feat(app-shell): the Studio option editor authors an option's lifecycle kind - #26931

Requested by Warren · project thread
Fixes #26729

Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:Studio 字段检查器的选项编辑器,在单选 select / radio 字段的每个选项上新增「生命周期类别」选择(未设 / 进行中 / 成功结束 / 失败结束),写入 spec 已发布的 kind。
• 为什么改:对象把这个字段选为 Lifecycle 字段时,spec 要求每个选项都声明 kind,而 Studio 原来没有任何地方能写它,保存一律 422,只能走 API。
• 风险与回滚:只动 Studio 编辑器与 i18n 文案,不改 spec、不改引擎;多值字段不画控件也不写 kind,改成多值类型时会去掉 kind;回滚即 revert 本 PR。
• 先看:ObjectFieldInspector.tsx 的 patchOptions 与 changeType(写入与去除 kind 的两处),其次 classifyOption(非法 kind 走异常行)。
• 两分钟验证:pnpm --dir ui exec vitest run packages/app-shell/src/views/metadata-admin/inspectors/ObjectFieldInspector.optionKind-26729.test.tsx,预期 16 个测试全过。
• 需要你做的:无,席位落地。

What changes

• kind becomes an owned key of the picklist option editor. Option gains kind?: SelectOptionKind; classifyOption moves it out of the rest bag into its own slot (the docblock's "the three parts move together": Option, readOptions, patchOptions). The vocabulary is read from the installed contract (SelectOptionSchema.shape.kind.unwrap().options), not re-listed.
• A picker per option row (data-testid="option-kind", the shared Radix Select): "No kind" / Open / Won / Lost. "No kind" removes the key outright — absent stays absent, which is what lets ObjectSchema tell "declared open" from "forgot the key" under stageField.
• Only where the spec accepts it. The picker is drawn, and patchOptions writes kind, only on a single-value select / radio (takesOptionKind: SINGLE_OPTION_TYPES and multiple !== true, the same test FieldSchema applies). On a multi-value field no picker is drawn and no kind is written; a kind authored there by other means (which FieldSchema already refuses) is not written back. A retype to a type that cannot carry one drops the kinds, as a retype already drops a picklist the new type cannot name; a retype to radio keeps them.
• An out-of-vocabulary kind (kind: 'closed') is a malformed row (kind-not-valid), carried verbatim like the other malformed shapes, rather than silently dropped.
• The lifecycle hint. When the object names this field as its stageField and an option still has no kind, the inspector shows one held-hint line saying so; it clears once every option has one.
• i18n. New designer.field.* rows in the studio namespace, en and zh: optKind, optKind.unset / .open / .won / .lost, optMalformed.kindNotValid, hint.stageKinds. No engine.studio.settings.* row. Patch changeset for @object-ui/i18n (app-shell is private).
• Not touched: ObjectSettingsPanel.tsx (the Lifecycle picker itself), the spec, the engine.

Why

The Lifecycle field picker refused a Studio-authored select (Option 'new' of field status declares no kind), and the option editor had no kind control: the triage grade asks for the control for the published option kind (#23809).

Tests

New ObjectFieldInspector.optionKind-26729.test.tsx (16 tests), every case ending at ObjectSchema / FieldSchema or the written document's keys:

• CONTROL: the Studio-built select is refused as the stage field at fields.status.options.{0,1,2}.kind.
• Picking Open / Won / Lost writes the three kinds, and ObjectSchema then accepts the object with stageField: 'status'; with one option left unset, the refusal names only that option.
• "No kind" removes the key; an existing kind round-trips through an edit of its own row and another row; the picker shows the stored kind.
• multiselect / checkboxes / select with multiple: true: no picker, no kind written, FieldSchema accepts.
• A kind authored on a multiselect is not written back; an out-of-vocabulary kind is a malformed row kept verbatim.
• A retype to Multi-Select drops the kinds; a retype to Radio keeps them; FieldSchema accepts both.
• The lifecycle hint appears while an option lacks a kind and clears once all have one; CONTROL: no hint when the object names another stage field.
• zh: the picker's name and the kind label come from the studio catalogue (await loadStudioCatalogue('zh') at import).

Ablation (one-shot, through scripts/ablation-replace.mjs, mutation committed-then-restored): deleting if (o.kind && kindAllowed) out.kind = o.kind; from patchOptions turned 5 of the 16 tests red (the three authoring cases, the round trip, the hint); restore verified blob == HEAD and git diff HEAD empty.

Acceptance notes

• ObjectSettingsPanel's Lifecycle picker is unchanged; the order marked that line optional and the inspector's hint covers what the picker would have said.
• Two whole-tree scans in ui/scripts/__tests__ (check-i18n-dead-keys.test.ts "on the real tree", check-vi-mock-inherit.test.ts "the sweep") each hit their 15 s test budget once on this shared box during the full vitest run scripts/ (a different test each run); each file passes alone. Load-sensitive, not this diff; CI's run is the verdict.

Verification

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic. (spec untouched)
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint passes. (CI owns pnpm lint)
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset, and both declaration lines are filled in here and in the changeset.
• 6. packages/spec touched: n/a.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English.

Generated by Claude Code