.
 - **A rename that cannot update an expression safely is refused, and nothing changes.** If an expression that reads the node does not parse, the message under the ID field names it. The rename is also refused when the old or new id is also a variable name in the flow, because the expression could mean either one.

.changeset/11865-access-panels-shared-select.md

@@ -2,7 +2,7 @@
 '@object-ui/app-shell': patch
 ---
 
-Studio's package record sharing overview picks each object's internal and external sharing model with the shared `Select`. That is the control Studio's object Settings tab already uses for the same two sharing models (objectui#11865, the overview's part of that card). A permission set's advanced facets keep their native selects for now: the shared `Select` adds bytes to the console's first load there, and the first-load budget has no room for them yet.
+Studio's package record sharing overview picks each object's internal and external sharing model with the shared `Select` (the external dial is gone in 18: Studio no longer offers `externalSharingModel`, see objectui#12353). That is the control Studio's object Settings tab already uses for the same two sharing models (objectui#11865, the overview's part of that card). A permission set's advanced facets keep their native selects for now: the shared `Select` adds bytes to the console's first load there, and the first-load budget has no room for them yet.
 
 The overview's two dials were browser-native selects, so they looked and behaved differently from Studio's other dropdowns. They now use the shared Radix `Select`: the same trigger, dropdown and keyboard behaviour.
 

.changeset/11865-object-settings-shared-select.md

@@ -2,7 +2,7 @@
 '@object-ui/app-shell': patch
 ---
 
-Studio's object Settings tab picks the sharing model, the external sharing model, the name field, the lifecycle field and the next highlight field with the shared `Select`, the control Studio's validation rule editor already picks with (objectui#11865, the object settings part of that card).
+Studio's object Settings tab picks the sharing model, the external sharing model (gone in 18: Studio no longer offers `externalSharingModel`, see objectui#12353), the name field, the lifecycle field and the next highlight field with the shared `Select`, the control Studio's validation rule editor already picks with (objectui#11865, the object settings part of that card).
 
 The five pickers were browser-native selects, so they looked and behaved differently from Studio's other dropdowns. They now use the shared Radix `Select`: the same trigger, dropdown and keyboard behaviour.
 

.changeset/11880-grid-filter-handoff.md

@@ -9,6 +9,6 @@ An `object-view` grid now exports the filter it shows, and returns to page 1 whe
 Two user-visible corrections, both on an `object-view` rendered by the registered renderer (a page node, the Studio's stored-view preview; a host that supplies `renderListView` already behaved this way):
 
 - **The export carries the view's filter.** A named view's filter and an authored `table.defaultFilters` narrowed the rows on screen, but the server-streamed export (CSV, XLSX, JSON) was handed no filter and downloaded every record of the object. It now downloads the rows the view shows, as it already did for `table.filter`.
-- **Changing the filter returns the grid to page 1.** When a named view's filter or `table.defaultFilters` changed while the grid stayed mounted, the next query kept the old page index, which could ask for a page the new result does not have. It now returns to page 1, as `table.filter` already did.
+- **Changing the filter returns the grid to page 1.** When a named view's filter or `table.defaultFilters` changed while the grid stayed mounted, the next query kept the old page index, which could ask for a page the new result does not have. It now returns to page 1, as `table.filter` already did (for a named view's filter only in 18, where `ObjectView` no longer reads `table.defaultFilters`, see objectui#12030).
 
 What did not change: which filter applies (each rung and each pair sends the same query and draws the same rows; in 18 the `table.defaultFilters` rung is retired, see objectui#12030), and a refused filter still draws the grid's malformed-filter panel and queries nothing. Nothing is added to or removed from any published type, schema or export.

.changeset/11880-repeater-datasource.md

@@ -5,5 +5,5 @@
 
 `element:repeater` reads the node-level `dataSource` binding first (objectui#11880, the repeater half of the objectui step of objectstack#11509, ruled A-narrow).
 
-- **Renderer (`@object-ui/components`).** A repeater bound only through `dataSource` now lists the records it names; it showed "No records" before. Its flat `properties.object` / `filter` / `sort` / `limit` stay read, as the fallback (spec 18 retires them as authored keys, see objectui#12030, and the renderer keeps this read for metadata written before), so a repeater that carries only the flat keys reads exactly as before. Where a node carries both, the precedence is the one `ElementDataSourceGate` applies to every object-bound block: the binding's `object`; the flat `filter` AND-combined with the binding's (and its saved view's), so neither is dropped; the binding's own `sort` and `limit`, else the flat key, else the saved view's. An unresolvable `view`, or a filter the merge refuses, shows the configuration-error panel and reads nothing. The registration now publishes the injected `dataSource` input; `object` stays required (until objectui#12085 dropped the flat query inputs).
+- **Renderer (`@object-ui/components`).** A repeater bound only through `dataSource` now lists the records it names; it showed "No records" before. Its flat `properties.object` / `filter` / `sort` / `limit` stay read, as the fallback (spec 18 retires them as authored keys, see objectui#12030, and the renderer keeps this read for metadata written before), so a repeater that carries only the flat keys reads exactly as before (superseded in 18 by objectui#12345: the repeater no longer reads the flat keys at all, alone or beside a binding, and reads its query from `dataSource` alone). Where a node carries both, the precedence is the one `ElementDataSourceGate` applies to every object-bound block: the binding's `object`; the flat `filter` AND-combined with the binding's (and its saved view's), so neither is dropped; the binding's own `sort` and `limit`, else the flat key, else the saved view's. An unresolvable `view`, or a filter the merge refuses, shows the configuration-error panel and reads nothing. The registration now publishes the injected `dataSource` input; `object` stays required (until objectui#12085 dropped the flat query inputs).
 - **Schema (`@object-ui/types`).** `ElementRepeaterBlockSchema` declares `dataSource` as `@objectstack/spec`'s `ElementDataSourceSchema`, by reference, and its bag's `object` stays required (until objectui#12056's waiver, and retired in 18, see objectui#12030). Clause-②: the strict authoring face (`StrictAnyComponentSchema`) now ACCEPTS a repeater carrying `properties.object` plus a well-formed `dataSource`, which it refused as an unrecognized key (since 18 it refuses `properties.object` by name, so the accepted node carries the binding alone); and the tolerant face (`safeValidateSchema`) now JUDGES the binding, so a repeater whose `dataSource` the spec's own node schema refuses (a string, a non-string `object`, an undeclared member, a record-form `filter`) is refused at `dataSource` where it was passed through unjudged. The TypeScript node type derived from the arm types the binding. A repeater whose bag omits `object` is refused at `properties.object` on both faces, as before; retired in 18, see objectui#12030: such a repeater is accepted.

.changeset/12085-element-binding-inputs.md

@@ -6,6 +6,6 @@
 
 **Narrowing, no widening.** The `element:repeater` registration drops its `object` (which was `required: true`), `filter`, `sort` and `limit` inputs. The `element:number` registration drops its `object` and `filter` inputs, which the metric has not read since objectui#11880. The injected `dataSource` input is unchanged on both. In the manifest these registrations feed, a repeater bound only through `dataSource` now draws no diagnostic (it drew `missing-required-prop`), and each of those six flat keys now draws `unknown-prop`. No export is added or removed, and no input is added.
 
-**Rendering is unchanged.** The repeater still reads its flat `properties.object` / `filter` / `sort` / `limit` as the binding's fallback, for metadata written before the retirement. That read is not published as an authoring surface. The metric reads its query from `dataSource` only, as before.
+**Rendering is unchanged.** The repeater still reads its flat `properties.object` / `filter` / `sort` / `limit` as the binding's fallback, for metadata written before the retirement (until objectui#12345 removed that fallback in 18, so the repeater reads its query from `dataSource` alone). That read is not published as an authoring surface. The metric reads its query from `dataSource` only, as before.
 
 objectstack's `sdui.manifest.json` is regenerated from these registrations at objectstack's next console pin bump; this release does not change that file.

.changeset/12334-objectstack-18-sync-major.md

@@ -0,0 +1,48 @@
+---
+"@object-ui/types": major
+"@object-ui/core": major
+"@object-ui/i18n": major
+"@object-ui/react": major
+"@object-ui/components": major
+"@object-ui/react-runtime": major
+"@object-ui/sdui-parser": major
+"@object-ui/fields": major
+"@object-ui/layout": major
+"@object-ui/data-objectstack": major
+"@object-ui/cli": major
+"@object-ui/create-plugin": major
+"@object-ui/runner": major
+"@object-ui/auth": major
+"@object-ui/permissions": major
+"@object-ui/mobile": major
+"@object-ui/plugin-ai": major
+"@object-ui/plugin-calendar": major
+"@object-ui/plugin-charts": major
+"@object-ui/plugin-chatbot": major
+"@object-ui/plugin-dashboard": major
+"@object-ui/plugin-designer": major
+"@object-ui/plugin-detail": major
+"@object-ui/plugin-editor": major
+"@object-ui/plugin-form": major
+"@object-ui/plugin-gantt": major
+"@object-ui/plugin-grid": major
+"@object-ui/plugin-kanban": major
+"@object-ui/plugin-list": major
+"@object-ui/plugin-map": major
+"@object-ui/plugin-markdown": major
+"@object-ui/plugin-report": major
+"@object-ui/plugin-timeline": major
+"@object-ui/plugin-tree": major
+"@object-ui/plugin-view": major
+"@object-ui/collaboration": major
+"@object-ui/app-shell": major
+"@object-ui/providers": major
+"@object-ui/console": major
+"object-ui": major
+---
+
+objectui moves to major 18, following `@objectstack/*` from 17 to 18 (objectui#12334). objectui's major is not a count of its own breaking changes: it tracks the major of the `@objectstack/spec` its packages depend on, so that the same major means compatible across the two projects. These packages already declare `@objectstack/*` on the 18 line (objectui#12030), so every package in the fixed release group crosses to 18 together. It is the one `major` that rule admits: objectui's own breaking changes ship as `minor`, described in their entries.
+
+The 18 line is released from Changesets pre mode on the `next` tag (`.changeset/pre.json`), as `@objectstack/*` 18 is: these releases are `18.0.0-next.N` on the npm `next` dist-tag, and they do not move `latest`. The first `18.0.0` on `latest` comes when `@objectstack` 18 is generally available and this repository leaves pre mode.
+
+What a consumer must do to take it is in the `@objectstack/*` 18 entry of this release (objectui#12030): move its own `@objectstack/*` packages to the 18 line at the same time.

.changeset/pre.json

@@ -0,0 +1,4 @@
+{
+  "mode": "pre",
+  "tag": "next"
+}

AGENTS.md

@@ -282,7 +282,7 @@ AGENTS.md 的「只跑受影响的包」指的是**用上面的路径过滤缩
 - minor/patch **独立演进**——objectstack 没动时不必跟发;objectui 自己的改动照常用 changeset 推进(从当前 major 起步,如 `11.0.0 → 11.1.0`)。
 - objectstack 跨 major(→12)时,下一次 objectui 发版一并把 major 提到 `12`。
 - 推论:**changeset 里不要声明 `major`** —— fixed 组任一 `major` 都会把全组推上去、脱离 objectstack 的节奏(如 17.x 期间被推到 18)。objectui 自身的破坏性变更也标 `minor`(在正文里写清 breaking 语义即可);唯一例外是跟随 objectstack 跨 major 的那一次同步升级。
-- **这一条现在由 CI 机械强制** —— `scripts/check-changeset-no-major.mjs` 在任一 changeset 声明 `major` 时退出非零,由 `.github/workflows/changeset-guard.yml` 跑(它是唯一以 `.changeset/**` 为**触发**路径的 workflow。这里曾写「`ci.yml`/`lint.yml` 都把 `**/*.md` 和 `.changeset/**` 列进 `paths-ignore`,只加 changeset 的 PR 不会启动任何 workflow」——**已不成立,别照它做判断**:objectui#3523 step 2 把 `paths-ignore` 从这两个 workflow 的 `pull_request` 上删掉了,今天它**只**留在 `push` 触发上;PR 的路径判断改在 job 内做 —— `ci.yml` 的 `Decide whether this change needs a full run` 步骤,`lint.yml` 有同名的一份。所以 md-only / changeset-only 的 PR **照常启动**这两个 workflow 并产出 required context —— 实测 PR #3856(只改一个 `.md`)起了 16 个 check、PR #4339(只给 AGENTS.md 加一行)起了 17 个。真正被跳过的是重活:那个 job 内开关的排除表逐条**等于** `push` 的 `paths-ignore`(仍含 `**/*.md` 与 `.changeset/**`),changeset-only 的 PR 里这两个 workflow 每一步都 skip。这才是 changeset-guard.yml 今天仍必须独立、且以 `.changeset/**` **反向**触发的理由:它跑在那个 job 内开关之外,免安装免构建,是唯一判得到 changeset-only PR 的东西。见 objectui#3523 step 2 / objectui#3857),`pnpm test` 里另有一条仓库状态断言兜底。跟随 objectstack 跨 major 的那一次发版设 `OBJECTUI_ALLOW_MAJOR=1` 放行。前情:objectui#3161/#3159/#3160/#3225 四个 changeset 在 17.x 期间标了 `major`(17 个包条目),足以把 39 个包发成 `18.0.0`。
+- **这一条现在由 CI 机械强制** —— `scripts/check-changeset-no-major.mjs` 在任一 changeset 声明 `major` 时退出非零,由 `.github/workflows/changeset-guard.yml` 跑(它是唯一以 `.changeset/**` 为**触发**路径的 workflow。这里曾写「`ci.yml`/`lint.yml` 都把 `**/*.md` 和 `.changeset/**` 列进 `paths-ignore`,只加 changeset 的 PR 不会启动任何 workflow」——**已不成立,别照它做判断**:objectui#3523 step 2 把 `paths-ignore` 从这两个 workflow 的 `pull_request` 上删掉了,今天它**只**留在 `push` 触发上;PR 的路径判断改在 job 内做 —— `ci.yml` 的 `Decide whether this change needs a full run` 步骤,`lint.yml` 有同名的一份。所以 md-only / changeset-only 的 PR **照常启动**这两个 workflow 并产出 required context —— 实测 PR #3856(只改一个 `.md`)起了 16 个 check、PR #4339(只给 AGENTS.md 加一行)起了 17 个。真正被跳过的是重活:那个 job 内开关的排除表逐条**等于** `push` 的 `paths-ignore`(仍含 `**/*.md` 与 `.changeset/**`),changeset-only 的 PR 里这两个 workflow 每一步都 skip。这才是 changeset-guard.yml 今天仍必须独立、且以 `.changeset/**` **反向**触发的理由:它跑在那个 job 内开关之外,免安装免构建,是唯一判得到 changeset-only PR 的东西。见 objectui#3523 step 2 / objectui#3857),`pnpm test` 里另有一条仓库状态断言兜底。跟随 objectstack 跨 major 的那一次同步升级由脚本按 manifest 自行判定放行,不靠开关:只有 fixed 组成员上的 `major`、且该组当前 major + 1 正好等于组内 manifest 在 `dependencies`/`peerDependencies`/`optionalDependencies` 里声明的 `@objectstack/spec` 范围的 major 时才放行,其余照旧拦(维护者裁决,objectui#12030 评论 6149296149;判据的权威表述是该脚本的文件头);Changesets pre 模式下已被 `changeset version` 消费、挪进 `.changeset/pre/` 的 changeset 不再读。前情:objectui#3161/#3159/#3160/#3225 四个 changeset 在 17.x 期间标了 `major`(17 个包条目),足以把 39 个包发成 `18.0.0`。
 - 这是约定优先于 semver 纯粹性的取舍(为可维护/好记),因此 objectui 的 major 不代表「它自身 API 的破坏性变更次数」。`@object-ui/site` 与 `@object-ui/example-*` 在 `ignore` 列表,不随组联动。
 
 ### 多 agent 协作纪律(并行修改本仓库,务必遵守)

content/docs/guide/ci-cd-pipeline.md

@@ -2576,8 +2576,13 @@ Runs `scripts/check-changeset-no-major.mjs`, which fails if any pending changese
 breaking changes of our own as `minor` and describe the break in the changeset body.
 
 The one release that legitimately bumps the major is the one following `@objectstack` across
-its major; it sets `OBJECTUI_ALLOW_MAJOR=1`. `pnpm test` asserts the same repository state, so
-the rule survives this workflow being skipped.
+its major, and the script admits it by itself, with no switch to set: a `major` on a `fixed`-group
+package passes only when the group's current major plus one equals the major of the
+`@objectstack/spec` range the group's own manifests declare in `dependencies`,
+`peerDependencies` or `optionalDependencies`; every other `major` is refused. In Changesets pre
+mode, a changeset that `changeset version` has consumed moves to `.changeset/pre/` and is no longer
+read. The script's header is the authoritative statement of the criterion. `pnpm test` asserts
+the same repository state, so the rule survives this workflow being skipped.
 
 #### Second job: Changeset Overwrite Report