Repository navigation
fix(app-shell): read the single AI quota pool and draw one ring (objectui#8524) - #10911
Merged
objectstack-fleet[bot] merged 6 commits intoSep 28, 2026
Merged
Conversation
The cloud usage endpoint now answers { pool, breakdown } (one quota pool plus
an observation-only split). useAiUsage parsed only the retired per-meter
{ meters: { build, dataChat } } answer, so it returned null and the ChatDock
indicator disappeared without any error.
- useAiUsage reads { pool, breakdown? } strictly. An absent breakdown is a
normal reading; the retired { meters } answer parses to null and sets error.
- AiUsageIndicator draws ONE ring from pool. The split is listed as text in
the popover (share of the same pool), only when a member was measured.
- New console.ai.usage.breakdownTitle row in every locale.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015AUunPkX7UTkCH9e7AdZo1
…ectui#8524) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015AUunPkX7UTkCH9e7AdZo1
…usage-single-pool
…usage-single-pool
…dy the indicator guard Comment wording only in useAiUsage (the optional breakdown comes from cloud#2133, not ADR-0015), and a behaviour-identical early-return guard in AiUsageIndicator. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015AUunPkX7UTkCH9e7AdZo1
…usage-single-pool
Contributor
|
changeset-claim-re-read
|
Contributor
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
This was referenced Sep 28, 2026
Contributor
Author
Contract reviewServed-tier: Inputs read: card #8524 (body and all five comments), PR #10911 (body, 15-file list, net diff against ① Derived judgmentsAccept set of
Public surface and behaviour:
Shipped prose —
Locale copy: PR body producer sentences: " Nothing in the diff or its prose is judged WRONG. ② Semver levelThe changeset declares
The changeset's level matches what the diff publishes and matches ③ Boundary flagsThe dev report (
Check-runs on the head: 43 in all, 40 Implemented-by: VERDICT: PASS Generated by Claude Code |
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
Sep 28, 2026
This was referenced Sep 28, 2026
objectstack-fleet
Bot
deleted the
claude/issue-8524-ai-usage-single-pool
branch
September 28, 2026 09:02
akarma-synetal
pushed a commit
to akarma-synetal/objectui
that referenced
this pull request
Oct 7, 2026
… family D input slice) (objectstack-ai#10907) Part of objectstack-ai#9256 Clause-②: yes **Clause-② `yes`, as the claim declared:** `input` accepted an authored `children` on both published faces and rendered nothing, with no render-time error or warning and no element; only the parser tier's `not-a-container` warning (objectui#9910) noticed it. It now refuses it by name. A published accept set narrows, so a contract review is owed before landing. This is the family-D slice that release `5864147075` on objectui#9256 lists, and nothing else: `input`, the family-D test header's superseded remarks, and `PartialSchema`'s docblock claim about generic `Omit` re-spellings. `br`, `hr` and `img` are not touched. objectui#9256 stays open. ## What changed - **`input` narrowed** in the family-D shape PRs objectui#9589 and objectui#10897 used: - TypeScript face (`InputSchema` in `packages/types/src/form.ts`): `body?: never` and `children?: never`, each with a docblock saying what the renderer reads and what it renders instead. - zod mirror (`InputSchema` in `zod/form.zod.ts`): two `retirementTombstone` members fed one string, `INPUT_NEITHER_CHANNEL`, built by `neitherContentChannelGuidance`. Both stay MEMBERS, so `zod-mirror-parity`'s key sets stay equal. - `body` was already refused on both faces by `BaseSchema` (objectui#6771). It is restated because that refusal names `children` as the remedy, and `input` does not read `children` either. - **`email` / `password` (`InputShorthandSchema`)**: no type or accept-set change. See "The shorthand's own pair" below. - **Texts (comments only):** - The family-D test header: 「family E, frozen」 (the `InputSchema` bullet), 「unattributable」 (the `DetailViewSchema` bullet) and the two `ui:calendar` remarks that still said plain `Omit` (the header bullet on the six bare-name hold-outs, and the comment above the slice-2 rows) now say what is true. No assertion in that file moved. - `PartialSchema`'s docblock in `packages/types/src/index.ts`: the sentence reading every generic re-spelling as collapsing is marked false, with `OmitDeclared` named as the counter-example. - **Bounded in-place fix, declared:** `partial-schema-collapse-pin.test.ts`'s header carried the same false claim, and even named the key-remapping spelling as one that collapses. It is corrected the same way, comment only. See "Texts" below for the four conditions and the evidence. - **New pin file** `packages/types/src/__tests__/content-channel-input-9256.test.ts`. - **One changeset** `.changeset/9256-input-content-channels.md`: `@object-ui/types` `minor` with an explicit BREAKING note and a migration line, the spelling PRs objectui#9589 and objectui#10897 used under this repo's no-major rule. - **Four pending changesets are corrected, prose only.** - Round 2 added a dated note to `6397-partial-schema-collapse-pin.md`, `9256-content-channel-family-d.md` and `8499-node-slot-registered-arms.md`. - Round 3 added a second note to the family-D entry and reworded the E3 entry's render claim in place. - Each carries a sentence a later change in this same release made false, or an overstatement, and each ships verbatim into the CHANGELOG. Every frontmatter block is byte-identical. See "Round 2" and "Round 3" below. ## Measurement, taken on `origin/main` `244d516df` before any edit **Instrument.** The TypeScript compiler API (6.0.3), ⛔ not grep. It builds one program per `tsconfig.json`: the 40 workspace packages, `apps/console` and the three examples, 44 programs over 2015 non-test source files. It ran on a BUILT tree (`turbo run build` 43/43, 0 cached) with 0 unresolved-module diagnostics. It files every `.body` / `.children` read (property access, string element access, object destructuring) under the declared type of its receiver: 520 channel reads were filed. - **`InputSchema`: 0 `body` / 0 `children` reads.** `InputShorthandSchema` is never a receiver: the `email` / `password` wrappers are `any`-typed and hand the node to the `input` renderer, which reads it as `InputSchema`. - **Lit controls fire in the same run:** `ButtonSchema` 1, `DivSchema` 1, `CardSchema` 1, `ContainerSchema` 1 (each `children`). - **Receiver reachability:** the same programs see `InputSchema` as the receiver of 23 reads over 17 keys: `defaultValue`, `description`, `error`, `id`, `inputType`, `label`, `max`, `maxLength`, `min`, `name`, `pattern`, `placeholder`, `readOnly`, `required`, `step`, `value`, `wrapperClass`. Neither channel is among them, so the zero is a reading, not blindness. **Every registration that claims `input`, bare and namespaced** (`check:registry-bare-names --json`, 1657 files, 424 claims, 0 contested): - `input`: sole claimant `ui:input`, a literal `ComponentRegistry.register('input', InputRenderer, …)` in `renderers/form/input.tsx`. The hop is typed (`InputRenderer` takes `schema: InputSchema`), not `any`. No `field:input` or other claimant exists; `ai:input` is a different key, a guarded protocol placeholder. - `email` / `password`: `ui:email` / `ui:password` own the bare keys; `field:email` / `field:password` stand down with `skipFallback: true`. **Does any path pass the node's content through? No.** - `SchemaRenderer` destructures `children` and `body` out of the props bag before the spread. - The only pass-through `InputRenderer` gives the native element is `toFormControlDomProps(inputProps)`, a whitelist pick of `SDUI_DOM_PASS_THROUGH_KEYS` plus `name` / `disabled`; `children` is not on it. - The `email` / `password` wrappers only spread `props` and `props.schema` into `InputRenderer`. - The registration declares no `children` slot input (objectui#9910). - No in-source host builds an `input` node by spreading another node. The literal `input` nodes in `form.tsx`, `cli`'s `init` scaffold and the dashboard / designer config panels are assembled key by key, and none carries a channel. - **The 75 `any`-typed channel reads were attributed file by file.** None can receive an `input` node's channel: - app-shell nav / search trees, and HTTP request, error or email `body` payloads; - React `children` destructured off component props (`view:simple`'s renderer, `react-page`'s wrapper), which `SchemaRenderer` never fills from a node; - the metadata-admin preview canvases, which read a page block's `properties.children` or a page draft's root `children` to list and append blocks for selection, not to render a node's channel; - the page-container and sectioning registrations in `containers.tsx`, `page.tsx` and `semantic.tsx`, and the html-elements factory, whose tag list excludes `input` by name (its own comment lists `input` among the tags it never registers); - `core`'s schema builder (card / grid / flex builders; `InputBuilder` has no children setter); - `record:alert`'s own `props.body`, the runner's nav items, and the vscode validator. - **Generic traversers** (`validateChildren` in core, `sdui-parser`'s parse / validate, `cli validate`) walk children to validate. They render nothing and are node-agnostic, as for every family-D row already landed. **Producers.** `pnpm census:body-dialect --keys input,email,password,ui:input,ui:email,ui:password,div,card,page,button` read 9016 files. The key population is printed by the tool itself, so objectui#9545's blindness does not apply. - Nodes on the six input keys: `input` 321, `email` 75, `password` 60, `ui:email` 1, `ui:password` 17, `ui:input` 0. **0 author `body`, 0 author `children`.** - The controls fire in the same pass: `children` on `div` 172, `card` 183, `page` 51, `button` 5. - Real producers: 0. Nothing was migrated. ## The shorthand's own pair: type-redundant now, kept on purpose `InputShorthandSchema` inherits `InputSchema` through `OmitDeclared` on the TypeScript face and `.omit()` on the mirror, so it now inherits the new pair. It already declared its own (PR objectui#10897). - **Redundant as a type:** on the built d.ts both are the same `?: never`, and the new `Eq` pin makes `tsc` hold that. - **Not redundant as text.** The repo's rule for a restated tombstone is the one `neitherContentChannelGuidance`'s docblock gives for restating `body` over `BaseSchema`: restate when the inherited message misdirects this node's author. The helper's contract spells the node and its route into the message. The inherited string names `input` and the `ui:input` route; the shorthand's names `email` / `password` and the wrapper route. - **Decision:** keep both members and rewrite their docblocks (TS) and comment (zod) to say they are type-redundant and why they stay. The shorthand's old `body` docblock gave a reason that became false with this change (that the inherited refusal names `children` as the remedy), and it is replaced. - **Pinned:** a CONTROL row asserts that an author of `email` / `password` reads the message naming those two nodes and not the `input` one, so deleting the restatement turns it red. - Counter-precedents, noted: `UiCalendarSchema` restates nothing and inherits `calendar`'s message, which names it. objectui#8072 deleted the shorthand's `wrapperClass` restatement once `InputSchema` carried the key; that was a live key with one description, not a refusal message naming a node. ## The `OmitDeclared` propagation, measured on the BUILT d.ts A compiler-API probe over `packages/types/dist/form.d.ts` reads the heritage of `InputShorthandSchema` on its own: the base type the checker gives for `OmitDeclared` of `InputSchema` minus `type` / `inputType`, before the interface's own members. | built d.ts | `InputSchema.children` | heritage `.children` | shorthand face `.children` | |---|---|---|---| | base `244d516df` | `SchemaNode \| SchemaNode[]` | `SchemaNode \| SchemaNode[]` | `undefined` (its own restated tombstone) | | this branch | `undefined` | `undefined` | `undefined` | A consumer-side probe resolves `@object-ui/types` the way `packages/components` does (`--traceResolution`: `packages/types/dist/index.d.ts`). There, `input` + `children`, `input` + `body` and `email` + `children` each answer their `@ts-expect-error`, and a lit control (`const n: number = 'x'`) fires. ## Red on base, then green — one-shot, the restore proven by state The predictions were written to a file BEFORE the mutation. The mutation was `git checkout 244d516` of the two declaration files (`form.ts`, `zod/form.zod.ts`), with every test at HEAD. It was proven by both blob hashes equal to BASE and by anchor counts: `INPUT_NEITHER_CHANNEL` 0, the new TS docblock marker 0. The run sat under `trap … EXIT INT TERM` with absolute paths. | reader | mutated (BASE declarations) | restored (HEAD `72dba894a`) | |---|---|---| | vitest, `content-channel-input-9256.test.ts` | **RED** exit 1: 6 failed / 13 passed | **GREEN** 19 / 19 | | `tsc -p packages/types/tsconfig.test.json` | **RED** exit 2: TS2578 on the `inputChildren` pin, TS2322 on the `Eq` pin for `children`; 0 elsewhere (`zod-mirror-parity` included) | **GREEN** exit 0 | | built d.ts probe (types rebuilt; dist markers 0 then 2) | `InputSchema` and heritage `children` accepted | refused | - **The 6 red rows:** `input.children` refused at its path, its message, `.describe()` equality, every-value, and the root + nested `AnyComponentSchema` row; plus the `input.body` message row. The base message is `BaseSchema`'s `Did you mean` pointer. - **Green on base, by design:** `input.body` refused at its path, every-value and `AnyComponentSchema` (`BaseSchema` already refused `body`), and the shorthand message CONTROL (its own pair is unchanged). - **One prediction was recorded as uncertain and read, not assumed:** `input.body` `.describe()` equality is GREEN on base, because `BaseSchema`'s alias refusal also describes with its own message. -⚠️ `?: never` is erased before vitest runs: the TypeScript half is read by `tsc` alone. `tsc --listFiles` counts the new pin file 1 under `tsconfig.test.json` and 0 under `tsconfig.json`, so it is a real assertion under the `type-check` script. - **Restore:** proven by both blob hashes equal to HEAD, `git diff HEAD` empty and `git status` clean. ## Texts - **Family-D test header.** The `InputSchema` bullet now records the plain-`Omit` hold-out ground in the past tense, the E3 slice's `OmitDeclared` repair, and this slice's narrowing, pinned in the new file. The `DetailViewSchema` bullet records that the E3 slice attributed the `any` hop directly and narrowed it there. The two `ui:calendar` remarks name `OmitDeclared` for the TS face, `.extend()` for the mirror, and say which file pins which half. The third `ui:calendar` remark (the old TRIPWIRE note) was already in the past tense and is unchanged. - **`PartialSchema`'s docblock.** The triage's reading is kept as the triage's. The re-spelling claim is marked FALSE, with `OmitDeclared` as the counter-example. The docblock now says that re-spelling the alias would narrow a published type, a contract change not made in a comment correction, and that objectui#6397 is closed. The alias itself is unchanged. - **Bounded in-place fix: `partial-schema-collapse-pin.test.ts` header** (comment only; its assertions are about the alias as written and do not move). All four conditions hold: 1. It is the same defect as the carried item: the same sentence, one hop away. The corrected docblock points readers at this file. 2. The fix is mechanical, and its shape is pinned by the carried correction. 3. No open PR touches the file. 4. Same gate family: types `tsc` and vitest, no new verification surface. **Evidence, a one-shot probe against the built d.ts (not shipped):** the exact spelling the header named as collapsing, `{ type } & { [K in keyof T as K extends 'type' ? never : K]?: T[K] }`, declares 62 properties at `ObjectGridSchema` and 28 at `ButtonSchema`, where the shipped `PartialSchema` declares 1 at each. It refuses `label: 42` on `ButtonSchema` and `objectName: 42` on `ObjectGridSchema` (both `@ts-expect-error` used; tsc exit 0). ## Gates at HEAD `72dba894a` (round 1; round 2's are above) (exit codes captured by redirect-then-capture; heavy runs through the shared verify lock) | gate | result | |---|---| | `@object-ui/types` build (tsc + vite + dist completeness) | exit 0, 134 emitted files verified | | `@object-ui/types` `type-check` (`tsc --noEmit` + examples + `tsconfig.test.json`) | exit 0 | | `pnpm exec vitest run packages/types/` (parity, the family-D / E3 / input pins, the 8072 and 8762 shorthand pins) | 265 files / 5873 tests, exit 0 | | `pnpm exec vitest run scripts/` (every `scripts/__tests__` gate suite, a superset of those whose corpus holds a touched file) | 177 passed + 2 skipped of 179 files / 5322 tests, exit 0 | | downstream consumer type-check: the packages naming `InputSchema` (`components`, `core`) plus those authoring `input` nodes or configs (`cli`, `plugin-dashboard`, `plugin-designer`) and `fields`, `react`, `types`, against the rebuilt `dist` | 8 × `type-check: Done`, exit 0 | | `@object-ui/types` lint (`eslint .`, JSON) | 338 files, 0 errors; the 9 `form.ts` warnings sit outside every added hunk | | `check:handler-key-reads` | exit 0; 128 arms, 50 / 50 reads judged, 0 unjudged | | `check:control-bytes` · `check:new-line-citations` (0 new) · `changeset:check` · `check-changeset-presence` · `check:changeset-claims` · `check:pending-changeset-literals` | exit 0 each | | `check:spec-symbols` · `check:component-surface-parity` · `check:readme-exports` · `check:registry-bare-names` · `check:prompt-keys` · `check:doc-types` · `check:test-path-roots` · `check:esm-specifiers` · `check:vi-mock-*` · `type-check:coverage` | exit 0 each | | `check:doc-snippets` (678 / 678) · `check:doc-examples` · `check:skill-examples` · `check:doc-fences` · `check:doc-example-ids` · `check:sdui-registration-pins` · `pnpm check` (CLI self-check) | exit 0 each | | governed guard `--test` over the 7 paths | NOT GOVERNED (lit control `AGENTS.md`: exit 3) | - **The gate list was derived by hand.** objectstack's `dispatch-gates.mjs` refuses an objectui answer by design (exit 2), so the list comes from objectui's `package.json` and `.github/workflows/`. - `check:changeset-claims` is report-only. It names 18 pending changesets that mention a touched file; each was read, and none is falsified as a record of its own change. The two that carry a claim this slice bears on are in the Acceptance notes. - **NOT MEASURED, left to CI:** the 8-way `pnpm test` shards beyond `packages/types` and `scripts/`, `test:dist`, and E2E. `check:sdui-registration-pins` read the console built at base; this diff touches no registration and no `sideEffects` array. ## Round 2: four prose edits, head `6eb3397b4` These come from contract review `5865087439` ③. The seat took them into this PR because each ships verbatim into the CHANGELOG. They are prose only: no code, no test logic, and every frontmatter block is byte-identical (the `name: bump` lines at base and head compare equal). 1. **`.changeset/9256-input-content-channels.md`, and the Clause-② sentence above.** - Old: "an authored child list on it rendered nothing: no error, no warning, no element." - New: "an authored child list on it rendered nothing, with no render-time error or warning and no element; only the parser tier's `not-a-container` warning (objectui#9910) noticed it." - Measured with a one-shot probe, not committed. It built the manifest from the live registry the way `container-declaration-ratchet.test.tsx` does, then ran `validateTree` and `compile`. - `input`, `ui:input`, `email` and `password` carrying `children`: each draws exactly `not-a-container`. - The same four without `children`: none. - `div` carrying `children` (control): none. - The JSX `input` tag with a child: `not-a-container`, severity `warning`. 2. **`.changeset/6397-partial-schema-collapse-pin.md`.** A dated note follows the "Why a pin and not a repair or a retirement" paragraph: a generic re-spelling of this `Omit` does not collapse, objectui#9256. - Spelling: `OmitDeclared` is spelled exactly as quoted, in `packages/types/src/form.ts`. - Counts: the re-spelled alias declares 28 properties at `ButtonSchema` and 62 at `ObjectGridSchema` on this head and on `main` `733fd5ac6`. The shipped alias declares 1 at each. - "Not true when written": at that entry's own commit (`c8ea8af9c`), under the same TypeScript 6.0.3 its lockfile pins, the same spelling declares 27 and 61. Those are the source counts in that entry's own table. Both re-spellings refuse `label: 42` (the `@ts-expect-error` is used), and a lit control on the shipped alias admits it. - "Later in this same release": the E3 entry and the 6397 entry are both pending on `main` (`git cat-file -e` exit 0 for each; control on an absent path: exit 128). 3. **`.changeset/9256-content-channel-family-d.md`.** A dated note follows the AMENDED paragraph, in that file's own convention, naming the hold-outs this card later narrowed in this release. - Six of the nine names, in `nine-holdouts-six-narrow.md`. - `list` and `timeline`, in `9256-list-timeline-content-channels.md`. This is one more than the reviewer's draft, measured. - `DetailViewSchema` and the shorthand faces, in the E3 entry. - `InputSchema`, in this PR. - A compiler-API probe over the source reads `body` / `children` as `undefined` with their own `?: never` members on `TextSchema`, `ImageSchema`, `IconSchema`, `TabsSchema`, `AccordionSchema`, `CalendarSchema`, `ListSchema`, `TimelineSchema`, `DetailViewSchema`, `InputShorthandSchema` and `InputSchema`. The zod faces are pinned in the family-D, E3 and input pin files. - `ButtonSchema` and `AppComponentSchema` still resolve `children` to `SchemaNode | SchemaNode[]` (`ButtonSchema`'s zod mirror declares its own `children` union; `AppComponentSchema`'s extends `BaseSchema`, which declares one). That is the note's "still accept `children`" sentence. 4. **`.changeset/8499-node-slot-registered-arms.md`.** "`inputType` is deliberately NOT declared on this arm" is false on `main`. - `form.zod.ts` declares `inputType: retirementTombstone(SHORTHAND_INPUT_TYPE_REFUSAL)` on `InputShorthandSchema`, and `form.ts` declares `inputType?: never`. Both files are identical on `main` and at the base. - A runtime probe: `inputType` is a member of the mirror shape. `{ type: 'password', inputType: 'text' }` is refused with `invalid_type` at path `inputType`, the message opening with the objectui#8762 guidance. Bare `{ type: 'password' }` (control) parses. - objectui#8762's entry landed a day after 8499's (2026-09-10 against 2026-09-09), and both are pending on `main`. - A dated note in the file's own form follows the `InputShorthandSchema` bullet, the same place its objectui#10756 note sits under the `HtmlElementSchema` bullet. | gate at `6eb3397b4` | result | |---|---| | `check-changeset-overwrite` | exit 0 (report-only), "1 changeset(s) added, 3 modified, 0 deleted". Each of the three modified entries prints the same declaration at base and now (`@object-ui/types: patch`, `minor`, `minor`), so no declaration is lost: its case 2, a correction on purpose. | | `changeset:check` · `check-changeset-presence` · `check:changeset-claims` · `check:pending-changeset-literals` · `check:control-bytes` · `check:new-line-citations` | exit 0 each | | the 24 `scripts/__tests__` suites that name `.changeset` (the changeset gates' own suites, the polarity census, control bytes, pending literals, docs and workflow suites) | 24 files / 913 tests, exit 0 | ## Round 3: two more prose edits, head `876f6d7ea` These come from delta review `5865466392` ③ flags 2 and 3. They are prose only, and both frontmatter blocks are byte-identical. 1. **`.changeset/9256-content-channel-family-d.md`: a second dated note, directly after the AMENDED paragraph** and before round 2's note. The headline is "the two twins no longer take `body` — objectui#6771". - Old, still in the AMENDED paragraph: "The two twins are unchanged and go on inheriting `body` as the content slot". - New note: objectui#6771 retired `body` on `BaseSchema` itself, so both twins refuse `body` on both faces. The TypeScript face does it through `BaseSchema`'s `body?: never`; each mirror does it by name, pointing at `requestBody`. - Measured, TypeScript: a compiler-API probe over the source gives `BaseSchema` its own `body?: never`. `ChatbotEnhancedSchema` and `ChatbotFloatingSchema` resolve `body` to `undefined` and declare only `children?: never` themselves. - Measured, zod, with a one-shot runtime probe (deleted): each twin's mirror refuses `body` with `invalid_type` at path `body`, and the message names `requestBody`. `requestBody` parses (control). - Provenance, from `git log -S`: `BaseSchema`'s `body?: never` and both twins' mirror tombstones were introduced by `2acd8e109` (objectui#6771, 2026-09-19). The AMENDED paragraph was written on 2026-09-17 (`c42554e94`), and `6771-retire-body-child-list-dialect.md` is pending on `main`. So "later in this same release" holds. - The reviewer's draft also cited objectui#9659. That card did not introduce the twin refusals, so the note cites objectui#6771 alone. - Round 2's note opens "most of the hold-outs above have since been narrowed". It is still true beside the new one: it names what this card narrowed, and it ends with `button` and `AppComponentSchema`, which still accept `children`. 2. **`.changeset/9256-e3-residual-content-channels.md`, in place.** - Old: "an authored child list on them rendered nothing: no error, no warning, no element." - New: "an authored child list on them rendered nothing, with no render-time error or warning and no element; only the parser tier's `not-a-container` warning (objectui#9910) noticed it." - Measured for all twelve keys, not just the two in round 2. A one-shot runtime probe (deleted) loaded every plugin registration (`plugin-grid`, `-form`, `-kanban`, `-map`, `-tree`, `-view`, `-gantt`, `-calendar`, `-charts`, `-detail`, and `components`). It built the manifest from the live registry and ran `validateTree`. - Each of the twelve is known to the registry, and none declares a `children` input. - Carrying `children`, each draws `not-a-container` (severity `warning`). Without it, none does. `div` carrying `children` (control) draws nothing. - `object-chart`, `email` and `password` also draw one `unknown-prop` in both legs, from the probe's own `objectName` fixture key. It is not part of the difference. | gate at `876f6d7ea` | result | |---|---| | `check-changeset-overwrite` | exit 0 (report-only), "1 changeset(s) added, 4 modified, 0 deleted". Each modified entry prints the same declaration at base and now, so no declaration is lost. | | `changeset:check` · `check-changeset-presence` · `check:changeset-claims` · `check:pending-changeset-literals` · `check:control-bytes` · `check:new-line-citations` | exit 0 each | | the 24 `scripts/__tests__` suites that name `.changeset` | 24 files / 913 tests, exit 0 | ## Serial constraints - **Re-mapped at branch time (`244d516df`), before each round's push, and again before round 3's edits.** At round 3 the open PRs are objectui#10921, objectstack-ai#10915, objectstack-ai#10914, objectstack-ai#10912, objectstack-ai#10911, objectstack-ai#10910, objectstack-ai#10908, objectstack-ai#10906, objectstack-ai#10901, objectstack-ai#10777 and objectstack-ai#10278. objectstack-ai#10904, objectstack-ai#10902 and objectstack-ai#10891 have merged since the first map. - None touches any of the five changesets this PR adds or edits. - One touches a source path here: PR objectui#10914 edits `packages/types/src/zod/form.zod.ts`, in two comment / `.describe()` hunks on the `wrapperClass` key (in `InputSchema` and in the `InputShorthandSchema` restatement note). Neither overlaps or touches this PR's hunks, and `git merge-tree` of this head (`876f6d7ea`) against objectstack-ai#10914's head is clean. - PR objectui#10908 edits `zod-mirror-parity.test.ts` and `zod/index.zod.ts`, neither of which this PR touches. - **The release PR objectui#5400** (1732 files, enumerated to the end, last updated 2026-09-18) touches none of the source paths. As release consumption it deletes pending changesets, three of the four edited here among them (not the E3 entry); that is not a hunk overlap. - **`origin/main` moved four commits since the base** (`9f0c84a44`, `7ea8118f7`, `8522396c0`, `733fd5ac6`). None touches `packages/types` or the changesets edited here, and `git merge-tree` of this head (`876f6d7ea`) against `733fd5ac6` is clean. `main` has not moved since. ## Acceptance notes: out of scope, not fixed here - **The `PartialSchema` docblock and its pin file still sequence the alias's repair on objectui#5155.** That card is closed as a duplicate of objectui#5250, whose ruling keeps the tolerant face on renderer props. objectui#8347 (open) is the card on removing the index signature from the authoring face. "Once objectui#5155 removes the root index signature" therefore names a card that will not do it. Noted, not edited: rewriting the sequencing is a judgement about objectui#5250 / objectstack-ai#8347, not a comment correction. Carrier: none. - **The same "no error, no warning, no element" sentence lives in published refusal messages and docblocks,** not only in changesets. Examples: the family-D `retirementTombstone` strings on the zod mirrors (the chatbot twins' included) and the matching TypeScript docblocks. Wherever the node's registration declares no `children` input, the parser tier's `not-a-container` warning did fire. Correcting those strings is a published-message change, not prose, and it is not this PR's. Noted, not edited. Carrier: none. --- _Generated by [Claude Code](https://claude.ai/code/session_01DuWo5bdP9SdVebamn99GGk)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #8524
Clause-②: no
What was wrong
After objectstack-ai/cloud#2144 the tenant runtime's
GET /api/v1/ai/usageanswers{ pool, breakdown }: one AI quota pool, plus an observation-only split of that same pool.useAiUsageonly parsed the retired per-meter{ meters: { build, dataChat } }answer, so its parser returnednullandAiUsageIndicatorrendered nothing. No ring, no error.Producer read at cloud
main96eb092,packages/objectos-runtime/src/ai-usage-endpoint-plugin.ts(symbolsAiUsageIndicator,AiUsageBreakdownFractions,AiUsageResponse,buildAiUsageResponse,unknownResponse):poolcarriesplanType,fraction(number or null),unmetered,resetKind,resetsAt,upgrade,topUp. These are the same fields each old meter carried.breakdownis always an object whose members are number or null. Every error path answers 200 with a neutral pool (fraction: null) and both membersnull.build/dataChatsplit read-only statistics that no quota decision uses. cloud#2133 lets the endpoint keepbreakdownbut does not require it, so its absence is a normal reading. The triage comment on the card says the same.What changed
useAiUsagereads{ pool, breakdown? }and only that shape.poolmust carry every declared field with its primitive type.resetKindandplanTypeare checked as strings, so a reset kind this build does not know still renders without a reset line (the objectui#7371 behaviour, still pinned).breakdown, or one withnullmembers, is a normal reading and never the parse-failure branch.{ meters }one included, givesusage: nulland now also setserror. There is no dual-shape read.AiUsageIndicatordraws ONE ring frompool.Build 30%andAsk 20%, each a percentage of the same pool. It is shown only when at least one member is a number, and it is never a second ring.ai-usage-cta.console.ai.usage.breakdownTitle, in every locale pack. The split rows reusemeterBuild/meterAsk..changeset/8524-ai-usage-single-pool.mddeclares a patch for@object-ui/app-shelland@object-ui/i18n.File surface is the claim's: the hook, the component, their two test files, the ten locale packs and one changeset. The hooks barrel is untouched (see Acceptance notes).
Tests
packages/app-shell/src/hooks/__tests__/useAiUsage.test.ts:{ pool, breakdown };breakdownabsent parses to{ pool }with no error;breakdownwith both membersnullparses as a normal reading;{ meters }answer givesusage: nullwitherrorset (the strict reader, pinned);breakdownmember that is not a fraction givesusage: null.packages/app-shell/src/layout/__tests__/AiUsageIndicator.test.tsx:nullgives one ring and no split;breakdownabsent gives one ring and no split;{ pool }answer with nobreakdownrenders the ring, and the{ meters }answer renders nothing once the hook reports its error.Reverse verification (after committing the fix; mutation and restore scripted with a trap): restoring the base versions of
useAiUsage.tsandAiUsageIndicator.tsxunder the new tests givesTest Files 2 failed (2),Tests 20 failed | 4 passed (24). The 4 that stay green are the shape-independent cases: null usage renders nothing, non-2xx, inert without apiBase, and refetch on nudge. The restore was proven by HEAD blob hashes and an emptygit diff HEAD.Local verification
Final reading at
12e1c1d. That head is this branch after mergingorigin/mainthree times; the merges brought no change to the files this PR touches. Each exit code was captured to a file before any pipe. The full app-shell suite ran once, atfd6907c(the fix commit):Test Files 843 passed | 1 skipped (844),Tests 8673 passed | 9 skipped. It was not re-run at the final head, because CI runs it.12e1c1d)node scripts/check-changeset-presence.mjspnpm changeset:check(fixed group + no-major)pnpm check:i18n-keyspnpm check:i18n-driftpnpm check:i18n-dead-keyspnpm check:control-bytespnpm check:new-line-citations(VERDICT 0 new citations)pnpm check:changeset-claimspnpm check:pending-changeset-literalspnpm check:vi-mock-specifiers/vi-mock-inherit/vi-mock-override-shapepnpm check:test-path-rootspnpm check:unreferenced-sources/phantom-deps/self-importturbo run build --filter=@object-ui/app-shell^...(28/28, all cache hits) thenpnpm --filter @object-ui/app-shell run type-check(verify lock)vitest runonuseAiUsage.test.ts+AiUsageIndicator.test.tsx(verify lock): 2 files, 24 tests passedeslint --format jsonon the 14.ts/.tsxfiles in this diff: 0 errors, 2 warnings that were already therepnpm check:eager-locale-cataloguesAlso run at
a5d7c25, after the first merge (which touched the locale packs): thepackages/i18n/suite plus the two touched tests plus the two ChatDock tests gave 76 files and 1234 tests passed, exit 0.tsc -p tsconfig.test.json --listFilesOnlyconfirms the package type-check compiles both touched test files.Lint was narrowed to the diff's
.ts/.tsxfiles. The narrowing is safe becauseeslint.config.jssets noparserOptions/project/projectService/ type-checked preset (zero hits, with a lit control), has noimport/*plugin rules, and none ofeslint-rules/*reads another file. Every rule therefore judges one file, and this diff cannot move a verdict on an untouched file. The count comes from--format json: 14 files, 0 errors, 2 warnings. Both warnings arereact-hooks/set-state-in-effecton effects this diff does not change, and linting the base blobs of those two files shows the same two warnings.NOT MEASURED:
check:eager-locale-catalogues: PREREQUISITE NOT MET, because it needs a built console bundle (apps/console/dist/eager-closure.json). CI builds it.Acceptance notes
titletooltip. The card allows "a tooltip or a detail view", and atitletooltip never shows on touch devices.breakdown, decided by the PM on the dev report (a PM decision, not a maintainer ruling: option A, keep strict): abreakdownthat is present but malformed is rejected like any other unreadable answer.erroris set and the ring hides. "Malformed" means not an object, a member that is neither a number nor null, or a literalbreakdown: null, which the producer never sends. The alternative, dropping a malformed split and keeping the ring, would be exactly the tolerant read the card's routing constraint forbids, and it would hide a future rename on the producer side. Only an absentbreakdown, or one with null members, means "no split".erroron an unreadable answer: the hook now setserrorwhen an answer does not parse. Before, it returnedusage: nullwith no error. No UI readserrortoday.AiMeterUsagekeeps its name and now documents the pool. The newAiUsageBreakdownlives in the hook file. Neither change touchespackages/app-shell/src/hooks/index.ts, which is outside the claimed surface. The package entrysrc/index.tsre-exports none of these types, henceClause-②: no..objectui-sha(the overlay layer inapps/objectos/Dockerfile). So this reader ships with a runtime that already serves{ pool, breakdown }. The one path where it could meet a{ meters }runtime is a hotfix branch that moves the pin without cloud#2144. That path is noted here and deliberately not answered with a dual read.Session:
https://claude.ai/code/session_015AUunPkX7UTkCH9e7AdZo1Generated by Claude Code