Repository navigation
decision: objectui#11168 — action:button.undoable is declared, but the block path never captures Undo for operation: 'update' (no row stash): deliver it on the block (B) or narrow it off the spec row (C)? #11754
Description
Ruled: 6030342264 · letter B · 2026-10-07T03:35Z
Filing gate: ② a decision only the maintainer can make. Filed by the triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015, session_01AavokzJ5DndAwitDXvKy4U), on the maintainer's instruction 「需要我决裁的问题，立决策卡」. ⛔ Not a claim. Once ruled, the ruling is recorded on this card and on the parent, and this card closes.
Parent: objectui#11168 (p2, pm:on-hold). Its Restart-when: is this ruling and nothing else: "the maintainer rules B or C on action:button.undoable" (5982135645). The question was raised in the slice-1 dev report (5907196304, open question 2) on 2026-09-30, and no decision card carried it until now.
维护者速读
页面上作者配置的"更新"按钮（action:button，operation: 'update'）可以声明 undoable，spec 也接受这个写法，但点完之后没有"撤销"按钮。原因是运行时要靠一份行快照才能做撤销，而按钮区块从来不写这份快照。记录详情页上的 api 写法则能正常撤销。
- B：改 objectui 的运行时，让按钮区块在有记录上下文时（记录页的当前记录、表格的当前行）也能撤销，然后把这个键发布出去。不破坏任何东西，现在就能做。
- C：请 spec 把
undoable从action:button这一行收窄掉。这是破坏性收窄，只能放进 v18；而且记录页上原本好用的api撤销会一并失去这个已声明的键。
推荐 B。回一个字母：B / C。
四棱（os-decision-facets）
- ① 项目长远合理性：B 让 spec 的规范写法（
operation: 'update'）在区块上也兑现，符合"声明即兑现"。C 用删除的办法消除差距，代价是丢掉一条今天就能用的路径。 - ② 实际业务拉动：撤销是用户直接看得到的能力，作者在页面上配的更新按钮今天会"静默没有撤销"。目前没有测到具体客户需求。
- ③ 防 AI 犯错：B 之后，AI 按 spec 写出的
undoable在区块上真的生效。C 之后，AI 写这个键会被拒，信号是响亮的，但能力也就没了。 - ④ 创业阶段不扩散：B 是 objectui 内部一处运行时改动，加上它的 pins，不新增键。C 要动一次 spec，而且必须等 v18。
Prior rulings read: undoable / ActionSchema fulfilment → ruling 5754211444 (objectstack batch #203 item 5, letter C, landed as objectstack-ai/objectstack#19635). Its rule: undoable: true is legal only on a shape some runtime fulfils (operation: 'update' through the framework runtime; type: 'api' through the console), and is refused elsewhere. That rule is why D (publish with a caveat) is out. It does not choose between delivering and narrowing on the block. check-prior-rulings on these terms: 0 further hits. Thread: objectui#11168 5907196304, 5907833984, 5982135645.
一句话问题
spec 对 action:button 声明了 undoable，但按钮区块上 operation: 'update' 这种规范写法从来做不出撤销。是让它在区块上兑现（B），还是从 spec 行里收窄掉（C）？
选项 × 代价
| 选项 | 做什么 | 代价 |
|---|---|---|
| A（现状） | 继续不发布，记账留在 objectui#11168 | 卡片永远挂着；按钮上声明了也不生效 |
| B | @object-ui/core 的 update 路径在区块调用时，从作用域里的记录（记录页的当前记录、表格的当前行）取撤销基准；有基准时生效，然后发布 undoable |
一处运行时改动加 pins。没有记录上下文的独立按钮仍然无法撤销，description 要写明这一点（这是声明写法所需的上下文，不是附带的免责说明） |
| C | 请 spec 从 action:button 行收窄掉 undoable |
破坏性收窄，要等 v18；记录页的 api 路径失去一个今天就兑现的已声明键 |
| D | 原样发布，加一句说明 | 声明了却做不到，5754211444 已排除这种做法 |
推荐 + 回退 + 置信缺口
- 推荐 B。 它兑现的是 spec 自己的规范写法，
5754211444已认定框架运行时兑现operation: 'update'。它不破坏任何东西，现在就能在 objectui 落地，并且不让任何已兑现的路径倒退。 - 两年后的样子： 区块、记录页、列表上同一个
undoable行为一致，只要有记录上下文就能撤销，与动作从哪个界面发起无关。 - 回退 C，在 v18 里做。
- 置信缺口： 没有量过有多少已存储页面在
action:button上写了undoable。B 的认领方先盘点一遍。 - 只看①选 B；②③④ 是否翻转：否。
裁后执行
- B： objectui#11168 回到
pm:queue，下一个切片做 B，加上发布和 pins。分诊记录裁决。 - C： 在 objectstack 立 spec 收窄卡，带
target:v18，并挡在 #15193 之后。objectui#11168 的这条记账改为指向那张卡。
Activity
objectstack-fleet commentedon Oct 7, 2026
Ruling: batch #284 item 1 · letter B · maintainer 「同意」 2026-10-07T03:33Z
Director seat, summon #35,
session_01VYToj6PQehTEKNrjGM9akg(GitHubos-zhuang; written asobjectstack-fleet[bot]via the relay). Batch #284 was presented in chat with this card's options B / C (A and D as the card lists them); the maintainer answered 「同意」, which takes the recommendation. Freshness gate: the body (no comments) and the parent's thread (objectui#11168: the hold correction 5982135645, triage's pointer 6029656672) were re-read before this record; nothing newer exists. Thread-read: none on this card.The ruling
B. The button block delivers
undoablewhere a record is in scope. The runtime already captures Undo whenever the invoking surface hands it the row (packages/core/src/actions/ActionRunner.ts:1788:action.undoable && rowRecord && writtenFields.length > 0), and the surfaces that do so today are the record page's declared-actions bar, the related-record bridge, the layout containers and the grid's rows; theaction:buttonrenderer hands it nothing. The change is at the block: when the block runs inside a record context (the record page's current record, a table's current row), it passes that record as the Undo baseline the runner already reads; the key is then published onaction:button, with its description stating the one limit (a standalone button with no record in scope offers no Undo, because there is no row to restore). ⛔ Not taken: C (a breaking narrowing of the spec row, v18 only, which would also contradict the closed table the 5754211444 ruling fixed, whereoperation: 'update'is the shape the framework runtime fulfils, and would strip an already-fulfilled key from the record page'sapipath), A (the entry stays booked forever) and D (publishing a key that does not deliver, which 5754211444 already excluded).Why B: the spec's closed table of fulfillable shapes is already ruled; B makes the block honour it. In two years the same
undoablebehaves the same on the block, the record page and the list: wherever a record is in scope, the action can be undone, whatever surface it started from.State
completedin this act; the pointer is recorded on objectui#11168, whoseRestart-when:(5982135645: "the maintainer rules B or C onaction:button.undoable") is met by this ruling, so the hold is released there in the same act (pm:on-hold→pm:queue, no assignee). The next slice on spec parity: the ten blocks @objectstack/spec 17.5.0 newly carries, with their unpublished inputs, refused arms and member pins — declare each by measurement (objectui#11111 decision 3 = B) #11168 does B with the publication and its pins; the claimant first counts the stored pages that declareundoableonaction:button(the card's unmeasured gap).Generated by Claude Code