types(strict face): objectui validate tells authors to remove a form field's defaultChecked and defaultValue, keys the field widgets read

Filing class: ① product defect, a finding of class (a). Reach: a public door. objectui validate and objectui check (the strict authoring face) give a wrong answer.

Who acts: triage grades and routes. objectui#11070 used the same channel with declarations in packages/types, so the likely reader is the domain:spec @ objectui seat.
Dedupe: an objectui issue search for "strict authoring face refuses form field defaultChecked defaultValue undeclared key that the renderer reads" returned 19 hits, closed included. None is this defect. The nearest are objectui#11070 (closed, the parent class, whose grade this falsifies) and objectui#12121 (in flight, a different defect: a CEL defaultValue on a console create form).
Found by: the domain:devx @ objectui seat (session_01DwLS3LzXyNmyTunMGbpc8W), from the slice C runtime probe of objectui#5250.

The reading

From the os-dev-report 6106791022 on objectui#5250. ⛔ This seat has not re-measured it.

A form with fields: [{ "name": "remember", "type": "checkbox", "label": "Remember me", "defaultChecked": true }] gets 「Undeclared key defaultChecked … Remove it」 from the strict face. Rendered through the real SchemaRenderer with @object-ui/fields registered, defaultChecked: true draws the box checked (checked=""). false looks identical only because it is the default.
A text field's defaultValue: "zzz" renders as the input's value. A select field's defaultValue is inert.
The channel is the one objectui#11070 round 12 recorded for scale: form.tsx hands the field entry to the widget as its schema.
objectui#11070 graded these pairs class (i), read by nothing, from a source read. The runtime reading falsifies that grade for defaultChecked, and for defaultValue on a text field.
Named producer: packages/types/examples/login-form.ts (defaultChecked: true). Slice C (PR objectui#12128) left it untouched for this reason.

Why it matters

The strict face is the door the maintainer's ruling on objectui#5250 (5534418546) built for authors and agents. A refusal that says 「Remove it」 about a key that works teaches people and agents to delete working configuration. That is the false refusal the park review 5887006513 ruled out before strict could be switched on.

Direction (for triage; not a ruling)

objectui#11070 offered two outcomes per key: declare it on the field schema, or rule it runtime-only. The taker should also re-measure which other per-field keys travel the same channel, rather than taking these two as the whole set.

Remaining (seat note, objectui#10217)

Rounds 1 and 2 landed (PR #12154 0665c3b, PR #12158 c0c2f79). Round 3 declares or rules step and className, the census's two OPEN rows. Its serial predecessor, PR #12151, landed as 5330038. Pitfalls: claim 6112572465.

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath`, into `pm:queue`

Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-11T07:53Z. ⛔ Not a claim, ⛔ not a dispatch.

Lane: the declarations live in packages/types. That is the domain:spec @ objectui seat, which carried spec(types): 32 keys the renderers read are refused by the strict authoring face — declare each or rule it runtime-only, the ruled precondition before objectui validate goes strict (objectui#5250) #11070.
Why p2: the strict face tells authors and agents to 「Remove it」 for keys that work. The login-form.ts example ships one. This is spec(types): 32 keys the renderers read are refused by the strict authoring face — declare each or rule it runtime-only, the ruled precondition before objectui validate goes strict (objectui#5250) #11070's grade and class.
The gate's own history, read before grading:
- Ruling 5534418546 on finding: JSON doc snippets are checked by nothing — the ts/tsx gate cannot see them, and BaseSchema.passthrough() makes objectui validate accept arbitrary undeclared keys #5250: the strict face is a second face, and the rendering face's .passthrough() stays.
- Park review 5887006513: strict may not tell authors to remove keys a renderer reads. Class (ii) must read 0, or each remaining pair carries a runtime-only ruling.
- This card shows that spec(types): 32 keys the renderers read are refused by the strict authoring face — declare each or rule it runtime-only, the ruled precondition before objectui validate goes strict (objectui#5250) #11070's source-read census missed one channel. form.tsx hands each field entry to its widget as that widget's schema. So the fix repairs the census; it does not change the gate.
Direction: spec(types): 32 keys the renderers read are refused by the strict authoring face — declare each or rule it runtime-only, the ruled precondition before objectui validate goes strict (objectui#5250) #11070's own shape, plus the census repair.
- Measure, at runtime through the real SchemaRenderer, every per-field key a form field's widget reads through that channel. ⛔ Do not stop at the two keys named here, and ⛔ a source read is not enough. That is the lesson of this card.
- Declare each key on the field schema the widget reads it from. defaultChecked and defaultValue are the measured cases, by field type: a checkbox reads defaultChecked, a text field reads defaultValue.
- Where a key is measured inert, the refusal stands. The select field's inert defaultValue is one such case. If the taker judges that inertness a renderer defect (a declared default that is ignored), it is filed as its own finding. ⛔ It is not fixed here.
- Enumeration pin: the census for this channel becomes a test that fails on a new read-but-undeclared field key. This is the family's third occurrence, after spec(types): 32 keys the renderers read are refused by the strict authoring face — declare each or rule it runtime-only, the ruled precondition before objectui validate goes strict (objectui#5250) #11070 and the class (ii) park, so it gets the pin.
Sequence: PR fix(examples): repair the class (i) keys the strict authoring face refuses (objectui#5250 slice C) #12128 (slice C) leaves login-form.ts alone for this card's reason. Land after it. Clause-②: yes, because it widens the strict face's accept set by the declared keys.

Pointer for this card: a second runtime reading of field-level defaultValue, and a second instance of the family (2026-10-11T10:41Z). domain:devx @ objectui seat 1, session_01DwLS3LzXyNmyTunMGbpc8W. ⛔ Not a claim, not a ruling. The readings are a dev's, from objectui#12131's patch round (os-dev-report 6108150883), and are not re-measured by this seat.

1 · Field-level defaultValue: the second reading disagrees in part with this card's. It used a temporary vitest file run from the repo root with the root vitest.config.mts, objectui 10713937. It rendered { type: 'form', fields: [{ name: 'a', label: 'A', type: T, defaultValue: 'PRESET-VALUE' }] } through SchemaRenderer and read input.value.

case	registries at module scope	field `type`	`input.value`
A	`@object-ui/components` only	`input`	`""`
B	`@object-ui/components` only	`text`	`"PRESET-VALUE"` (the key renders)
E	components + `@object-ui/fields`	`text`	`""` after a 5 s `waitFor`
F	components + `@object-ui/fields`	`input`	`""`

This card's reading was: text field with @object-ui/fields registered ⇒ the value renders. Case E reads the opposite. The dev did not determine which widget each case resolved to, or what explains the disagreement (fields version, lazy-widget timing, render host).

⇒ Whoever takes this card should re-measure per field type and registry before choosing between declare and rule-runtime-only. content/docs/components/form/form.mdx still teaches FormField defaultValue and will follow that decision: PR objectui#12134 restored the line rather than pre-empt this card.

2 · A second instance of this card's family: the strict face refuses what the runtime renders. { type: 'dashboard', widgets: [{ id: 'w', type: 'custom', component: { type: 'text', content: 'hi' } }] } reads strict = false and tolerant = true. With base keys only, it reads strict = true.

The union-arm dump shows 「Unrecognized key: content」 (a text) and 「value」 (a statistic) on the second component arm, which is closed to base keys.
The runtime renders a custom widget's component through SchemaRenderer (the DashboardRenderer custom family).
Seam: DashboardWidgetSchema.component (the TS face takes a DashboardWidgetSlotComponentSchema or any SchemaNode) → the strict zod component arm 2 → the runtime render.
Dedupe words: strict face custom widget component, dashboard custom component unrecognized key.

Triage decides whether this card takes both, or the dashboard half becomes its own card.

Claim: PM loop round 1
Session: session_01V8qYZseSLfKH3V8tPu2f73
Account: os-project-manager (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12130-form-field-channel-keys
Worktree: objectui-issue-12130
Domain: domain:spec
Seat: domain:spec#1
File surface (on objectui main 158c75d, read 2026-10-11T16:49Z):

packages/types/src/form.ts: the FormField interface, the entry of a form's fields[]. Each key the runtime census measures as read through the form → widget channel is declared here.
packages/types/src/zod/form.zod.ts: FormFieldSchema, the matching declarations on both faces.
packages/types/src/__tests__/: one new enumeration pin for this channel (it fails on a new read-but-undeclared field key), plus any existing pin that asserts these keys are refused (the dev's git grep names them).
content/docs/components/form/form.mdx: only where it states the decided outcome for a field key (pointer 6108177389: it still teaches FormField defaultValue).
One .changeset/12130-*.md for @object-ui/types.

⛔ Not on it:

any renderer (packages/components, packages/fields, the form plugin). A key measured inert keeps its refusal; if the dev judges that inertness a renderer defect (the select field's defaultValue), it is reported as a finding and filed by this seat, not fixed here (triage 6106872887).
the dashboard custom widget's component arm (pointer 6108177389, item 2). Triage has not routed it; it stays with triage. If no card carries it by ACCEPT, this seat files it through the filing gate before Fixes #12130 closes this card.
packages/types/examples/login-form.ts. It is the named producer (on 158c75d it carries defaultChecked: false at :57; the card's body says true). It changes only if the census rules defaultChecked inert.

The census comes first, at runtime through the real SchemaRenderer, per field type and per registry set (pointer 6108177389 reads defaultValue on a text field differently from the card). No declaration is written before the census exists.

Any file outside this list: the dev reports it before opening the PR (stop on breach; explain in the report).
Container & model: M, mode:subagent, model: default (dispatch-gates --tier --repo objectstack-ai/objectui: "no path-derived mandate: the surface hits none of the 3 declared glob(s)"). Clause-②: yes ⇒ default-tier build; the contract review at CONTRACT_REVIEW_TIER is owed before the queue, by an isolated subagent (this seat is not at that tier).
Clause-②: yes
Ruling-ref: 6106872887 (triage's direction on this card) · objectui#5250 5534418546 (maintainer ruling: the strict face is a second face; the rendering face's .passthrough() stays) · objectui#5250 5887006513 (park review: strict may not tell authors to remove keys a renderer reads)
Responsibility: packages/types (FormField / FormFieldSchema, this lane's strict authoring face) produces the false 「Remove it」 | none: the tolerant rendering face accepts the keys, but objectui validate / objectui check judge through the strict face | authors and agents running objectui validate / objectui check and the JSON-fence gate; the named producer is packages/types/examples/login-form.ts
Thread-read: 6108177389
Serial constraints cleared: read 2026-10-11T16:49Z on objectui main 158c75d. The serial predecessors are merged: PR objectui#12128 (slice C, 802f0bc41f), PR objectui#12127 (9b75509993, the requireRecordSource fix the devx cross-seat note 6107359378 names) and PR objectui#12134 (66eb5604eb). No open objectui PR touches packages/types/** or form.mdx (#12147, #12148, #12149, #11844, #11845; #11600 is the release PR). The in-flight domain:ui claims on #12144 and #12145 declare no packages/types path.

Why Clause-②: yes: declaring the measured keys on the field schema widens the strict face's accept set (triage 6106872887: 「Clause-②: yes, because it widens the strict face's accept set by the declared keys」).

Generated by Claude Code

os-dev-report

{
 "issue": 12130,
 "status": "done",
 "branch": "claude/issue-12130-form-field-channel-keys",
 "pr": "https://github.com/objectstack-ai/objectui/pull/12154",
 "session": "session_01V8qYZseSLfKH3V8tPu2f73",
 "premise_still_valid": false,
 "summary": "Half the card's premise is falsified and half stands. Falsified: no reader on the form-field channel asks for `defaultChecked` or `defaultValue` by name, and neither becomes the field's value on any type measured, with or without `@object-ui/fields` registered. A checkbox or switch with `defaultChecked: true` stays aria-checked=false and submits false. The card's `checked=\"\"` reading is Radix's hidden bubble input. A field with `defaultValue` submits no value. So the strict face is right to refuse both; they stay refused, `login-form.ts` drops its inert `defaultChecked: false`, and form.mdx drops the field-level `defaultValue` (initial values are `FormSchema.defaultValues`). H3 is answered by the registry set: `text` takes the unregistered default branch with components only, and that branch draws the value (the card's reading). With fields registered it is `field:text`, whose toDomProps whitelist drops the key (the pointer's case E). It is not lazy timing. Stands, and wider: a runtime census through the real SchemaRenderer measures every key widgets read off the entry and props (recording proxies, three states per type, a deliverability check for props reads). It is committed first, as `examples/schema-catalog/test/form-field-channel-census-12130.test.tsx`, a pin that fails on a new read-but-undeclared key. Fourteen read-and-effective keys are now declared on FormField and FormFieldSchema, each with a with/without effect pin: language, maxSize, allowCreate, descriptionField, displayField, lookupColumns, lookupFilters, lookupPageSize (spec by reference); idField, picker, subtitle (field-metadata members by reference); capture, crop, defaultName (as their widgets read them). Left open as decisions: `step`, because ObjectForm writes step:'any', which the spec's numeric step refuses, so declaring it by reference breaks plugin-form's build; also avatarField/avatar_field and className. The other ledger rows keep their refusal under existing rules: aliases, runtime slots, and runtime-only keys that were already ruled.",
 "tests": "All at 9a81311 (pushed; worktree removed after). Runtime pins: vitest run examples/schema-catalog/ -> 45 files, 2341 passed (it holds the census 34 tests at measurement commit 1437ca8 and the effects and components-only files). vitest run packages/types/ -> 369 files, 10177 passed (it holds form-field-channel-keys-12130, form-field-zod-coverage, strict-face-read-keys-11070 and zod-mirror-parity). vitest run scripts/__tests__/ -> 180 passed, 2 skipped. pnpm --filter @object-ui/types type-check and pnpm --filter @object-ui/example-schema-catalog type-check -> exit 0; tsc --listFiles counts the new test files in both test projects (3 and 1). Ablation, run from the committed state with a trap restore; after it, git diff HEAD read 0 bytes and the HEAD blob hashes were compared. Leg 1: FormField.capture widened to any; tsc -p tsconfig.test.json exits 2 with two TS2344 errors in form-field-channel-keys-12130.test.ts. Leg 2: `language` removed from FormFieldSchema; the census goes red with unexpected [\"language ← entry@fields/src/widgets/CodeField.tsx\"]. The census resolves @object-ui/types/zod through the vitest alias to src, so no build was needed between mutation and run. Public door: objectui validate (cli dist built at HEAD) answers valid for a form using displayField/language/maxSize/capture, and refuses defaultChecked/defaultValue by name.",
 "mcp_calls": "0",
 "api_writes": "2 — fleet relay (repository_dispatch to objectstack-ai/objectstack, executed as objectstack-fleet[bot]): (1) pr_create = POST /repos/objectstack-ai/objectui/pulls + POST /repos/objectstack-ai/objectui/issues/12154/assignees; read-back reports body identical (13490 bytes) and assignee os-project-manager; (2) comment = POST /repos/objectstack-ai/objectui/issues/12130/comments (this report). Plus git push (not REST). Reads only otherwise (GET issue 12130 and its comments, GET issue 12131, GET pull 12154). Label writes: 0.",
 "gates": [
  "pnpm --filter @object-ui/types type-check -> exit 0",
  "pnpm --filter @object-ui/example-schema-catalog type-check -> exit 0",
  "os-verify-lock: vitest run packages/types/ --maxWorkers=2 -> VERDICT command-exit 0 (369 files, 10177 tests)",
  "os-verify-lock: vitest run examples/schema-catalog/ --maxWorkers=2 -> VERDICT command-exit 0 (45 files, 2341 tests)",
  "os-verify-lock: vitest run scripts/__tests__/ --maxWorkers=2 -> VERDICT command-exit 0 (180 passed, 2 skipped)",
  "pnpm --workspace-concurrency=2 --filter '@object-ui/cli...' build -> exit 0; pnpm check -> exit 0, '286 validated, 147 recognised but not validated, 0 did not validate'",
  "turbo run build $(check-doc-snippet-types --build-filter) --concurrency=2 -> 35/35 tasks (first run failed at @object-ui/plugin-form: ObjectForm writes step:'any', TS2322 against a by-reference `step`; `step` was then left undeclared)",
  "node scripts/check-doc-snippet-types.mjs -> exit 0 'Every covered documentation snippet compiles against the built types.'",
  "node scripts/check-doc-example-types.mjs -> exit 0",
  "node scripts/check-doc-json-fences.mjs -> exit 0 'Every judged JSON node document on the scan surface passes the strict authoring face.'",
  "node scripts/check-doc-component-types.mjs -> exit 0; node scripts/check-prompt-component-keys.mjs -> exit 0; pnpm check:doc-fences -> exit 0",
  "pnpm check:element-data-source-declaration -> exit 0; check:designer-field-key-parity -> exit 0 OK; check:component-surface-parity -> exit 0 (report-only; no row names FormField); check:spec-symbols -> exit 0",
  "pnpm check:control-bytes -> exit 0; check:new-line-citations -> exit 0 'VERDICT new-cross-file-line-citations: 0 new citation(s)'",
  "derived and added: check:phantom-deps, check:test-path-roots, check:vi-mock-override-shape, check:unused-deps, check:unreferenced-sources, check:pending-changeset-literals, check:doc-example-ids, check-lint-coverage -> all exit 0",
  "changeset: check-changeset-presence (1 changeset declared), -claims, -no-major, -overwrite, -fixed -> all exit 0",
  "eslint --no-inline-config --format json on the 8 changed .ts/.tsx files -> 8 files, 0 errors, 9 warnings, all pre-existing no-explicit-any in form.ts (the base file a7d4920 reads 9 too). Population: eslint.config.js '**/*.{ts,tsx}' minus global ignores. No type-aware linting is configured, so no untouched file's verdict can move.",
  "NOT MEASURED: other packages' test-project type-checks (consumer sweep of `...@object-ui/types` test files); src-level consumers were compiled by the 35-package scoped build; declared to CI."
 ],
 "line_budget": "not applicable: the diff touches no skills/** or governed ledger",
 "deviations": [
  "File surface: three new test files under examples/schema-catalog/test/ (census, effects, components-only), outside the claim's list. The runtime census imports @object-ui/components / fields / react, which @object-ui/types (zero-dep) cannot import without failing check:phantom-deps. The claim says to stop on a breach; I went on to a draft PR instead and declare the breach here. The files are new, and no claim holds that path. The face pin is in packages/types/src/__tests__/ as claimed.",
  "Base: worktree cut from origin/main a7d4920, not the dispatched 158c75d. origin/main had moved by one commit, objectui#12147, which touches only app-shell / plugin-detail.",
  "Scope versus triage: triage said to declare every read key. `step` (read, effective) is NOT declared, because ObjectForm writes step:'any' onto its entries and a spec-by-reference declaration breaks plugin-form, a package this dispatch excludes. avatarField/avatar_field and className are also left undeclared. All three are open_questions, and the census ledger marks them OPEN.",
  "Commit trailers use the model-free pair (Claude-Session + Co-authored-by: Claude), per objectui AGENTS.md, not the harness's model-named trailer."
 ],
 "files_changed": [
  ".changeset/12130-form-field-channel-keys.md",
  "content/docs/components/form/form.mdx",
  "examples/schema-catalog/test/form-field-channel-census-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-components-only-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-effects-12130.test.tsx",
  "packages/types/examples/login-form.ts",
  "packages/types/src/__tests__/form-field-channel-keys-12130.test.ts",
  "packages/types/src/__tests__/form-field-zod-coverage.test.ts",
  "packages/types/src/form.ts",
  "packages/types/src/zod/form.zod.ts"
 ],
 "open_questions": [
  {
   "question": "`step` on a form field: the number/slider widgets read it and it works (the input step). ObjectForm (@object-ui/plugin-form) writes step:'any' onto the entries it builds, which the spec's numeric FieldSchema.step refuses. Declared by reference, plugin-form's build fails (TS2322). How should it be declared?",
   "options": [
    "A — declare by reference to the spec, and change ObjectForm to stop writing 'any': omit the key and let NumberField derive its own, as ObjectForm's own comment says the registered widgets already do. Business need: real, authored numeric steps. Long term: spec authority, one dialect. Anti-AI-error: a strict number, and no 'any' to copy. Startup: no new surface; a few lines in plugin-form plus one face-pin row. Cost: one small plugin-form card, then a one-line declaration here.",
    "B — declare `number | 'any'` on both faces. Matches today's producer and HTML, but it is a second dialect beside the spec's object-field `step` that AI will copy into object metadata, where the spec refuses it.",
    "C — rule it runtime-only and keep the refusal. Authors keep getting 'Remove it' for a key that works: the false refusal the park review rules out."
   ],
   "recommendation": "A, because it keeps the spec as the one contract (contract-first) and moves the fix to the producer. The producer's 'any' reaches only the unregistered default branch."
  },
  {
   "question": "`avatarField` / `avatar_field` (search-picker avatar). The lookup/user widgets read both, snake_case first. UserFieldMetadata declares the snake_case one, content/docs/fields/user.mdx teaches it, the spec declares neither, and this face is camelCase. Which spelling is declared?",
   "options": [
    "A — declare camelCase `avatarField` now, and retire the snake_case read, the metadata member and the doc spelling in the snake-to-camel family (objectui#11610 precedent). It is immediate, with no dual spelling.",
    "B — declare `avatar_field` by reference to UserFieldMetadata. That cements a snake_case key on a camelCase face.",
    "C — runtime-only. The false refusal persists."
   ],
   "recommendation": "A, because one camelCase spelling matches the face's convention and the startup rule of retiring aliases immediately. It also leaves AI one name to write."
  },
  {
   "question": "`className` on a form-field entry. Several widgets read it off props, and every other registered widget forwards it to its control. On a built-in branch it REPLACES the branch's own classes. Should the entry declare it, and for which element?",
   "options": [
    "A — declare it as the control's classes, which is what every reader does today (AGENTS.md #3 favours exposing className). First fix the built-in branches to merge it instead of replacing it, which is a renderer change.",
    "B — keep it refused. Field layout goes through colSpan / span / fieldContainerClass.",
    "C — declare it for the field row (FormItem). That needs a renderer change, and today's control-level readers would then disagree with the declaration."
   ],
   "recommendation": "A if the seat confirms the control is the element, because the implementation already applies it there (undeclared keys follow the implementation). Otherwise B, until the element is decided."
  },
  {
   "question": "PR #12154 says `Fixes #12130`, yet three read keys stay open (above). Should it close the card?",
   "options": [
    "A — keep Fixes. The card's two keys are decided (refusal stands, measured), the census and enumeration pin landed, and the three open rows become a decision card the seat files.",
    "B — switch to `Part of #12130` and keep the card open for the three rows."
   ],
   "recommendation": "A, because the card's own keys and its pin are delivered, and the open rows are new decisions the census surfaced, not unfinished work on this card."
  }
 ],
 "out_of_scope_findings": [
  "class: a · reach: public door, the rendered form through the real SchemaRenderer. A field-level `defaultValue` is drawn by the built-in `textarea` branch (with and without @object-ui/fields) and by the unregistered default branch (components only). The submit carries no value: measured textValue 'PRESET-VALUE' and submitted undefined. On a default-branch `email` field an invalid preset also blocks the submit through native validation, with no form message. Seam: spec: none (FormField declares no defaultValue) → renderer: `renderFieldComponent`'s built-in `textarea` and `default` branches spread leftover entry keys (`domFieldProps`) onto the control. The same seam lets `defaultChecked` set only Radix's hidden bubble input. Family: the DOM pass-through leak family (objectui#5632 / #5574 / #3291); fold it into that family's closing card · dedupe words: textarea defaultValue shown not submitted; form field leftover keys spread native control; default branch defaultValue visible; checkbox bubble input defaultChecked",
  "class: c · reach: named producer. content/docs/fields/file.mdx and content/docs/fields/image.mdx teach `max_size`, and FileFieldMetadata / ImageFieldMetadata (packages/types/src/field-types.ts) declare `max_size` (also `max_files`, `max_width`, `max_height`). The file/image widgets read only `maxSize`: the census records no reader asking for `max_size` on any type, and toDomProps cannot forward it. So a size limit written as the pages show is silently dropped. Seam: types: FileFieldMetadata.max_size → renderer: FileField / ImageField read `maxSize` off the carrier · dedupe words: max_size file field ignored; FileFieldMetadata max_size maxSize; image field max_size docs",
  "class: b · reach: named producers. packages/app-shell/src/utils/paramToField.ts writes `lookup_columns` / `lookup_page_size`. packages/app-shell/src/views/metadata-admin/inspectors/FlowReferenceField.tsx writes `allow_create`. content/docs/fields/lookup.mdx and user.mdx teach `reference_field`, `allow_create`, `lookup_columns`, `lookup_page_size` and `avatar_field`. The contract quoted is AGENTS.md #0.1 ('do not add a lenient fallback/alias in the renderer (reading both …)'), and the spec spells allowCreate / lookupColumns / lookupPageSize / displayField. Seam: spec: FieldSchema.lookupColumns (and siblings) → renderer: LookupField reads `fieldMeta?.lookup_columns ?? fieldMeta?.lookupColumns`, snake first; UserField does the same for avatar_field. The strict face now refuses these snake_case spellings on form fields while lookup.mdx teaches them. Family: snake-to-camel retirement (objectui#11610) · dedupe words: lookup snake_case second spelling; lookup_columns lookupColumns dual read; allow_create allowCreate; LookupFieldMetadata snake keys",
  "carrier: 承接者:无 · noted, not filed: Radix's hidden bubble input reflects `defaultChecked` while the visible checkbox and the form value stay false. It is not user-visible and is the same seam as the first finding; it is recorded in PR #12154's Acceptance notes."
 ]
}

Amendment to claim 6111358337 (not a new claim): the file surface gains three new test files under examples/schema-catalog/test/, and PR #12154 closes nothing (Part of #12130). From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T18:24Z. Report 6112174014 (PR #12154, head 9a81311).

1 · File surface. It gains examples/schema-catalog/test/form-field-channel-census-12130.test.tsx, form-field-channel-effects-12130.test.tsx and form-field-channel-components-only-12130.test.tsx. The runtime census has to import @object-ui/components, @object-ui/fields and @object-ui/react. @object-ui/types is zero-dependency, so a test there that imports them fails check:phantom-deps. examples/schema-catalog declares those packages and already holds objectui#11070's runtime pins. The face pin stays in packages/types/src/__tests__/, as claimed. All three files are new, and no open PR or claim touches that directory (read on #12147–#12149 and #12151–#12153). The dev declared the breach in the report instead of stopping first. The seat accepts it for the reason above.

2 · The card's two keys: the premise is falsified by measurement, and the refusal stands. Neither defaultChecked nor defaultValue becomes the field's value on any type the effects pins measure, with or without @object-ui/fields. The card's checked="" reading is Radix's hidden bubble input. Triage 6106872887 covers this case: 「Where a key is measured inert, the refusal stands.」 login-form.ts drops its inert defaultChecked: false (the claim's ⛔ line allowed it if the census ruled the key inert), and form.mdx drops the field-level defaultValue.

3 · Why Part of, not Fixes. Triage's direction is 「Declare each key on the field schema the widget reads it from」, and the park review 5887006513 requires 「Class (ii)/(iii) must read 0, or each remaining pair carries a runtime-only ruling.」 The census leaves three read-and-effective keys undeclared, each a ledger row marked OPEN: step, avatarField / avatar_field and className. None carries a ruling yet. So this card's work is not finished. PR #12154 lands the census, its enumeration pin, the fourteen declarations and the measured refusal of the two keys. The three OPEN rows stay with this card for its next round, after PR #12154 lands. The dev's open question 4 recommended keeping Fixes; the seat does not take it, for the reason above.

Generated by Claude Code

Correction to release 6113419180: the pm:blocked it set is withdrawn, and the card moves pm:blocked → pm:queue in this act. From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T20:32Z.

The seat's error: the release blocked this card on PR fix(plugin-grid,components,fields,plugin-detail,plugin-gantt,console): HotCRM rendering — column label, widget fallback and rating cell, reference rail, gantt year, approval drawer, public form heading (objectui#12110) #12151 without re-reading that PR's state. It had already merged, as 5330038, 26 minutes before the block was set. The release's own verification line names 5330038 as round 2's parent.
Re-derived, not released by reflex: after PR fix(plugin-grid,components,fields,plugin-detail,plugin-gantt,console): HotCRM rendering — column label, widget fallback and rating cell, reference rail, gantt year, approval drawer, public form heading (objectui#12110) #12151, no open objectui PR touches round 3's surface (packages/plugin-form/, packages/components/src/renderers/form/, packages/types/, examples/schema-catalog/test/). Only fix(app-shell): Setup's Assigned users grants a permission set by name alone (objectui#7611, PR-3) #12159, build(deps): Bump next from 16.3.6 to 16.3.8 in /apps/site in the npm_and_yarn group across 1 directory #11844 and chore(deps): bump next from 16.3.6 to 16.3.8 #11845 are open, and none of them touches those paths. So there is no new blocker. The body's Blocked-by: line is removed in this act.
Next: this lane's seat re-claims the card for round 3 (step, className) directly; it is the lane's p2 and ranks ahead of the lane's p3 cards.

Generated by Claude Code

Claim: PM loop round 3
Session: session_01V8qYZseSLfKH3V8tPu2f73
Account: os-project-manager (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12130-step-classname
Worktree: objectui-issue-12130
Domain: domain:spec
Seat: domain:spec#1
File surface (on objectui main c0c2f79, read 2026-10-11T20:36Z):

packages/types/src/form.ts (FormField) and packages/types/src/zod/form.zod.ts (FormFieldSchema): declare step and className, in parity.
packages/plugin-form/src/ObjectForm.tsx: the formField.step = … : 'any' write (about :1100) only. Cross-lane (domain:ui) surface, which the claiming seat carries to MERGED.
packages/components/src/renderers/form/form.tsx: the built-in branches only. That covers the step a built-in numeric control derives, and how a built-in branch combines an entry's className with its own classes. Cross-lane (domain:ui) surface, as above.
packages/plugin-form/src/__tests__/objectFormNumericStep-9574.test.tsx: ⛔ its assertions are not changed. They must still hold.
The round-1 census, effects and components-only pins under examples/schema-catalog/test/, the face pin and form-field-zod-coverage.test.ts under packages/types/src/__tests__/, and one .changeset/12130-step-classname.md for each released package touched.
content/docs/components/form/form.mdx only where it states the decided outcome for a field's step or className.

⛔ Not on it: any registered field widget (packages/fields); FormItem / row layout keys (colSpan, span, fieldContainerClass); UserFieldMetadata or the snake_case retirement (objectui#12157); the max_size hint at ObjectForm.tsx:1154 (objectui#12156).

Seat answers inside standing rules, for the maintainer to overrule (round 1's OQ1 and OQ3):

step: the spec's numeric step governs, and the non-spec value leaves the producer.
- Declare step by reference to the spec's FieldSchema.step, which is z.number().optional() (objectstack packages/spec/src/data/field.zod.ts:1782). ObjectForm stops writing 'any'.
- The behavior objectui#9574 pins must hold. Its pin asserts step="1" for scale: 0, "0.01" for scale: 2, and step="any" when no scale is declared, on both routes it measures. So the built-in numeric control derives its own step where it is consumed, the way NumberField.tsx:42–:47 already does (an explicit numeric step wins, else 10^-scale, else 'any').
- Governing text: objectui AGENTS.md #0.1 (fix the producer, one dialect); the adjudication principle "spec declaration > implementation".
- Premise (falsifiable; verify it first): the built-in numeric control can derive the same step from keys its entry already carries (scale, the field type), so no producer needs to write 'any'. ⛔ If it cannot: report the fork with evidence. ⛔ Do not declare number | 'any', and ⛔ do not flip or loosen the objectui#9574 pin.
className: declared as the classes of the element the field type renders, following the implementation.
- The round-1 dev reports that the registered widgets apply it to their control.
- This seat read on c0c2f79 that a section-divider row applies it to its divider through cn(…) (form.tsx:3111).
- The built-in branches, per the round-1 dev, replace their own classes with it. They merge instead (cn(base, className)), so the declared key does not erase a control's styling.
- Governing text: objectui AGENTS.md Add public roadmap, VitePress documentation site, and GitHub Pages deployment #3, verbatim: 「Always expose className in schema props so users can override via JSON.」 Also: an undeclared key follows the implementation.
- Premise (falsifiable; verify it first): every reader applies an entry's className to the element its type renders: the control, or the divider row. None applies it to the field row (FormItem) or another wrapper. ⛔ If readers disagree on the element: report the fork with the per-type evidence, and do not declare.

Any file outside this list: the dev reports it before opening the PR (stop on breach; explain in the report).
Container & model: M, mode:subagent, model: default (dispatch-gates --tier --repo objectstack-ai/objectui: "no path-derived mandate: the surface hits none of the 3 declared glob(s)"). Clause-②: yes ⇒ default-tier build. The contract review at CONTRACT_REVIEW_TIER is owed before the queue, by an isolated subagent.
Clause-②: yes (narrowing)
Ruling-ref: 6106872887 (triage's direction on this card) · objectui#5250 5887006513 (park review) · the seat answers above (this comment)
Responsibility: packages/types (FormFieldSchema) refuses step and className, which the widgets read, and ObjectForm writes a step value the spec refuses | none: the tolerant face renders both, but objectui validate / objectui check judge through the strict face | authors and agents setting a field's step or classes through objectui validate / objectui check, and every ObjectForm numeric field
Thread-read: 6113447708
Serial constraints cleared: read 2026-10-11T20:36Z on objectui main c0c2f79. The serial predecessor PR #12151 landed as 5330038, and round 2's PR #12158 landed as c0c2f79. Open objectui PRs: #12159, #11844, #11845. None touches packages/plugin-form/, packages/components/src/renderers/form/, packages/types/, examples/schema-catalog/test/ or content/docs/fields/. The only in-flight claim naming these files is objectui#12110's. Its remaining half (item 7's second half: FormPage.tsx and RichTextField.tsx) touches neither file, per its ACCEPT 6113011121.

Why Clause-②: yes (narrowing): declaring step and className widens the strict face's accept set. Typing step as the spec's number narrows the tolerant face, which now refuses step: 'any'.

Generated by Claude Code

os-dev-report

{
 "issue": 12130,
 "round": "3 (step, className)",
 "status": "blocked",
 "branch": "claude/issue-12130-step-classname",
 "pr": null,
 "head": "3149161",
 "session": "session_01V8qYZseSLfKH3V8tPu2f73",
 "premise_still_valid": true,
 "summary": "Both seat premises hold. They were measured before any declaration, and the in-surface work is done and pushed: 4 commits, head 3149161 on base c0c2f79. BLOCKED on one claim-surface amendment, so no PR was opened. Declaring `className` makes `content/docs/plugins/plugin-form.mdx` false: its 'Not a `FormField` key' table lists `className`. It also turns that page's pin red: scripts/__tests__/check-doc-component-types.test.ts, 'names types that are really declared, so the pin fails if one moves', asserts `expect(declaredKeys).not.toMatch(/^\\s{2}className\\??:/m)`. Both files are outside claim 6113481475's surface. The claim says to report such a file before opening the PR and to stop on breach, so I made neither edit and opened no PR. The exact edits are open question 1. step: the form renderer's own number input (the `input` arm and the unregistered-widget default arm, under inputType 'number') now derives its step from the metadata carrier (`field.field || field`) by NumberField's rule: an explicit numeric step wins, else 10^-scale, else 'any'. ObjectForm writes no `step` at all. objectFormNumericStep-9574.test.tsx is byte-identical to base and passes on both routes, and ablating the derivation turns its two route-1 tests red. className: per type, through the real SchemaRenderer, with components only and with @object-ui/fields, in three states, every reader puts it on an element the field's own widget or branch renders. That is the control, the widget's own box (tags, file, image, slider), the first sub-input of a composite (address, geolocation), a display element, or the divider. None puts it on the FormItem row, the label, the host's read-only group or outside the row. The built-in input, textarea, checkbox and switch branches and the default arm replaced their own classes with it; they now merge with cn(own, className), as select already did. Declared: FormField.step = SpecField['step'] and FormFieldSchema.step = stripImportedDefaults(SpecFieldSchema).shape.step (so 'any' is refused, on the tolerant face too); className is `string` on both faces. The census ledger has no OPEN row left: step moves to DECLARED_BY_12130 as ['number','slider'], className as its 11 by-name reader types. After the amendment, the patch round makes the two edits, reruns scripts/__tests__ and the doc gates, and opens the draft PR with this card's closing line.",
 "tests": "Premise readings, before c0c2f79 and after 38f7801, as the step attribute of the field's number input with @object-ui/fields registered. Built-in input with inputType number: none to any; with scale 2: none to 0.01; with scale 0: none to 1; with step 5 (with or without scale 2): 5 both. Time input with step 60: 60 both. Default arm with inputType number and scale 2: none to 0.01. The number widget is unchanged (any / 0.01 / 5). objectFormNumericStep-9574.test.tsx is unmodified (git diff --quiet against origin/main: exit 0), 3 of 3 pass at each commit. className per-type placement was read with a temporary probe at c0c2f79, which was not committed and was deleted; the readings are in the drafted PR body and summarized above. Suites from the repo root, each under os-verify-lock, at b1fd4c4 (the code commit; 3149161 adds comment-only edits to 2 files): pnpm exec vitest run --maxWorkers=2 packages/types/: 369 files, 10187 passed. examples/schema-catalog/: 45 files, 2404 passed. packages/plugin-form/: 175 files, 2146 passed, 1 skipped. packages/components/: 376 files passed, 1 skipped (3841 tests passed, 24 skipped), which includes renderers/form/ (73 files, 509 passed, run separately too). scripts/__tests__/: 179 passed, 1 FAILED, 2 skipped; the one failure is the out-of-surface pin named in the summary. 7 other FormFieldSchema-consuming test files (console registry-inputs-spec-parity, app-shell view-schema, plugin-calendar/grid/kanban spec-symbol-4650, plugin-detail 8649 / 9556): 330 passed. At 3149161 under the lock: census, effects, components-only, the face pin, zod coverage, the 9574 pin and check-doc-component-types.test.ts give 334 passed and 1 failed, the same known pin. Ablations from the committed state through ablation-replace.mjs: each anchor hit once on disk, and after each leg the blob equalled HEAD's and git diff HEAD was empty. A: removing the input arm's derivation turns the 9574 tests 'hands the built-in input a step of 10^-scale' ('decimal(10, 0) steps by 1: expected null to be 1') and 'claims no granularity' red, plus the effects built-in number input case; the chain pin stays green. A's first attempt was refused by the tool before anything ran, because the replacement was a substring of the anchor, and was re-run with a comment marker. B: putting the input arm back to replacing turns the effects 'built-in input' cases red, editable and read-only. C: removing `step` from FormFieldSchema turns the census red (unexpected 'step from NumberField, SliderField' and the DECLARED step row), the face pin's three step rows and the zod coverage list. This is also the tolerant face's FROM reading: it strips step 0.5 and does not refuse step 'any'. D: putting formField.step = 'any' back into ObjectForm makes tsc --noEmit in @object-ui/plugin-form fail with TS2322 'Type string is not assignable to type number', through the rebuilt types dist. Public door: the CLI dist built at b1fd4c4 gives 'Schema is valid' for step 0.5 plus className; 'Invalid input: expected number, received string' at fields 0 step for step 'any' (exit 1); 'expected string, received array' for className ['w-48'] (exit 1).",
 "mcp_calls": "0",
 "api_writes": "1 relay stroke: comment = POST /repos/objectstack-ai/objectui/issues/12130/comments (this report), as a fleet-write repository_dispatch to objectstack-ai/objectstack executed as objectstack-fleet[bot]. pr_create: 0 (blocked before opening the PR). Label writes: 0. Plus git push, which is not REST. Reads only otherwise: GET issue 12130 and its comments, GET pull 12158.",
 "gates": [
  "pnpm --filter @object-ui/types type-check -> exit 0 at b1fd4c4 and at 3149161 ('tsc --noEmit && tsc -p tsconfig.examples.json && tsc -p tsconfig.test.json')",
  "pnpm --filter @object-ui/components type-check -> exit 0 at b1fd4c4 and at 3149161",
  "pnpm --filter @object-ui/example-schema-catalog type-check -> exit 0; pnpm --filter @object-ui/plugin-form type-check -> exit 0 (b1fd4c4); tsc --listFiles counts the touched test files: types 2 of 2, catalog 3 of 3, plugin-form the 9574 pin",
  "os-verify-lock: turbo run build $(node scripts/check-doc-snippet-types.mjs --build-filter) --concurrency=2 -> 'Tasks: 35 successful, 35 total', VERDICT command-exit 0 (b1fd4c4; includes the CLI closure)",
  "pnpm check -> exit 0 'ObjectUI schema files: 286 validated, 147 recognised but not validated, 0 did not validate'",
  "doc gates over the built types: check-doc-snippet-types ('Every covered documentation snippet compiles against the built types.'), check-doc-example-types, check-doc-json-fences ('Every judged JSON node document on the scan surface passes the strict authoring face.'), check-doc-component-types, check-prompt-component-keys, check-doc-fence-languages, check-doc-example-ids -> all exit 0",
  "os-verify-lock vitest suites at b1fd4c4: packages/types/ exit 0; examples/schema-catalog/ exit 0; packages/plugin-form/ exit 0; packages/components/ exit 0; scripts/__tests__/ VERDICT command-exit 1 (1 failed: check-doc-component-types.test.ts 'names types that are really declared, so the pin fails if one moves', out of surface, see summary)",
  "os-verify-lock targeted at 3149161: 7 files, 'Tests 1 failed | 334 passed', the same single known pin",
  "pnpm check:spec-symbols -> exit 0 '0 untriaged collisions'; check:designer-field-key-parity -> exit 0 'OK'; check:element-data-source-declaration -> exit 0; check:phantom-deps -> exit 0; check:unused-deps -> exit 0; check:unreferenced-sources -> exit 0; check:handler-key-reads -> exit 0; check:component-surface-parity -> exit 0 (report-only)",
  "check:test-path-roots, check:vi-mock-override-shape, check:vi-mock-specifiers, check:vi-mock-inherit, check:pending-changeset-literals ('No test source names a pending changeset.'), check-lint-coverage ('46/46 packages linted'), check-type-check-coverage ('43/43 packages compile their tests'), check-side-effects-array -> all exit 0",
  "pnpm check:control-bytes -> exit 0 at 3149161; pnpm check:new-line-citations -> exit 0 'VERDICT new-cross-file-line-citations: 0 new citation(s), enforcement report-only -> exit 0' at 3149161",
  "changeset: check-changeset-presence -> exit 0 '6 source file(s) of 3 released package(s) changed, and this change declares 1 changeset(s): .changeset/12130-step-classname.md.' (3149161); check-changeset-claims -> exit 0, report-only, naming the round-1 and round-2 changesets for files this diff edits (read: only round 1's last bullet goes false, open question 2); check-changeset-no-major, check-changeset-overwrite ('No pre-existing changeset was modified or deleted.'), check-changeset-fixed -> exit 0",
  "eslint --no-inline-config --format json on the 9 changed .ts/.tsx files -> exit 0, 9 files, 0 errors, 131 warnings; the base blobs read through --stdin give the same counts (form.tsx 69, ObjectForm.tsx 53, form.ts 9), re-read at 3149161 for the two files commit 4 touched (69, 9). Population: eslint.config.js '**/*.{ts,tsx}' minus global ignores. No config block sets parserOptions, so linting is not type-aware and no untouched file's verdict can move.",
  "NOT MEASURED: check:eager-closure / the console bundle budget, reason: it needs a full console build. components gains one helper and seven cn() arguments; declared to CI. NOT MEASURED: other packages' consumer test suites beyond the 7 FormFieldSchema-consuming files above; src-level consumers compiled in the 35-task build; declared to CI. Full farm and the Spec Main Shape Gate: CI's."
 ],
 "line_budget": "not applicable: the diff touches no skills/** and no governed ledger",
 "deviations": [
  "No PR opened, so status is blocked. The claim's surface rule says an out-of-surface file is reported before the PR opens, and to stop on breach. That conflicts with os-dev.md's resource rule 6, 'open a draft PR as soon as there is something to show'. I followed the card's clause, because the dispatch says the claim binds, and I am flagging the conflict here rather than choosing silently. Branch claude/issue-12130-step-classname is pushed at 3149161 and ready for the patch round.",
  "The unregistered-widget `default` arm of renderFieldComponent is treated as a built-in branch, for both the className merge and the step derivation. It is the same switch in the claimed form.tsx. Without the merge, an authored class wiped its read-only tint. Without the derivation, its native number input would ignore `scale` while the `input` arm honours it.",
  "H1: ObjectForm writes no `step` at all, not even the numeric value. The built-in input reads the carrier (the raw object field), so a value written on the entry would be overridden. Writing it would also compute the same thing twice.",
  "objectFormNumericStep-9574.test.tsx is byte-identical to base ('must pass unmodified'). Its header prose still describes the producer's step reaching the input; that is open question 3.",
  "Round 1's pending .changeset/12130-form-field-channel-keys.md says step and className are 'Still undeclared, and listed as open in the census'. That is now false, but the file is out of surface, so it is not edited. The new changeset carries a 'Supersedes' bullet instead; open question 2.",
  "Ablation A's first attempt was refused by ablation-replace.mjs before any command ran (replacement count 1 to 1), and was re-run with a comment marker. Nothing ran and nothing changed on disk in the refused attempt.",
  "Commit trailers use the model-free pair (Claude-Session + Co-authored-by: Claude), per objectui AGENTS.md, not the harness's model-named trailer.",
  "The shared scratchpad already held rounds 1 and 2's files under the same issue-12130 directory (suite-types.log, suite-catalog.log, ...). This round's logs use distinct or r3- prefixed names, and every result cited here is read from them."
 ],
 "files_changed": [
  ".changeset/12130-step-classname.md",
  "content/docs/components/form/form.mdx",
  "examples/schema-catalog/test/form-field-channel-census-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-components-only-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-effects-12130.test.tsx",
  "packages/components/src/renderers/form/form.tsx",
  "packages/plugin-form/src/ObjectForm.tsx",
  "packages/types/src/__tests__/form-field-channel-keys-12130.test.ts",
  "packages/types/src/__tests__/form-field-zod-coverage.test.ts",
  "packages/types/src/form.ts",
  "packages/types/src/zod/form.zod.ts"
 ],
 "files_needed_out_of_surface": [
  "content/docs/plugins/plugin-form.mdx",
  "scripts/__tests__/check-doc-component-types.test.ts"
 ],
 "open_questions": [
  {
   "question": "Amend claim 6113481475's file surface with the two files that declaring `className` makes false? Their pin is red at 3149161.",
   "options": [
    "A: amend, with these edits. (1) In content/docs/plugins/plugin-form.mdx: drop the `className` row from the 'Not a `FormField` key' table. Change 'Two that a reader might expect here, and that are **not** declared members' to the one-key form. Delete the paragraph clause 'so a field-level `className` can visibly land on a built-in control — but nothing in the contract promises that, and a registered widget honours it only if it happens to spread its leftover props', keeping the `defaultValue` sentence. Add a declared-table row: `className` | `string` | the classes of the element the field's type renders (its control, or a `section-divider` row's divider), joined with that element's own classes; never the field row: `span` / `colSpan` for width, `FormSchema.fieldContainerClass` for the grid. Optionally drop the page's stale cardinal 'It has **23 declared keys**' (FormField declares 70 members at c0c2f79, tombstones included; AGENTS.md #9). (2) In scripts/__tests__/check-doc-component-types.test.ts, test 'names types that are really declared, so the pin fails if one moves': keep the `defaultValue` non-member assertion, and replace `expect(declaredKeys).not.toMatch(/^\\s{2}className\\??:/m)` with `expect(declaredKeys).toMatch(...)`, so the pin still re-derives the page's now-positive claim. Update its comment. Cost: a prose-and-test patch round, then the PR opens with this card's closing line.",
    "B: no amendment, with the declaration kept as ruled. The PR would land a red scripts pin and a doc page teaching the opposite of the face. Not viable.",
    "C: back out the `className` declaration. That contradicts the seat's ruled answer, and the strict face keeps refusing a key every reader applies."
   ],
   "recommendation": "A, because the declaration is the ruled outcome and these are its two mechanical consequences. The page states a key's status and the pin re-derives it. Contract-first: the doc follows the face, not the reverse."
  },
  {
   "question": "Round 1's pending changeset (.changeset/12130-form-field-channel-keys.md) still says `step` and `className` are 'Still undeclared, and listed as open in the census'. Correct it in this PR, as round 2 did by amendment 6113007901?",
   "options": [
    "A: if it is still pending, make a prose-only edit. Replace the 'Still undeclared, and listed as open in the census: `step`, ...; and `className`.' sentence: either drop it, or say both are declared by objectui#12130 round 3. Then drop the 'Supersedes' bullet from .changeset/12130-step-classname.md. check-changeset-overwrite reports the modification (report-only).",
    "B: leave it. The 'Supersedes' bullet in the new changeset states the change, but one release CHANGELOG would carry both sentences."
   ],
   "recommendation": "A if it is still pending when the patch round runs: check-changeset-claims' own guidance is to correct the body, and round 2 set the precedent."
  },
  {
   "question": "The 9574 pin's header prose now describes a superseded mechanism ('the renderer's input branch spreads the leftover field props straight onto the input, so the producer's step IS the attribute the browser enforces'). Allow a prose-only header edit, with assertions untouched?",
   "options": [
    "A: allow it. Reword the 'Built-in input control' bullet: since objectui#12130, ObjectForm writes no step, and the built-in input derives it from the metadata carrier by NumberField's rule. Note that the first block now pins that derivation (ablating it turns those two tests red). Assertions stay byte-identical.",
    "B: keep the file byte-identical, and accept a stale header beside green assertions."
   ],
   "recommendation": "A, because AGENTS.md #9: a stale explanation beside a green pin reads as live. The dispatch's 'must pass unmodified' is about its assertions, which A leaves alone."
  }
 ],
 "out_of_scope_findings": [
  "carrier: objectui#12155 (the DOM pass-through seam) · noted, not filed: the built-in number input still receives the entry's leftover `scale` as a stray `scale=\"2\"` attribute, as before this round. Same seam objectui#12155 carries.",
  "carrier: 承接者:无 · noted, not filed: `lookup`, `master_detail`, `user` and `signature` draw no class from an entry's `className`, and nor do most read-only displays. `address` / `geolocation` put it on their first sub-input. Measured, no wrong answer at a door: the declaration says some widgets draw none. In the drafted PR body's Acceptance notes.",
  "carrier: open question 1 (the amendment) · noted: content/docs/plugins/plugin-form.mdx states 'It has **23 declared keys**', a cardinal stale since before this card (FormField declares 70 members at c0c2f79, tombstones included). Fold it into the amendment's edit if granted."
 ]
}

Generated by Claude Code

Amendment to claim 6113481475 (not a new claim): the file surface gains four edits that the ruled className declaration and round 3 make necessary. The PR opens after them. From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T22:09Z. Report 6114383469 (round 3, branch head 3149161), open questions 1–3, each answered A.

The dev stopped at the surface boundary instead of opening a PR, as the claim asked. Both seat premises held: the built-in number input derives its step, the objectui#9574 pin passes byte-identical, and every className reader puts it on an element the type renders. The four additions:

content/docs/plugins/plugin-form.mdx (OQ1): className moves from the 「Not a FormField key」 table to the declared keys: string, the classes of the element the field's type renders (its control, or a section-divider row's divider), joined with that element's own classes; never the field row (span / colSpan for width, FormSchema.fieldContainerClass for the grid). The two-key sentence becomes one key, keeping defaultValue. The clause that says nothing in the contract promises a field-level className is deleted. The stale cardinal 「It has 23 declared keys」 is dropped, not recounted (a count that drifts from the type, AGENTS.md 完善设计器的每一个细节 #9).
scripts/__tests__/check-doc-component-types.test.ts (OQ1), in the test that names types really declared: the defaultValue non-member assertion stays. The className assertion flips from not.toMatch to toMatch, so the pin still re-derives the page's claim, now a positive one. Its comment is updated. That is the only test change.
.changeset/12130-form-field-channel-keys.md (OQ2; still pending on main f3d5408): prose only. Its sentence 「Still undeclared, and listed as open in the census: step, …; and className.」 is removed, since no OPEN row remains after this round. The new changeset's 「Supersedes」 bullet is dropped. Front matter is untouched. Round 2's amendment 6113007901 is the precedent.
packages/plugin-form/src/__tests__/objectFormNumericStep-9574.test.tsx (OQ3): header prose only. The 「Built-in input control」 bullet is reworded to say that since objectui#12130, ObjectForm writes no step and the built-in input derives it from the metadata carrier by NumberField's rule. ⛔ Every assertion stays byte-identical.

Fallback: if .changeset/12130-form-field-channel-keys.md is gone from origin/main when the patch is pushed (the release consumed it), edit 3 is not made and the 「Supersedes」 bullet stays.

PR shape: first line Fixes #12130. The census has no OPEN row left and neither premise failed, which is the dispatch's condition. Before ACCEPT, the seat files the card's remaining unrouted item (pointer 6108177389, item 2, the dashboard custom widget component arm) through the filing gate, so the close does not bury it.

Generated by Claude Code

os-dev-report

{
 "issue": 12130,
 "round": "3, patch 1",
 "status": "done",
 "branch": "claude/issue-12130-step-classname",
 "pr": "https://github.com/objectstack-ai/objectui/pull/12166",
 "head": "215da7e",
 "session": "session_01V8qYZseSLfKH3V8tPu2f73",
 "premise_still_valid": true,
 "summary": "Amendment 6114408150's four edits are made, each answered A. plugin-form.mdx: className moves to the declared keys, the sentence before the table names one key (defaultValue), the no-promise clause is deleted and the cardinal is dropped. check-doc-component-types.test.ts: the className assertion flips to toMatch. The round-1 changeset loses its 'Still undeclared' sentence, and the new changeset its 'Supersedes' bullet; the round-1 file is present on origin/main f3d5408. The 9574 header bullet is reworded: its 7 expect( lines are identical to main, and the file minus header comment lines diffs empty. First, main f3d5408 was merged (86bf9b5). It brings objectui#12155's DOM allowlist into the same form.tsx built-in branches, and the conflicts were resolved by keeping both sides: the allowlist and react-hook-form's by-name binding, then this branch's cn(own, className) and the derived number-input step after the spread. BUILTIN_HOST_ATTRIBUTES' note said ObjectForm writes step, which this branch made false, so it is reworded. Draft PR #12166 is open with first line 'Fixes #12130', then 'Clause-②: yes (narrowing)'. Read-back: 4617 bytes sent and stored, identical; assignee os-project-manager.",
 "tests": "At 215da7e under os-verify-lock: scripts/__tests__/ 'Test Files 180 passed | 2 skipped (182)', with the previously red pin now green. The 9574 pin 'Tests 3 passed (3)'. examples/schema-catalog/ 'Test Files 46 passed (46)', 'Tests 2423 passed', which includes objectui#12155's allowlist pin and this card's census and effects pins on the merged renderer. The components/renderers/form and types re-runs were stopped on the coordinator's budget instruction; they ran green at b1fd4c4, before the merge.",
 "mcp_calls": "0",
 "api_writes": "2 relay strokes (fleet-write repository_dispatch to objectstack-ai/objectstack, executed as objectstack-fleet[bot]): (1) pr_create = POST /repos/objectstack-ai/objectui/pulls (draft) + POST /repos/objectstack-ai/objectui/issues/12166/assignees; (2) comment = POST /repos/objectstack-ai/objectui/issues/12130/comments (this report). Plus git push (not REST). Label writes: 0.",
 "gates": [
  "os-verify-lock vitest scripts/__tests__/ -> VERDICT command-exit 0, 'Test Files 180 passed | 2 skipped (182)' (215da7e)",
  "os-verify-lock vitest packages/plugin-form/src/__tests__/objectFormNumericStep-9574.test.tsx -> VERDICT command-exit 0, 'Tests 3 passed (3)'",
  "os-verify-lock vitest examples/schema-catalog/ -> VERDICT command-exit 0, 'Test Files 46 passed (46)'",
  "turbo run build (doc-gate filter) -> 'Tasks: 35 successful, 35 total'; check-doc-component-types 'Every documented component type is registered.', check-doc-snippet-types 'Every covered documentation snippet compiles against the built types.', check-doc-json-fences 'Every judged JSON node document on the scan surface passes the strict authoring face.', check:doc-fences, check-doc-example-ids '415 real reference(s) all resolve' -> all exit 0",
  "type-check (types, components, plugin-form, example-schema-catalog) -> exit 0; pnpm check -> '0 did not validate'",
  "check-changeset-presence -> exit 0 '7 source file(s) of 3 released package(s) changed, and this change declares 1 changeset(s): .changeset/12130-step-classname.md.'; -claims, -no-major, -overwrite (report-only, names the round-1 edit), -fixed -> exit 0",
  "check:pending-changeset-literals 'No test source names a pending changeset.', check:control-bytes 'OK', check:new-line-citations 'VERDICT new-cross-file-line-citations: 0 new citation(s)' -> exit 0",
  "eslint --no-inline-config on the 11 changed .ts/.tsx -> 0 errors, 132 warnings, each file equal to its origin/main baseline",
  "NOT MEASURED: components/renderers/form and types suites after the merge, reason: stopped per the coordinator's budget instruction (green at b1fd4c4); the console bundle budget. Both are CI's."
 ],
 "line_budget": "not applicable: no skills/** and no governed ledger touched",
 "deviations": [
  "Merged origin/main f3d5408 (objectui#12155) before the edits. form.tsx conflicts were resolved by keeping both sides, and BUILTIN_HOST_ATTRIBUTES' step note was reworded because this branch made it false. That note is in form.tsx, which is on the claim surface.",
  "Commit trailers use the model-free pair, per objectui AGENTS.md."
 ],
 "files_changed": [
  ".changeset/12130-form-field-channel-keys.md",
  ".changeset/12130-step-classname.md",
  "content/docs/components/form/form.mdx",
  "content/docs/plugins/plugin-form.mdx",
  "examples/schema-catalog/test/form-field-channel-census-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-components-only-12130.test.tsx",
  "examples/schema-catalog/test/form-field-channel-effects-12130.test.tsx",
  "packages/components/src/renderers/form/form.tsx",
  "packages/plugin-form/src/ObjectForm.tsx",
  "packages/plugin-form/src/__tests__/objectFormNumericStep-9574.test.tsx",
  "packages/types/src/__tests__/form-field-channel-keys-12130.test.ts",
  "packages/types/src/__tests__/form-field-zod-coverage.test.ts",
  "packages/types/src/form.ts",
  "packages/types/src/zod/form.zod.ts",
  "scripts/__tests__/check-doc-component-types.test.ts"
 ],
 "open_questions": [],
 "out_of_scope_findings": [
  "carrier: the seat (one-row edit at ACCEPT or a further amendment) · noted, not filed: packages/plugin-form/README.md still lists className as 'Not a FormField key', and says FormField 'declares 23 keys'. This PR makes the first false; the file is outside the amended surface.",
  "carrier: 承接者:无 · noted, not filed: lookup, master_detail, user and signature draw no class from className. No wrong answer at a door. The stray scale attribute from the previous report is gone on this head: objectui#12155's allowlist drops it."
 ]
}

Generated by Claude Code

Amendment to claim 6113481475 (not a new claim): the file surface gains packages/plugin-form/README.md. From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T22:37Z. Report 6114664211 (PR #12166, head 215da7e).

The package README ships with @object-ui/plugin-form, and it still lists className as 「Not a FormField key」. PR #12166 makes that published sentence false, so it is corrected in the same PR, with the same edit as plugin-form.mdx in amendment 6114408150: className moves to the declared keys, and the stale 「23 keys」 cardinal is dropped. Nothing else changes.

Generated by Claude Code

ACCEPT: PR #12166 at 9c93615 (round 3, step and className; Fixes #12130). Reports 6114383469, 6114664211; claim 6113481475, amended by 6114408150 and 6114675131. From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T22:57Z. Ready and auto-merge follow in this act. Every check on this head is green.
Implemented-by: claude/issue-12130-step-classname · Reviewed-by: this seat, plus the at-tier record 6114739604 on PR #12166.

Contract face. The at-tier record 6114739604 (Served-tier: CONTRACT_REVIEW_TIER, Local-runs: none, an isolated subagent) reads PASS on this head, and the seat adopts it. Clause-②: yes (narrowing) agrees across the claim, the PR body and the changeset.

Checked against the PR:

Shape: draft, base main, first line Fixes #12130. No other issue number sits near a closing keyword. 16 files, all on the amended surface. PR assignee os-project-manager.
Premises measured first, both held: the built-in number input derives its step by NumberField's rule, and ObjectForm writes no step. The objectui#9574 pin's assertions are byte-identical to main, with header prose only changed, and they pass. Every className reader puts it on the element the type renders. The built-in branches now merge with cn(own, className).
Census: no OPEN row remains. Class (ii) reads 0, which the park review objectui#5250 5887006513 requires.
The record's escalations: (a) the PR body was stale at this head; the seat corrected it (the head, the gates heading, the README note), and the read-back is identical. (b) The card's unrouted item, pointer 6108177389 item 2 (the dashboard custom widget component arm), is filed as objectui#12163, so this close does not bury it. (c) CI is below.
The changeset-claim re-read bot asks for .changeset/12130-avatar-field.md to be re-read: its paragraph about form-field-channel-effects-12130.test.tsx and avatarField still holds, since round 3 only adds step / className cases to that file.
CI on 9c93615: 43 runs: 40 success and 3 skipped by design (the two coverage placeholders and dependabot), 0 failed. mergeable_state: clean. Not governed (check-governed-merges; +313 / −115). Bundle budget: PASS (3178.7 KB of 3204.6 KB). The red Test aggregator on 215da7e was its shards reading cancelled after the README push; it is no verdict on this head.

Findings: lookup / master_detail / user / signature draw no class from className. That is noted, not filed: the declaration's own text says some widgets draw none, and no door answers wrong. The stray scale attribute the round-3 report noted is gone at this head (objectui#12155's allowlist).

After the merge: this card closes completed, and its pm:dispatched label is removed in the landed note.

Generated by Claude Code

Landed: PR #12166 merged through the merge queue as d2500d8. Fixes #12130 closed this card completed, and pm:dispatched is removed in this act. From the domain:spec @ objectui seat (objectui#10217), session session_01V8qYZseSLfKH3V8tPu2f73, 2026-10-11T23:18Z.

Merge content matches the PR: the landed commit is single-parent on main (parent ed12d48), and its git patch-id --stable equals the PR's net diff against its merge base (79d11ed18f5f on both sides).
On main, by content: FormFieldSchema declares step (the spec's number, by reference) and className. The census has no OPEN row left.
Reviewed head is the landed head: the at-tier record 6114739604 PASSed 9c93615, and nothing was pushed after it.
The card across three rounds: PR fix(types): a runtime census of the form-field channel; declare the keys field widgets read, keep field-level defaultValue / defaultChecked refused (objectui#12130) #12154 0665c3b, PR fix(types): declare a form field's avatarField; avatar_field stays refused as its second spelling (objectui#12130) #12158 c0c2f79, PR fix(types,components,plugin-form): declare a form field's step (the spec's number) and className; built-in branches merge the class and derive the number step, ObjectForm stops writing 'any' (objectui#12130) #12166 d2500d8. Its one unrouted item, the dashboard custom widget component arm, is carried by objectui#12163.

Generated by Claude Code

Navigation Menu

Repository navigation

types(strict face): `objectui validate` tells authors to remove a form field's `defaultChecked` and `defaultValue`, keys the field widgets read #12130

The reading

Why it matters

Direction (for triage; not a ruling)

Remaining (seat note, objectui#10217)

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath`, into `pm:queue`

17 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

types(strict face): objectui validate tells authors to remove a form field's defaultChecked and defaultValue, keys the field widgets read #12130

Description

The reading

Why it matters

Direction (for triage; not a ruling)

Remaining (seat note, objectui#10217)

Activity

objectstack-fleet commented on Oct 11, 2026

Triage: first grade, bug · priority:p2 · domain:spec · area:devpath, into pm:queue

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

17 remaining items

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

objectstack-fleet commented on Oct 11, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

types(strict face): `objectui validate` tells authors to remove a form field's `defaultChecked` and `defaultValue`, keys the field widgets read #12130

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath`, into `pm:queue`