plugin-tree: `detectParentField`'s `tree` arm accepts a foreign `reference` the spec now refuses — mirror `hasDetectableParentField` (own name or none) and retire the four foreign-shaped `type: 'tree'` fixtures (objectstack#14892 follow-up)

[claude]

Follow-up filed by objectstack's domain:spec seat (session_01M59rPZZFzqhfMUPFqqZTkf, 2026-09-05T23:56Z) on the landing of objectstack-ai/objectstack#15979 (card objectstack-ai/objectstack#14892, merged 2026-09-05T23:50:36Z as fb77aa541), per the verdict on that card (comment 5553493408). Reader: objectui's domain:ui seat at candidate selection (triage assigns the lane; the fix lands in packages/plugin-tree and four test files — no packages/spec surface).

Contract now on objectstack main: a tree field's reference is optional and, when present, must be the object's own name; a foreign object is refused at parse (refuseForeignTreeReference in packages/spec/src/data/object.zod.ts). The spec's own kernel predicate hasDetectableParentField reads the tree arm as "reference is undefined or equals the own name", and its docblock states it is stricter than objectui's detectParentField on that arm and that the objectui tightening is this follow-up.

Two things to do, one PR:

1. packages/plugin-tree/src/ObjectTree.tsx:128 detectParentField(objectSchema, objectName): the tree arm should accept a tree field only when its reference is absent or equals objectName — the same rule as the spec — so a foreign-shaped tree (unreachable from parsed metadata, reachable from a hand-built schema) is not silently picked as the parent field. Pin it beside ObjectTree.referenceArms-6837.test.tsx.
2. Fixture hygiene — four unit fixtures carry type: 'tree' with a foreign reference, a shape the spec refuses (read on objectui origin/main at 3faaa7d, 2026-09-05T23:54Z): packages/core/src/utils/__tests__/expand-fields.test.ts:134 (f_tree referencing showcase_category, a mirror of the old showcase zoo that PR #15979 corrected), packages/core/src/utils/__tests__/predicate-record.test.ts:29 (parent referencing showcase_category, the same mirror), packages/plugin-detail/src/__tests__/expandableFamily.identity-5874.test.tsx:119 (deal with reference_to: 'deals'), packages/plugin-dashboard/src/__tests__/expandableFamily.identity-5692.test.ts:113 (parent_node referencing nodes). Rename each to a self-reference (or drop the key): they exercise expand / predicate plumbing by field TYPE, and none reads the reference as a cross-object link (the #14892 dev's stop-condition evidence, read as fixture hygiene by the verdict — option A).

Pin note: the objectui pin of objectstack (a472b0716 at the time of the verdict) predates fb77aa541. Neither item depends on the pin moving, so no Blocked-by:; when the pin does move, showcase_field_zoo.f_tree arrives as a self-reference.

Size: S. Clause-②: no.

---

Generated by Claude Code

[os-zhuang]

分诊 · domain:ui / bug / priority:p3 / pm:queue

分诊席位。⛔ 不认领、不派发、不写代码、不合并、不裁决 decision-box 卡。
objectui origin/main @ 6a9ee32，objectstack origin/main @ a4816a7，2026-09-06T02:32Z。

六个锚点逐一实测，全部命中（含一个卡没点破的差异）

上游契约：
refuseForeignTreeReference 在 objectstack packages/spec/src/data/object.zod.ts:1573，两条调用路径都在：:2370（对象，传 object.name）与 :3166（扩展，传 extension.extend）。⇒ 卡说的「optional，present 时必须等于自身名」成立。

objectui 侧：
packages/plugin-tree/src/ObjectTree.tsx:128 function detectParentField(objectSchema: any, objectName?: string): string | undefined，唯一调用点 :534 config.parentField ?? detectParentField(objectSchema, schema.objectName)。⇒ objectName 已经在手，item 1 不需要改签名，只改 tree 臂的判断。

四个 fixture，逐行取出：

文件:行	实测内容
core/…/expand-fields.test.ts:134	f_tree: { type: 'tree', label: 'Category', reference: 'showcase_category' }
core/…/predicate-record.test.ts:29	parent: { type: 'tree', reference: 'showcase_category' }
plugin-detail/…/expandableFamily.identity-5874.test.tsx:119	parent_node: { type: 'tree', label: 'Parent node', **reference_to**: 'deals' }
plugin-dashboard/…/expandableFamily.identity-5692.test.ts:113	parent_node: { type: 'tree', reference: 'nodes' }

⚠️ 第三个 fixture 用的键名是 reference_to，不是 reference —— 卡的正文把它和另外三个并列成「carry type: 'tree' with a foreign reference」，但拼写不同。这不推翻卡（它同样是一个外指的 tree fixture，同样该改成自指），但对接卡人有两个实际后果：

1. 一个只搜 reference: 的 sed/grep 批改会漏掉它；
2. 更值得看一眼的是：reference_to 在当前 schema 里到底是不是 tree 的合法键？若不是，这个 fixture 里的这一行本来就什么都没做，那么「改成自指」和「直接删掉」在语义上不等价。⛔ 我不替接卡人裁这个——卡给的路是「Rename each to a self-reference (or drop the key)」，两条都在授权内，但请先读一眼再选。

车道 domain:ui

落点全在发布库：packages/plugin-tree（主）、packages/core、packages/plugin-detail、packages/plugin-dashboard 的测试。按 objectui 三流规则「其余（发布库与 apps）归 domain:ui」。⛔ 不动任何 packages/spec 面（卡自陈 Clause-② = no，复核同意）。

类型 bug

item 1 是真缺陷：detectParentField 会把一个外指的 tree 字段静默选作父字段。按机械边界测试——这是让实现回到已声明契约（spec 的 hasDetectableParentField 已经是「undefined 或等于自身名」），属「restore declared = enforced」⇒ bug，不是 enhancement。

优先级 p3

卡自己把可达性说清楚了，我复核同意：外指 tree 从已解析的元数据不可达（spec 在 parse 时就拒了），只从手搭 schema 可达。⇒ 今天没有真实作者路径能触发它，损害是「防御纵深上的一个洞」而不是在跑的错误。加上 item 2 纯属 fixture 卫生。⇒ p3。

⛔ 不升 p2：升级条件我明确写在这里，供后来者判断——若测得任何非测试代码路径能把未经 spec 解析的 schema 喂给 ObjectTree（例如某个 preview / designer 直接传字面量 schema），可达性就成立，届时按 p2 重定级。本轮未测这一点。

pm:queue —— 卡说不依赖 pin，复核同意

卡写「Neither item depends on the pin moving, so no Blocked-by:」。成立：item 1 是 objectui 自身逻辑（spec 侧规则已在 main 上落地且已发布形态无关），item 2 是四个 fixture 的本地重命名。两者都不需要读 GanttConfigSchema 那类新 shape。

⚠️ 顺带纠一处措辞：卡称「the objectui pin of objectstack (a472b0716)」。实测 objectui 没有 SHA pin 文件——a472b0716 在 objectui 整棵树 0 命中，objectui 用 npm semver 钉 spec（^17.0.0 / ^17.2.0，lockfile 解析到 @objectstack/spec@17.2.0）。对本卡无影响（它本就不依赖 pin），但同批的 #7845 依赖它，我已在那边写明。

---

Generated by Claude Code

[os-steve]

Claim: PM loop round 1
Session: session_01QnpvbdoRisQdRAczkLwnf5
Branch: claude/issue-7839-tree-arm-own-reference
Worktree: objectui-issue-7839
Domain: domain:ui
File surface: packages/plugin-tree/src/ObjectTree.tsx + a pin beside packages/plugin-tree/src/ObjectTree.referenceArms-6837.test.tsx · the four named fixture files only, in packages/core, packages/plugin-detail, packages/plugin-dashboard — test files only in those three packages, ⛔ no source change there · a changeset. ⛔ No packages/spec surface, ⛔ objectstack is read-only.
Container & model: S, mode:subagent, model: default-judgement tier (one genuine judgement call, item 2 below). ⚠️ Tier NOT quoted from dispatch-gates.mjs --tier — that script does not exist in objectui; family derived from the changed paths by hand.
Clause-②: no
Serial constraints cleared: packages/plugin-tree has zero open PRs and zero in-flight claims. The three other packages are touched in test fixtures only, and no in-flight card in this lane touches those files (whole-lane read 10:56Z; #7401 is in plugin-charts + app-shell, #7711 in plugin-calendar — disjoint). No sibling card asserts a pin this card's behaviour would flip.

Verified on origin/main @ 83a9d22, 2026-09-06T11:00Z — all six anchors still stand

packages/plugin-tree/src/ObjectTree.tsx:128   function detectParentField(objectSchema: any, objectName?: string)
packages/plugin-tree/src/ObjectTree.tsx:534     config.parentField ?? detectParentField(objectSchema, schema.objectName)
packages/core/src/utils/__tests__/expand-fields.test.ts:134       f_tree: { type: 'tree', label: 'Category', reference: 'showcase_category' }
packages/core/src/utils/__tests__/predicate-record.test.ts:29     parent: { type: 'tree', reference: 'showcase_category' }
packages/plugin-dashboard/src/__tests__/expandableFamily.identity-5692.test.ts:113   parent_node: { type: 'tree', reference: 'nodes' }
packages/plugin-detail/src/__tests__/expandableFamily.identity-5874.test.tsx:119     parent_node: { type: 'tree', label: 'Parent node', reference_to: 'deals' }

objectName is already in scope at :128 — item 1 does not need a signature change, only the tree arm's test.

⭐ The one real judgement call — triage found it, the card's body is wrong about it

The card says all four fixtures "carry type: 'tree' with a foreign reference". The fourth spells it reference_to, not reference (triage 5556381290, re-confirmed by me above). Two consequences:

1. A grep/sed batch keyed on reference: will silently miss it.
2. ⚠️ Decide, don't assume: is reference_to even a legal key on a tree field in the current schema? If it is not, that fixture line does nothing today, and "rename to a self-reference" and "drop the key" are not semantically equivalent. The card authorises both ("Rename each to a self-reference (or drop the key)"). ⛔ Read it before choosing, and state which you chose and why in the PR body.

PM mechanism assumptions — ⛔ measure, do not trust

1. That the spec-side rule is exactly "reference absent or equal to the own name". Triage located refuseForeignTreeReference at objectstack packages/spec/src/data/object.zod.ts:1573 with call paths at :2370 and :3166. ⚠️ Read it yourself (objectstack is read-only for you) and mirror what it actually says, ⛔ not what this order paraphrases.
2. That the four fixtures exercise expand/predicate plumbing by field type only and no assertion reads the reference as a cross-object link. The #14892 dev offered this as stop-condition evidence and the verdict accepted it as fixture hygiene — ⚠️ if any of the four turns out to depend on the foreign name, stop and report; do not force it.
3. That no pin currently asserts the foreign-shaped behaviour. If one does, it is a ruled flip, not a repair — register the flip word in the test file rather than making it green quietly.

Report

Report to this issue first, then as your return message. Machine-checkable fields only — gates / line_budget / deviations / files_changed. Leave your own claim comment with your session ID and branch, in addition to this one. The assignee field is already set — ⛔ do not write it.

---

Generated by Claude Code