[finding] `ObjectFormSection` still declares `className` / `gridClassName` after objectstack#13626 retired every authored-metadata read of them

[os-support-ai]

Observation only, filed unassigned and not queued — surfaced while implementing the objectstack#13626 ruling (PR #7198). Not graded here.

The shape

objectstack#13626 (maintainer ruling 2026-09-01, director decision batch C: retire the reads) removed all seven as any reads of className / gridClassName off the form-view section family, on the grounds that @objectstack/spec deliberately does not declare either key on the form-view/section authorable surface.

ObjectFormSection in @object-ui/types (packages/types/src/objectql.ts) still declares both keys, with doc comments describing an effect they no longer have:

/**
 * Custom CSS class for the section's wrapper (Card, when the form variant
 * renders sections as cards; the divider header, for the flat/simple path).
 */
className?: string;

/**
 * Custom CSS class for the section's field grid. Only used by form variants
 * that render sections as Card chrome (Modal/Split/Tabbed/Wizard).
 */
gridClassName?: string;

ObjectFormSchema.sections is typed ObjectFormSection[], so this is the type an author's section metadata is checked against. After #7198 an author can write either key, have it type-check, read a doc comment promising a wrapper class — and get nothing.

That is the mirror image of the original finding rather than its resolution: objectstack#13626 was "not declared on one side, consumed on the other"; what remains is "declared on one side, consumed nowhere" — the declared-but-inert shape of the enforce-or-remove class.

Why it was left alone in #7198

Deliberately out of scope, not overlooked:

• The ruling's action was specifically the cast reads. Removing two members from a published type in @object-ui/types is a different action against a different surface, and the maintainer weighed three routes (retire the reads / declare the keys / write the status quo down) — not this fourth one.
• The declarations are not fully dead. The per-layout section types (ModalFormSectionConfig, TabbedForm's FormSectionConfig, SplitFormSectionConfig, DrawerFormSectionConfig, WizardStepConfig) declare their own copies and their components still consume them — ModalForm.tsx:668, TabbedForm.tsx:469, SplitForm.tsx:414, WizardForm.tsx:1025 — for programmatic React mounts, which the authorable boundary does not govern. Only the ObjectFormSection (authored-metadata) declaration is now inert.

Why it is worth recording

The live consequence is small but real, and it is the exact hazard that makes the #7198 pin behavioural rather than a source grep:

Because the keys are still declared, restoring consumption needs no cast. A later "cleanup" writing a plain className: s.className type-checks and silently puts the retired read back. packages/plugin-form/src/__tests__/sectionStyleKeysRetired-13626.test.tsx is what catches that; the type system will not.

The question, not answered here

Should ObjectFormSection drop the two members (making the authored-metadata type agree with the authorable boundary, at the cost of a breaking type change to @object-ui/types and a doc pass over the per-layout section types), or keep them with doc comments corrected to say they are inert for authored metadata and honoured only on direct programmatic mounts?

Refs: objectstack#13626 (the ruling) · #7198 (the implementation) · packages/types/src/objectql.ts · packages/plugin-form/README.md "Section styling is not authorable"

[yinlianghui]

Evidence for triage (spec@objectui seat, session session_01V3hPr7riucnMfhcHY86Msd, 2026-09-02; no grading, no label change). Measured on objectui origin/main 67dadd6 and objectstack origin/main.

• Premise still true. packages/types/src/objectql.ts:1132 className?: string and :1138 gridClassName?: string on ObjectFormSection, doc comments still describe an effect nothing delivers. Read sites of those two members via ObjectFormSection: zero — ObjectForm.tsx keeps only deliberate-omission comments naming #13626 (lines 293/317/350/384/450); control: s.columns / section.columns reads still hit (5+2+3+2 across ObjectForm/ModalForm/SplitForm/TabbedForm), so the zero is real.
• Sibling per-layout config types are NOT the same case: ModalFormSectionConfig (ModalForm.tsx:89/95), SplitFormSectionConfig (SplitForm.tsx:57/64), TabbedForm FormSectionConfig (:92/98), WizardStepConfig (:141/146), DrawerFormSectionConfig (:90) are read uncast (ModalForm.tsx:667-668/697, SplitForm.tsx:393/414, TabbedForm.tsx:469, WizardForm.tsx:1024-1025) — live for programmatic mounts (AppContent.tsx:1044, useActionModal.tsx:253, plugin-designer/ObjectManager.tsx:262); ObjectForm rebuilds sections key-by-key and drops both keys (ObjectForm.tsx:395-445). A fix scoped to ObjectFormSection must leave those five alone.
• Upstream ruling to inherit (objectstack#13626, maintainer 2026-09-01, director batch C 「同意」, verbatim): "retire the reads … Declaring the keys was weighed and not adopted: it would formally invite free Tailwind strings into authored metadata, the exact class the boundary exists to keep out." Spec side: FormSectionSchema (view.zod.ts:2485) is a strictObject with neither key; gridClassName has zero hits in packages/spec/src. Nothing in a spec release is involved — this is objectui-internal, not release-gated.
• Blast radius of retiring the two declarations: @object-ui/types published surface (changeset minor per the fixed-group rule, never major); packages/plugin-form/src/__tests__/sectionStyleKeysRetired-13626.test.tsx:91-92,99-100 authors both keys on an ObjectFormSection-typed fixture and would fail type-check — must be rewritten to a cast/literal in the same PR; docs: zero gridClassName under content/docs, plugin-form/README.md:379-407 already teaches "section styling is not authorable".
• Not measured: downstream consumers outside this repo constructing ObjectFormSection literals with these keys.

Reading: same-family as #13626 (mother ruling: do not declare), fix = delete two members + fixture update, S-size, packages/types only. Grading is triage's.

---

Generated by Claude Code

[yinlianghui]

Graded and dispatched in one pass via the maintainer direct-dispatch channel (spec@objectui seat, session session_01V3hPr7riucnMfhcHY86Msd, 2026-09-02).

Provenance three-piece: who — the maintainer; verbatim — 「加快进度，并发5个子任务」 and 「继续了，你怎么停掉了，并发5个加快进度」 (live chat, 2026-09-02, this PM session); where — the maintainer direct-dispatch channel (SKILL.md 多仓协调 4). The first-touch grading that would normally be triage's is taken here under that instruction, on the same-family meta-rule: the mother card objectstack#13626 carries the maintainer ruling of 2026-09-01 (verbatim: "retire the reads … Declaring the keys was weighed and not adopted: it would formally invite free Tailwind strings into authored metadata, the exact class the boundary exists to keep out"), and this card is the declared-but-inert remainder of that same defect on the same surface. Evidence on the ref is in comment 5503267499. Triage may re-grade with pm:retriage; this seat yields if it does.

Direction inherited: retire the two declarations on ObjectFormSection (the authored-metadata type), matching the spec's strict FormSectionSchema and the ruling's rationale; the five per-layout config types stay (live for programmatic mounts). Type: Task. Size: S. Clause-②: yes (narrows a published @object-ui/types surface) ⇒ contract-review tier, PR parks draft + needs:contract-review, in-seat review.

State: finding → pm:dispatched (assign + label swap together). Claim follows.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 2 (spec@objectui, 2026-09-02 shift)
Session: session_01V3hPr7riucnMfhcHY86Msd
Branch: claude/issue-7200-object-form-section-inert-style-keys
Worktree: objectui-issue-7200
Domain: domain:spec
File surface: packages/types/src/objectql.ts (ObjectFormSection members className / gridClassName) + its zod mirror if one declares them (packages/types/src/zod/objectql.zod.ts) + packages/plugin-form/src/__tests__/sectionStyleKeysRetired-13626.test.tsx (fixture only) + one changeset; JSDoc-only touch on the five per-layout section config types is allowed, their members are NOT removed (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: fable — Clause-② content limb (narrows a published declared surface); the objectstack dispatch-gates.mjs --tier derives nothing for objectui paths, tier is this seat's judgment
Clause-②: yes
Serial constraints cleared: in-lane in-flight = #5213 (packages/types/src/widget.ts, in the merge queue) — disjoint. Cross-lane in-flight on packages/types: PR #7240 (crud.ts, disclosure.ts, form.ts + zod mirrors), PR #7241 (data-display.ts + zod mirror) — file lists read, neither touches objectql.ts. PR #7198 (the mother card's objectui half) merged 2026-09-01T11:11Z — its pin test is the fixture named above. No sibling card's pins assert these two members.

---

Generated by Claude Code

[yinlianghui]

Claim (dev agent, dispatched by spec@objectui PM loop round 2)
Session: session_01V3hPr7riucnMfhcHY86Msd
Branch: claude/issue-7200-object-form-section-inert-style-keys (pushed empty at 64d624d as the write-route probe)
Worktree: objectui-issue-7200
File surface: packages/types/src/objectql.ts (ObjectFormSection members className / gridClassName + JSDoc), packages/plugin-form/src/__tests__/sectionStyleKeysRetired-13626.test.tsx (fixture only), one changeset, one pin test under packages/types/src/__tests__/. Five per-layout section config types keep their members.

---

Generated by Claude Code

[yinlianghui]

os-dev-report

{
 "issue": 7200,
 "status": "done",
 "branch": "claude/issue-7200-object-form-section-inert-style-keys",
 "pr": "https://github.com/objectstack-ai/objectui/pull/7278",
 "premise_still_valid": true,
 "summary": "Removed the className and gridClassName members (and JSDoc) from ObjectFormSection in packages/types/src/objectql.ts, leaving a deliberate-omission note naming objectstack#13626 and the two pins; the five per-layout section config types keep their members untouched. Added a compile-time pin (packages/types/src/__tests__/object-form-section-style-keys-undeclared.test.ts: ts-expect-error on annotated literals, a directive-free contrast row, and a Record-of-keyof census that fails in both directions), rewrote the #7198 fixture to keep authoring both keys through a fixture type widened at its boundary (ObjectFormSection intersected with the two string keys), and reworded the three sentences the removal made false (ObjectForm.tsx tabbed-arm comment, plugin-form README section + contributor note). Zone-2 checks: (1) zero authored reads confirmed, columns control hits; (2) no zod mirror exists at all — ObjectFormSchema in zod/objectql.zod.ts (passthrough BaseSchema.extend) declares no sections key, so no runtime refusal pin is possible and no tombstone door exists; (3) PM assumption partly false in a harmless direction: the fixture literal was passed through `as any` and would NOT have failed type-check after the removal — it was widened explicitly anyway so the authored-undeclared intent is typed; (4) docs re-verified, only plugin-form README touched; (5) objectql.ts last touch still 03380aa, no churn on any edited file between BASE 64d624d and origin/main at report time. PR parked draft with needs:contract-review (label written via MCP union write, read back intact with the four labeler labels). Sanitizer note: the PR body's `Record of keyof ObjectFormSection to true` generic was eaten to bare `Record` in two places (prose still coherent); one duplicate session footer appended by the create path. Changeset: @object-ui/types minor with breaking semantics in the body. Worktree removed without force after the PR opened.",
 "tests": "All readings on head 1f3f33f (working tree byte-identical to the commit; type-checks and the types vitest re-run post-commit). Dependency closure built first under the shared lock (plugin-form^... VERDICT command-exit 0, 88s; later console^... BUILD_EXIT=0). @object-ui/types type-check (tsc --noEmit && tsc -p tsconfig.examples.json && tsc -p tsconfig.test.json): exit 0, 0 TS errors, script name echoed. @object-ui/plugin-form type-check (tsc --noEmit && tsc -p tsconfig.test.json): exit 0, 0 TS errors. @object-ui/console type-check: TC_EXIT=0 after the closure build (the first exit 2 was TS2882 unbuilt-plugin precondition, not a verdict). vitest from repo root: packages/types/ 'Test Files 85 passed (85) / Tests 1067 passed (1067)' (VERDICT command-exit 0); packages/plugin-form/ 'Test Files 81 passed (81) / Tests 821 passed | 8 skipped (829)' (VERDICT command-exit 0, held 187s). Rebuilt dist: packages/types/dist/objectql.d.ts gridClassName count 0, columns control in the ObjectFormSection block 1. Ablation (direction predicted first: RED with exactly two TS2578 plus one census error): re-inserted both members into objectql.ts, mutation confirmed on disk by grep -c ABLATION-7200 = 2 and grep -c 'gridClassName?: string' 0 to 1 plus git diff --stat HEAD = 1 file +5; tsc -p tsconfig.test.json on the mutated tree: TSC_EXIT=2 with TS2578 at test lines 97 and 107 and TS2739 at 137 'missing the following properties ... className, gridClassName' — observed = predicted, nothing else red; restore via git checkout HEAD -- absolute path under an EXIT INT TERM trap, proven by empty git diff HEAD --stat and git hash-object = HEAD blob 1b57cd56776e0d2719146cf0d1dbc8aee56b1931 (non-empty, matching), marker count 0 after restore; no dist rebuild needed because the pin imports ../objectql from source. Gates (judgement lines): check-changeset-presence '3 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)'; check-changeset-no-major 'No changeset declares a major bump'; check-changeset-fixed 'All workspace packages are in the changeset fixed group'; check-changeset-overwrite 'No pre-existing changeset was modified or deleted'; check:spec-symbols '1333 files scanned against 4959 spec export names ... 0 untriaged collisions'; check:control-bytes 'OK (scanned 6004 tracked text file(s))'; eslint --no-inline-config on the four edited ts/tsx files: 0 errors (72 pre-existing no-explicit-any warnings in objectql.ts, none on touched lines); governed-surface guard on the six changed paths 'NOT GOVERNED'. check:doc-snippets: not owed (no covered snippet types a section literal); ran twice anyway, both exit 2 PRECONDITION NOT MET (unbuilt packages; the second run wanted only packages/cli) — NOT MEASURED, CI runs it. CI on the PR: in_progress at report time (report delivered at draft-PR time per contract).",
 "mcp_calls": "6 — add_issue_comment (claim), create_pull_request, pull_request_read x2 (body read-back + label read-back), issue_write (label union), add_issue_comment (this report). Issue body and comments were read from the public web page's embedded JSON payload (REST returned 403 for the session; gh absent), zero MCP reads.",
 "open_questions": [],
 "out_of_scope_findings": []
}

---

Generated by Claude Code

[yinlianghui]

PM verdict: ACCEPT + in-seat contract review PASS — PR #7278 (spec@objectui seat, session session_01V3hPr7riucnMfhcHY86Msd; reviewed against the PR diff and the GitHub record, not the report's narrative).

Tier fuse read before this review: get_session.external_metadata.last_served_model = claude-fable-5-1 (03:00Z) vs CONTRACT_REVIEW_TIER = claude-fable-5 (scripts/pm/dispatch-gates.mjs:7692) — at tier; review done in-seat per the maintainer ruling of 2026-08-31 (「项目经理是fable 或者可以派fable的子任务都可以自己contract view吧?」). No dev flag and no contract fork, so no isolated second-opinion sub-agent was owed.

Contract judgment list

1. Derived judgments (accept-set / public-surface deltas, each named):
   • ObjectFormSection loses optional members className and gridClassName (packages/types/src/objectql.ts). Effect: an annotated TS literal (or an ObjectFormSchema.sections entry) carrying either key is now an excess-property compile error; JSON metadata is unaffected because ObjectFormSection has no zod mirror (ObjectFormSchema in zod/objectql.zod.ts declares no sections), and the renderer already ignored both keys since PR refactor(plugin-form): retire the form-view section className/gridClassName reads #7198. Correct — this is the ruled direction of objectstack#13626 (do not declare) applied to the remainder; the type now matches the spec's strict FormSectionSchema.
   • The five per-layout config types (ModalFormSectionConfig, SplitFormSectionConfig, TabbedForm FormSectionConfig, WizardStepConfig, DrawerFormSectionConfig) keep their members — correct fence, they are read uncast for programmatic mounts outside the authorable boundary.
   • No ?: never tombstone — correct: no parse door exists, and a tombstone would still be a declaration in the published .d.ts, which is what the ruling declined.
   • Pin: @ts-expect-error on annotated literals + directive-free contrast row + Record<keyof ObjectFormSection, true> census (fails both directions); ObjectFormSection has no index signature, so the directives are load-bearing — ablation in the PR body shows exactly the predicted two TS2578 + one TS2739, nothing else. Accepted.
   • refactor(plugin-form): retire the form-view section className/gridClassName reads #7198 fixture keeps authoring both keys through a boundary-widened type — the behavioural pin stays meaningful. Accepted.
2. Semver: @object-ui/types minor with the breaking semantics in the changeset body — matches the fixed-group rule (never major). Consistent.
3. Boundary flags: open_questions empty, out_of_scope_findings empty; the one PM assumption the dev falsified (the fixture was as any-cast, so it would not have failed type-check) is harmless and was still handled explicitly. Nothing to escalate.

Checklist readings: draft ✓ · base main (64d624d) ✓ · first line Fixes #7200 ✓ · 6 changed files all inside the claimed surface ✓ · report comment 5503615293 with the literal os-dev-report first line ✓ · consumer-package readings real (types 1067/1067, plugin-form 821 passed / 8 skipped, three type-checks exit 0) ✓ · path face not governed ✓.

Carriers: needs:contract-review cleared on both the PR and this card in this same pass (the dual-carrier machine check check-clause2-carriers.mjs --pair 7278 cannot read GitHub from this container — exit 3, environment cannot answer — so the pair was verified by hand: both carried the label before this clearing, both cleared now). Landing: ready + auto-merge only after the full farm is green by name; CI was in_progress at 02:52Z.

---

Generated by Claude Code