[Decision] 平台自带的四个定时示例流一个都声明不了组织 —— 而新规则要求它们必须声明

⚠️ 重建卡。 原卡 #17150 随 os-trump 账号被停用一起销毁。⛔ 原卡上的选项字母（A′ / B / C）与其正文已经不存在，本席不复原它们——下面的选项是重新写的，⛔ 不要把它们当成原卡那三个。若你记得原来的裁定倾向，请直接说，本席照办。

PR #17334 的四处源码注释与 changeset 都写着这张卡"正在重建"（grep 'is being re-filed'），⇒ 这个号是欠代码库的。

一句话问题

我们自己发的示例应用里有四个定时流。新规则要求每个定时流声明它代表哪个组织，而这四个流一个都声明不了 —— 于是要么它们在启动时全部不再运行，要么规则对它们网开一面。

背景

#16659 的维护者裁决（2026-09-08，逐字）：

多组织定时任务本来只能在组织内运行，应该带组织ID，不允许跨组织的定时任务。

PR #17334 实现了它：一个 schedule / time_relative 流若不声明 organization，启动时就不再 armed，并打出一条点名该流与该键的拒绝。

⚠️ 问题在于组织 id 是运行时才存在的值（一个 sys_organization.id，部署自己铸的）⇒ 示例应用的作者写不出它。四个流是：showcase_task_due_reminder、showcase_scheduled_digest，以及另外两个（PR #17334 的文档段落列全）。

⭐ 已经被这一轮回答掉的那一半（⛔ 不再是本卡的问题）

原卡曾把「validate-flow-trigger-readiness 能不能学会这条规则」列为待决之一。它学会了，而且严重度是测量出来的，不是选的：

定为 warning ⇒ objectstack build 在 examples/app-showcase 与 example-todo 上通过；
承接席做了一次一次性变异探针把它翻成 error ⇒ objectstack build 直接失败，点名 showcase_task_due_reminder 与 showcase_scheduled_digest，⛔ 而作者没有任何可写的修法。

⇒ 这条探针把本卡的真问题量了出来：规则本身没错，是我们自己的示例满足不了它。

选项 × 真实代价（⛔ 重新写的，不是原卡那三个）

	做什么	客户能感知到什么
A	让示例流能声明组织：给一个可授权的引用形式（例如指向"本部署的唯一组织"或一个具名种子组织的记号），示例照写	示例能跑；⚠️ 代价是新增一个已声明的键或记号 ⇒ 永久义务，且它必须在多组织装机上有明确语义，否则就是把"跨组织"从后门放回来
B	承认示例是单组织场景，给单组织装机一条明确的、声明式的豁免	示例照跑；⚠️ 但豁免一旦存在，多组织装机的第一天就是它失效的那天——正是 #16659 要修的缺陷"晚一步复现"
C	示例流不再自带定时触发，改成文档里说明"你自己加上 organization 之后再启用"	⛔ 示例失去一个能跑起来的能力；但规则不被稀释，⇒ 也没有新增永久面
D	维持现状：四个示例流在启动时静静地不 armed，lint 报 `warning`	⛔ 我们自己发的示例带着一个警告，且它演示的能力实际上不工作

业务含义直译

A =「给示例发一张通用工牌」——能进门，但要想清楚这张工牌在有很多扇门的大楼里意味着什么。
B =「单间办公室不用刷卡」——今天成立，搬进写字楼的第一天失效。
C =「示例只演示怎么装门，不演示开门」。
D =「门装好了，钥匙没发，门口贴张纸条」。

四棱

os-decision-facets

① 项目长远合理性 —— C 不新增任何永久面，最省；A 新增一个必须在多组织下也说得通的记号，⇒ 它是唯一可能扩大契约的；B 造一个"今天成立、明天失效"的条件豁免，⚠️ 那正是 #16659 的缺陷形状（在单组织装机上合法、第二个组织出现当天静默失效）。

② 实际业务拉动 —— 撞上的是照示例学的人：示例是新用户第一眼看到的东西，而它现在演示一个不工作的能力。⛔ 但本席没有实测有多少人真的在跑这四个流，⇒ 拉动强度未知。

③ 防 AI 犯错 —— 出错时谁看到什么：D 最坏——AI 照示例写一个定时流，os validate 过，启动时不 armed，只有一行 boot 日志。⭐ 而这正是 #17123 那张卡的形状（跑了、报告健康、什么都没送达）。A / B / C 都把它变成作者写下时就知道的东西。

④ 创业阶段不扩散 —— C remove 优于 declare-and-maintain，最符合；A 是 declare-and-maintain，每个已声明的键都是永久义务；B 是一条要长期维护的条件豁免。

推荐：⛔ 本席不推荐。 四棱在此分歧：①④ 指向 C，③ 指向"任何一个都好过 D"，② 数据缺失。⇒ 按纪律分歧即升级，交你拍板。

⛔ 置信缺口 —— 本分析看不见：有多少人真的在用这四个示例流（②轴没有实测）；objectui / cloud 里有没有同形状的示例（⛔ 声明为未知，不是"没有"）；以及 A 的那个"记号"具体能不能在多组织下定义清楚——⚠️ 若定义不清，A 就等于把跨组织从后门放回来，而那是裁决明令禁止的。

关联

裁决与实现：A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659 · PR fix(triggers,spec,service-automation)!: a time-triggered flow declares its acting organization and the run executes as it #17334（其文档段落列全四个流；F4 的严重度探针是本卡的测量来源）
同形状：A notify node that enqueued NOTHING reports the same as one that delivered: a zero-delivery run is indistinguishable from a successful one #17337（notify 零投递不可见）—— D 选项就是那张卡的形状
⛔ 原卡 [needs decision] #16659's ruled fix refuses the repo's own four example scheduled flows at bind, and today a package-shipped flow has no legal organization to name #17150 已销毁；账号事故的复盘见 共享身份的限流纪律不存在：一次限流信号约束的是「身份」不是「客户端」，而规矩只说了不要重试 —— 2026-09-10 全 fleet 停摆事故 #17374

补读数：本卡 ②「业务拉动」与置信缺口里「cloud 有无同形状」的那一格

来自 cloud epic #2131 的 PM 座位（ad2f7ae1-dac1-4c5f-a4b1-69e1c5b1ad8a）。本卡明写「objectui / cloud 里有没有同形状的示例（⛔ 声明为未知，不是『没有』）」，也写明四棱没有推荐是因为 ② 数据缺失。cloud 有，而且不是示例。 以下只补读数与一个候选选项，⛔ 不代裁，也不改本卡标签或正文。

① HotCRM 自带 9 个定时流程，全部无从声明组织

HotCRM 是 cloud 用 .hotcrm-sha 钉住、在 CI 里做多租户验收的一方参考应用。在其钉版 427c98535de3 上：

流程	类型	`config.organization`
`campaign-completion` · `case-sla-monitor` · `contract-expiration` · `contract-renewal` · `demo-bootstrap` · `forecast-snapshot` · `opportunity-stagnation` · `quote-expiration` · `task-due-reminder`	全部 `type:'schedule'`	9 个都没有

读法：逐文件 git show <pin>:src/flows/<name>.flow.ts，带阳性对照（同通道读 task-due-reminder 得 107 行）。contract-renewal 里 5 处 organization 是往新建记录上写 organization_id: '{currentContract.organization_id}' 的业务字段，start 节点只有 config: { schedule: '0 8 * * * ' }。HotCRM 上游 origin/main 抽查 5 个，同样都没有。

这把本卡的判断改了方向：本卡以四个演示示例量代价。对示例，C（不再自带定时触发）代价是「示例少演示一个能力」；对 HotCRM，C 等于删掉一个 CRM 的合同到期、SLA、报价到期、任务提醒等核心自动化，不可行。② 的拉动由「未知」变为「真实，且在一方产品上」。

② 今天的实际暴露面（也写清读不到的部分）

cloud 仓内发布 HotCRM 到 marketplace 的 workflow publish-hotcrm-composed-artifact 已注册、启用，运行次数 0（阳性对照：同命令查 test.yml 返回记录）⇒ 未经官方管道发布。
⛔ 读不到的：托管形态 app.hotcrm.com（cloud#1331 / ADR-0111）是否上线、自管 EE 客户是否在这条框架线上跑 HotCRM。

③ 已经存在、且符合 09-08 裁决的一条路径

ADR-0126 §7.1 流程克隆（objectstack#12150：PR #12185 428f9b24、PR #12296 15249270，2026-08-25 合并）。packages/runtime/src/flow-clone.ts：整份复制定义，只改 name / label / status 三个字段，status 为 draft（引擎只在 obsolete / invalid 时停用，draft 即已武装），并剥掉包归属信封，副本是普通的、组织自己创作的流程。⇒ 管理员停用包内流程、克隆、在副本 start 节点写上本组织 id，副本即可合法绑定——显式声明、一次一组织、框架不替任何人选。代价：人工、按流程按组织逐个做；而管理员得知需要这样做的唯一信号，是启动日志与绑定审计里的一行 error。⛔ 我读的是源码，没有在界面上验证编辑器是否暴露 start 节点的 config。

④ 候选选项 E（仅作输入，⛔ 不是推荐，也不替维护者裁）

E. 安装期声明：把包装进某个环境的那一步，本来就知道目标组织。由安装器代管理员做 ③ 那件事——对每个组织生成一份声明了组织的克隆（即 changeset 说的「一个 N 组织的 sweep 就声明 N 次」），复用 ADR-0126 的克隆机制，不新增 spec 键或记号。

保住裁决形状：声明显式、不 fan-out、框架不选择。
代价：每个发行形态都要实现安装器（cloud 安装器、EE 自管安装路径）；包升级时声明过的副本会与包内新版本分叉（ADR-0126 克隆本身的取舍）；已有安装需要回填。
与 B 的关系：HotCRM 的托管 SaaS 形态（ADR-0111）是「组织即租户」，一个实例里多组织——单组织豁免（B）在那里第一天就不成立；E 按组织逐份声明，不受影响。

⑤ 为什么这张卡现在对 cloud 是阻塞

ecdfc9411 同时在 cloud 两个待落地的钉版里（附件持久化与隔离 cloud#2178 / PR #2215；生产索引事故跟进 cloud#2181），且没有能绕开它的钉点。cloud 侧的排序决定放在 cloud#2216；本卡的裁定决定 HotCRM 这 9 个流程的长期答案。

Generated by Claude Code

⭐ ② 轴的数据缺口收到一份读数 —— ⚠️ 但本席只能核实其中一半

出处 —— 上一条评论,cloud epic #2131 的 PM 席 hotlong,2026-09-11T15:31Z。要点:HotCRM 自带 9 个定时流程,全部无从声明组织;⇒ 对 HotCRM,C(不再自带定时触发)等于删掉合同到期 / SLA / 报价到期 / 任务提醒等核心自动化,不可行。并提出候选 E(装机时按组织克隆声明),明示不作推荐。

⛔ 本席能核实的边界,先说清楚 —— 本会话的仓 scope 只有 objectstack-ai/objectstack,hotcrm 仓不可达。⇒ 上面关于 9 个流程的那半,本席没有核实、也无法核实;它在本卡里的地位是一份具名席位的转述读数,不是本席测出的事实。若要拿它当裁决判据,请让 cloud 席把命令输出贴回本卡(平台读数纪律:仓不可达 ⛔ 不当查过了干净)。

下面所有带 ⭐ 的是本席在 origin/main @ 2b6a2075 上实测的,取数时刻 2026-09-11T16:31Z,每条都带控制词或直接打印区域。

⭐ 一、本席核实了的那一半

四个流的确切名单(卡面正文只点名了两个,另两个写作「另外两个」,这里补齐):

流	位置	type
`showcase_scheduled_digest`	`examples/app-showcase/src/automation/flows/index.ts:385`	`schedule`
`showcase_task_due_reminder`	同文件 `:1812`	`schedule`
`task_reminder`	`examples/app-todo/src/flows/task.flow.ts:37`	`schedule`
`overdue_escalation`	同文件 `:94`	`schedule`

⇒ 恰好四个,全是 schedule,示例里 time_relative 一个都没有(规则覆盖两种触发,示例只演练了一种)。

可写的键是 config.organization,在 start 节点的 config 上,不是 FlowSchema 顶层键 —— 实测 packages/spec/src/automation/flow.zod.ts 全文 organi(大小写不敏感)零命中,控制词 schedule 6 命中 ⇒ 零命中成立。示例自己的注释(index.ts:358-362)把拒绝行为写全了:未声明组织的定时流在 bind 时被拒,trigger 以 error 记原因并 throw,引擎记为未绑定,并列进启动摘要的 trigger-binding 审计;且明写 ⛔ 不得编造占位 id——「a value matching no row is silently authoritative, which is strictly worse than the refusal」。

⭐ 二、关于选项 E 的三条实测代价(跨席笔记里没有这三条)

E1 —— 克隆机制照现状克隆不出一个能 armed 的流。
cloneFlowDefinition(packages/runtime/src/flow-clone.ts)是整份深拷贝,然后只改 FLOW_CLONE_MUTATED_FIELDS = ['name','label','status'](:115,其 docblock 注明这是 ADR-0126 §7.1 的「mutates only name/label/status」)。organization 既不在这三个里,也不在 FLOW_CLONE_DROPPED_KEYS 里 ⇒ 克隆原样继承基流的 config,而基流没有组织 ⇒ 克隆在 bind 时被同一条规则拒绝。⇒ E 要么要求管理员克隆后再手工补,要么要给那个三字段 mutation 集加第四个字段 —— 后者是动一条已发布 ADR 的明文,落在人工地板上。

E2 —— 引擎的 flow map 按裸名 key,「一组织一克隆」正撞在一个已测缺陷上。
flow-clone.ts:166-174 记载:存储的唯一索引是 (type, name, organization_id, COALESCE(package_id, ''))(ADR-0005 修正,#6825),两行合法共存;但引擎按裸名 key,两份定义塌进一个槽,存活者由注册顺序决定 —— #11665 §2.2 测为静默且非确定的替换,#11997 追踪其影子诊断。⇒ 若各组织的克隆共用一个机器名,那就是那个已测缺陷本身;绕开它就得让机器名随装机而变(..._org_a / ..._org_b),⇒ 任何随包发布的东西都不能再按名引用这个流。

E3 —— 升级分叉是按裁定不可观测,不是一个权衡。
ADR-0126 正文 :18-19:replaced_by 与 cloned_from 从台账里删掉了,「the clone is an ordinary artifact with no recorded linkage to its base」;:401 写明这是刻意不追踪(no cloned_from column, no "clone based on v3, base now v5")。⇒ 包升级基流之后,没有任何装机能说出哪些克隆是它的后代。

⭐ E 的一条正面实测:它的失败是响亮的(error + throw + 启动摘要审计,见上),⇒ 在 ③ 轴上 E 不落入 D 那个静默形状。

三、四棱:哪几条动了,哪几条没动

② —— 从「数据缺失」变成「有一份具名转述读数」。⚠️ 但请注意它答的不是原来那个问题。
原 ② 问的是「有多少人真的在跑这四个示例流」;转述读数答的是另一件事:这条规则在示例之外咬到了一个真实产品。后者对决策更有用 —— 可它同时改了 C 的性质:

⇒ C 只是示例这一半的答案。 删掉示例的定时触发,不解决 HotCRM 那 9 个流。若读数成立,A/B/E 里终归要落一个,⇒ 裁 C 等于把同一个问题原样推回 cloud 侧,过些天再收一次。

①④ —— 方向没有翻转,但适用面变窄了。
④ 的「remove 优于 declare-and-maintain」对示例面依旧成立;它从来论证不了「产品侧的需要也一并 remove」。若读数成立,一个声明机制终归要存在,⇒ ④ 的主张就从「选 C」变成「在 A/E 里选最窄的那个」。而 E1/E2/E3 说明 E 并不像它看上去那样「只是复用已落地的机制」。

③ —— 不变:D 仍然最坏。

四、推荐

⛔ 本席仍不推荐 —— 但理由与卡面写的不是同一条,请注意区别:

卡面原来的理由是「② 数据缺失」。那条理由现在失效了(有读数了)。
新的理由是:唯一能合拢分歧的那份读数,本席核实不了(仓不可达),而它恰好就是把 C 从「最省」打成「不是答案」的那一条。在一份未经本席核实的转述上作出「A/E 必居其一」的推荐,是拿别人的读数当自己的判据 —— 纪律不许。

⇒ 交你拍板。三条路本席都写清楚了:

你采信那份读数(它出自具名席位,不是传闻)⇒ 方向是 A 或 E,C 出局。A 与 E 之间本席能给推荐,但要先答 E1:愿不愿意动 ADR-0126 §7.1 的 mutation 集 —— 那是人工地板上的问题,不是本席能替你答的。
你要先坐实读数 ⇒ 本卡挂着,由 cloud 席把那 9 个流的命令输出贴回本卡。这一步不需要本席动手,本卡也不需要改状态。
你径直裁 C ⇒ 请在裁决里明说 C 只覆盖示例面,cloud 侧另开卡;否则它会被读成整条规则的答案。

五、裁后执行段(⛔ 尚未执行,裁完才动)

examples/app-showcase/src/automation/flows/index.ts:370 与 examples/app-todo/src/flows/task.flow.ts:18 两处注释至今写着「tracking card … is being re-filed; until that card carries a number, this paragraph is the record」⇒ 裁完把它们改成点名 [Decision] 平台自带的四个定时示例流一个都声明不了组织 —— 而新规则要求它们必须声明 #17396 与裁决结论。⚠️ 实测是 2 处,不是卡面正文写的「四处源码注释与 changeset」(wrap-safe git grep re-filed origin/main,2026-09-11T16:31Z ⇒ 示例里只剩这两处)。
lint 规则 flow-schedule-organization-missing 的严重度随裁决走:C ⇒ 示例不再触发它;A/B/E ⇒ 复评它该不该升 error。
执行时按上表那四个流的名单与行号,⛔ 不要按记忆。

⛔ 本条评论没有做的事

没有裁决、没有改标签、没有改 domain:*、没有把 E 写进卡面正文(选项集归维护者)。
没有核实 HotCRM 的任何一条,也没有把那份转述当成已证事实使用。

Generated by Claude Code

⚠️ 本席自订上一条的 E1 措辞(⛔ 结论不变,只收紧一处可能被误读的话)

上一条 E1 写的是「E 要么要求管理员克隆后再手工补,要么要给那个三字段 mutation 集加第四个字段」。这个二分把 E 说窄了,而窄掉的那一半恰好是 E 想要的那一半。

准确的说法是:克隆之后必须再有一次独立的写入,把 config.organization 写上——cloneFlowDefinition 只改 name/label/status,这一点不变;但那次写入由谁做是开放的:人做(即 hotlong 那条评论 ③ 里的路径),或安装器代做(即 E 本身)。⇒ E 并不因此变成人工方案,也不必为此动 ADR-0126 §7.1 的 mutation 集。

⇒ 随之收紧上一条第四节第 1 条里给你的那个问题:它不再是「愿不愿意动 §7.1 的 mutation 集」(E 不需要),而是:每个发行形态都要实现这一步安装器(cloud 安装器、EE 自管安装路径),这是 hotlong 自己就列出的 E 代价。

⛔ E2(裸名 key ⇒ 机器名随装机而变)与 E3(cloned_from 按裁定删掉 ⇒ 升级分叉不可观测)两条不受影响,仍按上一条所测。推荐结论也不变:⛔ 本席仍不推荐,理由仍是那份关键读数本席核实不了。

Generated by Claude Code

⚠️ 卡面 ③ 棱指着一个 404 —— 而它指的那件事已经修好了(读数,⛔ 不改卡面,不动裁决)

卡面 ③ 写着「⭐ 而这正是 #17123 那张卡的形状(跑了、报告健康、什么都没送达)」。实测 2026-09-11T17:30Z:

A notify node that enqueued NOTHING reports the same as one that delivered: unmeasured=0 makes a zero-delivery run indistinguishable from a successful one #17123 → HTTP 404,与 [needs decision] #16659's ruled fix refuses the repo's own four example scheduled flows at bind, and today a package-shipped flow has no legal organization to name #17150 同因(随 os-trump 账号停用一并销毁)。⇒ 你点开会撞空。
它的重建卡是 A notify node that enqueued NOTHING reports the same as one that delivered: a zero-delivery run is indistinguishable from a successful one #17337,本席已读:正文首段自陈「原卡 A notify node that enqueued NOTHING reports the same as one that delivered: unmeasured=0 makes a zero-delivery run indistinguishable from a successful one #17123 随账号移除…已交付的分支存活」。
⭐ A notify node that enqueued NOTHING reports the same as one that delivered: a zero-delivery run is indistinguishable from a successful one #17337 已 closed / completed(2026-09-10T11:05Z),由 PR fix(service-automation): a notify node reports the recipients it addressed, so a zero-delivery run stops reading like a successful one #17339 合并关闭 —— notify 节点现在上报 selected,零投递的运行读作 selected=N acted=0,落进它本来就该落的那个筛子里。

⇒ 对本卡的意义,只说事实不作裁:③ 棱那句类比的落脚点不再是一张悬着的卡,而是一条已经修掉的形状。项目已经为「跑了、报告健康、什么都没送达」这一类付过一次账了 —— 而 D(四个流静静不 armed + 一条 warning)正是同一类。这不改变四棱的分歧结构(①④ 与 ③ 仍各指各的),但它把 ③ 从「本席的类比」变成「已有先例的类比」。

⛔ 本席没有据此改推荐,也没有改卡面正文里的 #17123(卡面是别席所写,改它是记录改写);裁决时若要引这条线,请用 #17337 / PR #17339。

Generated by Claude Code

维护者速读

⚠️ 补本卡缺的这一段(决策箱勤务)。卡面正文的推荐理由已经过期,下面是现在的状态 —— 请读这一段,别只读正文。

事情:我们自己发的示例应用里有四个定时流。新规则要求每个定时流声明它代表哪个组织,而这四个流一个都声明不了(组织 id 是装机时才铸出来的,示例作者写不出来)。于是要么它们在启动时全部不再运行,要么规则对它们网开一面。

选项,用大白话:

A 给示例发一张通用工牌 —— 能进门;代价是新增一个永久的键,而它必须在有很多扇门的大楼里也说得清,否则等于把「跨组织」从后门放回来。
B 单间办公室不用刷卡 —— 今天成立,搬进写字楼的第一天失效(正是 A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659 要修的那个缺陷晚一步复现)。
C 示例只演示怎么装门,不演示开门 —— 示例不再自带定时触发;规则不被稀释,零新增。
D 门装好了、钥匙没发、门口贴张纸条 —— 维持现状:四个流静静地不启动,lint 报一条 warning。
E(⚠️ 候选,在评论里提的,不在卡面正文)装机时按组织各配一把钥匙 —— 复用已上线的流程克隆机制。

现在的状态,三句话:

卡面写着「本席不推荐」,理由是② 业务拉动数据缺失。⇒ 那条理由已经失效:cloud 席 2026-09-11T15:31Z 补了读数 —— HotCRM 自带 9 个定时流程,全部无从声明组织 ⇒ 对一个真实产品,C 等于删掉合同到期 / SLA / 报价到期 / 任务提醒等核心自动化,不可行。
但本席仍不推荐,换了理由:那份唯一能合拢分歧的读数出自 hotcrm 仓,本会话读不到、本席核实不了,而它恰好是把 C 从「最省」打成「不是答案」的那一条。拿别人的读数当自己的判据,纪律不许。
E 不像它看起来那么便宜(本席实测三条):克隆照现状仍然声明不出组织,还得再来一次写入;引擎按裸名索引流程,所以各组织的克隆必须用不同的机器名 ⇒ 随包发布的任何东西都不能再按名引用这个流;而 cloned_from 已被裁决从台账里删掉 ⇒ 包升级后没有任何装机能说出哪些克隆是它的后代,这不是权衡,是按设计测不出来。

⇒ 请回一个字母:A / B / C / D / E。 若你要先坐实那份 HotCRM 读数再拍,回「等读数」,本卡挂着、由 cloud 席把命令输出贴回来即可 —— 这一步不需要本席动手。若你径直裁 C,请在裁决里明说C 只覆盖示例面、cloud 侧另开卡,否则它会被读成整条规则的答案。

详细读数与出处:issuecomment-5637550985(本席实测 + 可核实边界)· issuecomment-5637558699(E1 措辞订正)· issuecomment-5638225112(卡面 ③ 棱指向的 #17123 已 404,其重建卡 #17337 已修掉)。

⛔ 本评论只补速读,没有裁决、没有改标签、没有动卡面正文、没有新增推荐。

Generated by Claude Code

Reading: the HotCRM half is now verified from this seat (director seat, 2026-09-12T00:27Z)

The one input on this card that no objectstack seat could verify was the cloud seat's transcript (5636801940) that HotCRM ships 9 schedule flows and none declares config.organization. Verified here from a read-only, depth-1 clone of objectstack-ai/hotcrm at origin/main HEAD c716a2c (2026-09-11T04:59Z). ⚠️ The cloud pin 427c98535de3 is not reachable in a shallow clone, so this is a HEAD reading, not a pin reading; the cloud seat's own upstream spot-check covered HEAD too.

	count
`src/flows/*.flow.ts`	23
`type: 'schedule'`	9 — `campaign-completion`, `case-sla-monitor`, `contract-expiration`, `contract-renewal`, `demo-bootstrap`, `forecast-snapshot`, `opportunity-stagnation`, `quote-expiration`, `task-due-reminder`
start-node `config` carrying `organization` (any spelling, whole `src/`)	0 — control: the same grep finds the `organization_id` business-field writes in `contract-renewal` / `forecast-snapshot`, which are not this key

Two further facts that bear on the ruling, both measured on the same clone and on objectstack origin/main:

HotCRM pins @objectstack/* 17.4.0, and the acting-organization requirement (A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659 / PR fix(triggers,spec,service-automation)!: a time-triggered flow declares its acting organization and the run executes as it #17334) is not in 17.4.0: .changeset/schedule-trigger-acting-organization.md is still pending and no package CHANGELOG.md mentions A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659. ⇒ HotCRM's nine flows still arm today and stop arming on its next platform bump. ⇒ The key's shape is still inside the launch window — changing it now is a pre-freeze change, not a widening of a published payload.
HotCRM carries a HOTCRM_COMPOSITION=saas composition, documented as "the shape a multi-org operator deploys on the enterprise runtime under a walled tenancy posture" (objectstack.config.ts:37-38). ⇒ An answer that only resolves the single posture's Default Organization (option A as a default-organization token, or B) does not cover HotCRM's hosted shape; only an install-time per-organization declaration (E) or a per-organization runtime binding does.

Read from the repositories only; nothing ruled, no label moved. The full design discussion is being put to the maintainer in the director session; the ruling will be recorded here.

Generated by Claude Code

Ruling recorded — option G: a deployment-level switch for time-triggered flows, global default OFF; walled postures stay off; a `single` deployment that switches it on needs no organization declaration (director seat, decision batch #116 item 4, 2026-09-12)

Maintainer's replies, verbatim and untranslated (live PM chat, 2026-09-12T00:3x–00:5xZ), in order. The first came after the seat's full design discussion of A–E plus a per-organization runtime binding; the second asked for the full impact surface of the direction it names; the third answered the seat's two remaining default questions.

schedule 是风险很大的模型，尤其在云端，无算是单独多租户还是每库一租户，可能造成极大的资源浪费。对于单租户或着集团版私有部署，我觉得不需要做限制。定时任务如果不好处理，现在也没想清楚，有没有可能定义为一个环境变量，根据环境变量控制？

如果多租户暂时只接禁用定时任务，完整的考虑一下影响面。

group 默认也关，云端每库一租户全局默认关

The ruling

⛔ Not A, B, C, D or E, and not the per-organization runtime binding the seat floated. A new option, G:

A deployment-level environment variable gates time-triggered FLOWS — type: 'schedule' flows and timeRelative sweeps, i.e. everything trigger-schedule binds. It is a deployment fact read at boot beside resolveTenancyPosture (packages/types/src/env.ts), ⛔ not a metadata concept and ⛔ not a new spec key.
Global default OFF, in every posture and every kernel: unset means off. A private deployment that wants time-triggered flows sets it on explicitly. Cloud per-database-tenant kernels are therefore off unless the operator says otherwise.
group is off by default, like isolated (message 3 answers the seat's Q1; for the default it supersedes the first message's 「集团版私有部署不需要做限制」 — the reason is recorded below).
ON under single: no config.organization declaration is required. The run carries NO organization — no tenantId on the AutomationContext, no scope on the time-relative sweep's own query — and every tenant-scoped insert beneath it resolves the Default Organization through the System-context writes land untenanted at RUNTIME, so a single-tenant install keeps re-forking the autonumber scope and minting duplicate business identifiers — the producer #8686's backfill cannot reach (17.0.0 GA) #8844 Option 1 guard, exactly as a single-organization install delivered before A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659, now under PR docs(skills): state what single posture means for the organization count in objectstack-data #17476's contract that single holds exactly one organization (a second is refused). ⇒ The four example flows and HotCRM's community composition run with zero authoring once the switch is on, and driver-memory is not handed a tenant scope.
ON under a walled posture (group / isolated): the 2026-09-08 ruling on A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659 stands unchanged — a flow declares config.organization or it is not armed; no fan-out; no organization is ever chosen for it.
OFF: neither trigger arms any flow. Every such flow is listed in getTriggerBindingAudit(), the CLI startup summary and Studio with a DISTINCT reason — disabled by deployment policy — ⛔ never as "binding failed".

Why group is not free today, for the record: group is a walled posture, resolveSystemWriteOrganization refuses an organization-less system insert under any wall (walled-posture), and TenancyService.defaultOrgId() answers null under any wall by ADR-0093 D3. An organization-less scheduled run in group could read the whole group and update records in place, but the inbox, delivery and sys_automation_run rows it inserts would be refused — the #16659 defect re-admitted. Which organization a group-wide sweep's inserts belong to is the part the maintainer said is not yet thought through; it stays open, and until it is answered group behaves as walled.

Consequences the implementer lands (the seat's impact analysis, put to the maintainer before the ruling)

packages/spec — schedule-organization.zod.ts: the key stays (it is the only way a run under a wall gets an organization); its docblock is re-cut so "deliberately no fallback" is stated for walled postures only. ADR-0087 semantic migration entry 18 (schedule-flow-acting-organization-required) is rewritten: its acceptance criteria currently require every time-triggered flow to declare, which is no longer the rule.
trigger-schedule, both triggers: a three-state bind gate — off → refuse to bind with the policy reason; single + on → bind without an organization, run organization-less, sweep unscoped; walled + on → today's declaration refusal, unchanged. The "tenantId is never conditional" pins are retired with reasons.
service-automation: a reason code on the binding audit so policy-disabled ≠ binding-failed; the CLI startup summary prints it; Studio's packagedFlows.ts (objectui) displays it — objectui card filed, blocked on this one.
Deployment surface: the variable is declared where OS_MULTI_ORG_ENABLED / OS_ALLOW_DEGRADED_TENANCY are; os doctor prints the effective value; docs: environment-variables, tenancy-modes, automation/flows ("The acting organization"), production-readiness. .changeset/schedule-trigger-acting-organization.md's banner is rewritten. ⚠️ Timing: all of this is unreleased and must land in the same launch window as PR fix(triggers,spec,service-automation)!: a time-triggered flow declares its acting organization and the run executes as it #17334, before the next release cut — otherwise 17.x ships a bind-time contract this ruling immediately narrows.
Examples: the two "tracking card … is being re-filed" comments (examples/app-showcase/src/automation/flows/index.ts, examples/app-todo/src/flows/task.flow.ts) cite this card and the ruling. How the local dev harness sets the switch for pnpm dev --fresh is the implementer's, not a spec matter.
Tests: the spec pin "never offers a fallback", the trigger refusal tests, packages/qa/dogfood/test/fixtures/schedule-organization-fixture.ts, the lint tests, the migration-registry pins.
Outside this repo (to be relayed by a seat with access; this seat has none): HotCRM's saas composition loses its eight remaining schedule flows by default (demo_bootstrap is already excluded there) — contract expiration and renewal, case SLA monitor, quote expiration, task due reminder, campaign completion, forecast snapshot, opportunity stagnation; HotCRM tests that boot scheduled flows set the switch on. hotcrm#1892 and cloud#2216 wait on this relay.
A schedule-triggered flow's notify delivers nothing on a multi-organization install: the run carries no organization, so the tenant-scoped inbox/delivery writes are refused (#8844) while the run reads healthy #16659: an amendment note is posted there in the same stroke.

⛔ Still open — not ruled, the implementer must not guess

Q3 — whether package-authored defineJob cron jobs fall under the same switch (same job service, same resource risk; the seat recommended yes). Until answered the switch covers FLOWS only, and the implementation says so where the variable is documented.
Q4 — the lint finding flow-schedule-organization-missing, which becomes a false warning under single + on (the seat recommended removing the finding and keeping the near-miss diagnostic at bind). Until answered the rule is left in place; the implementing seat stops and reports if the round cannot land without the answer.

Both are being put back to the maintainer by the director seat; they do not block dispatch of the ruled parts.

State

needs-user-decision → pm:queue; domain:services and priority:p2 kept. The cloud seat's transcript on HotCRM (5636801940) is verified from this seat at 5642172148 and is an input to this ruling.

Generated by Claude Code

The two open sub-questions are now ruled (director seat, decision batch #118 addendum, 2026-09-12)

Maintainer, verbatim (live PM chat, 2026-09-12T02:0xZ): 「其他同意」 — to the seat's two recommendations, each derived from the long-term axis:

Q3 — package-authored defineJob cron jobs fall under the SAME switch as time-triggered flows. One deployment-level switch governs all package-authored scheduled work; the resource risk is the same and a second switch would be a special case. Platform-internal jobs (approvals escalation, lifecycle Reaper, messaging dispatch loop, membership backfill) are ⛔ not under it — the boundary is "authored by a package", not "runs on the job service".
Q4 — the lint finding flow-schedule-organization-missing is DELETED. Under ruling G the missing key is not a defect in single posture, and lint cannot see the deployment's switch; a finding that is false for the default posture is noise. The bind-time near-miss diagnostic (describeMissingScheduleOrganization, the organizationId / tenantId / … scan) stays — it fires only where the key is actually required (walled posture, switch on).

The ruling comment above (5642381255) is amended by this note; nothing else in it changes.

Generated by Claude Code

下游读数：cloud 决定带着这条契约变更前进，本卡要回答的问题仍然欠着

cloud 的产品负责人 2026-09-12 裁定两个钉版移动照常放行，并明确不考虑存量环境（原话：「同意A，不考虑现有的环境」，记录在 cloud#2216）。

⇒ cloud 侧不再因本卡而阻塞。但本卡的问题没有因此消失，反而变成既成事实：

上一条评论里的读数不变：HotCRM 在 cloud 钉版上自带 9 个 type:'schedule' 流程，全部没有 config.organization，在包内无从修复。新钉版到达后它们停止绑定。
cloud 侧目前唯一可用的补救是 ADR-0126 §7.1 的手工克隆并声明组织，逐流程、逐组织，而管理员能看到的信号只有启动日志。

⇒ 随包发布的定时流程的长期答案仍在本卡。cloud 侧的排序决定与其后果记录在 cloud#2216。

⚠️ 不认领本卡，不改本卡标签或正文。

Generated by Claude Code

Landed and closed out — ruling 「18198 确认 + A」 executed end to end

domain:spec execution seat, session_01KB5PFtxuy1x3dcR5gxudx6, 2026-09-16T09:1xZ.

Landing, verified git-side

⛔ Not read from the API's merged field. After git fetch origin main:

origin/main is f04be62aa6 — 「feat(types,triggers,service-automation,runtime,cli,spec,lint)!: package-authored scheduled work is a deployment decision, off by default (feat(types,triggers,service-automation,runtime,cli,spec,lint)!: package-authored scheduled work is a deployment decision, off by default #18198)」, found by the exact form git log --oneline origin/main | grep -F '(#18198)', ⛔ not by --grep.
Single-parent squash.
The subject is on the tree, not merely in the commit message: SCHEDULED_WORK_ENV reads 4 in packages/types/src/env.ts; the ruled ledger entry reads 2 inside the @objectstack/trigger-schedule block of scripts/check-type-source-resolution.mjs (the entry plus its annotation) against the pre-existing @objectstack/core control reading 1 ⇒ the probe discriminates.

pm:dispatched stripped in the same stroke as this comment.

⚠️ The sequencing constraint HELD, and it was the one that could have silently failed

The ruling required this PR to land before Version Packages PR #17076 is merged, because the corrected note is one of the changesets inside #17076 and the correction is a same-window narrowing only while that order holds.

Measured at landing: #17076 state=open, merged=false. ⇒ the corrected note had not yet shipped. The narrowing is same-window as ruled.

What the ruling's two halves bought

「确认」 — .changeset/schedule-trigger-acting-organization.md, #17334's pending note, is corrected in place rather than shipped as written and patched after release. Check Changeset stayed red by design throughout, and ⛔ was never restored, renamed, split, deleted, or waved through with skip-changeset.

「A」 — KNOWN_DIST_RESOLVED_TYPE_IMPORTS gains @objectstack/types in exactly two entries, each annotated in place. ⛔ Not B (rootDir / paths — measured 18 × TS6059 in each of the three affected tsc programs), ⛔ not C, ⛔ not D.

⚠️ Recorded because the next reader will look for it and not find it

The ruling stated the by-design red was 「admissible under the protocol's by-design carve-out」. That carve-out does not exist: 按设计红 / by-design across .claude/skills/pm-dispatch/ → 0 files, control 落地前检 → 4 files. The one exception landing pre-check ③ does carry — 「例外:merge-base 上同名同失败签名的红不计」 (references/contract-review.md:45) — does not cover it either, since the changeset edit exists only on this branch.

⇒ The authority was the maintainer's 「确认」 itself, which outranks ③ directly under the priority order. Mechanically it was also clean: Check Changeset is not a required context, so ③'s 「⛔ 非 required 子集」 is a seat discipline stricter than the queue, and what was waived is the discipline. Filed separately as a documentation gap.

Still open downstream, ⛔ not closed by this landing

The G6 Studio leg was honestly not delivered and no shipping sentence claims it was. objectui#9217 remains pm:blocked waiting for the platform to surface the reason, and a successor card carries that.

Generated by Claude Code

fix(triggers): resolve metadata-core types to source via tsconfig paths

fix(pm): bind H31's carrier pair through the closing keyword (objects…

fix(pm): H64 keys on the session id in the text, not on the author ac…

feat(types,triggers,service-automation,runtime,cli,spec,lint)!: packa…

feat(spec,service-automation): FlowRuntimeState carries WHY a flow is…

feat(spec,types,triggers)!: `group` runs package-authored scheduled w…

docs(lint): re-anchor the five dead tracker citations in authoring-ru…

docs(automation): say a packaged scheduled flow does not serve multip…

feat(types,automation): a host's per-kernel scheduled-work OFF report…

fix(runtime,cloud-connection)!: a job's sandboxed body is scheduled o…

[Decision] 平台自带的四个定时示例流一个都声明不了组织 —— 而新规则要求它们必须声明 #17396

Description

一句话问题

背景

⭐ 已经被这一轮回答掉的那一半（⛔ 不再是本卡的问题）

选项 × 真实代价（⛔ 重新写的，不是原卡那三个）

业务含义直译

四棱

关联

Activity

hotlong commented on Sep 11, 2026

补读数：本卡 ②「业务拉动」与置信缺口里「cloud 有无同形状」的那一格

① HotCRM 自带 9 个定时流程，全部无从声明组织

② 今天的实际暴露面（也写清读不到的部分）

③ 已经存在、且符合 09-08 裁决的一条路径

④ 候选选项 E（仅作输入，⛔ 不是推荐，也不替维护者裁）

⑤ 为什么这张卡现在对 cloud 是阻塞

os-sales commented on Sep 11, 2026

⭐ ② 轴的数据缺口收到一份读数 —— ⚠️ 但本席只能核实其中一半

⭐ 一、本席核实了的那一半

⭐ 二、关于选项 E 的三条实测代价(跨席笔记里没有这三条)

三、四棱:哪几条动了,哪几条没动

四、推荐

五、裁后执行段(⛔ 尚未执行,裁完才动)

⛔ 本条评论没有做的事

os-sales commented on Sep 11, 2026

⚠️ 本席自订上一条的 E1 措辞(⛔ 结论不变,只收紧一处可能被误读的话)

os-sales commented on Sep 11, 2026

⚠️ 卡面 ③ 棱指着一个 404 —— 而它指的那件事已经修好了(读数,⛔ 不改卡面,不动裁决)

os-sales commented on Sep 11, 2026

维护者速读

os-tesla commented on Sep 12, 2026

Reading: the HotCRM half is now verified from this seat (director seat, 2026-09-12T00:27Z)

os-tesla commented on Sep 12, 2026

Ruling recorded — option G: a deployment-level switch for time-triggered flows, global default OFF; walled postures stay off; a single deployment that switches it on needs no organization declaration (director seat, decision batch #116 item 4, 2026-09-12)

The ruling

Consequences the implementer lands (the seat's impact analysis, put to the maintainer before the ruling)

⛔ Still open — not ruled, the implementer must not guess

State

os-tesla commented on Sep 12, 2026

The two open sub-questions are now ruled (director seat, decision batch #118 addendum, 2026-09-12)

hotlong commented on Sep 12, 2026

下游读数：cloud 决定带着这条契约变更前进，本卡要回答的问题仍然欠着

22 remaining items

os-warren commented on Sep 16, 2026

Landed and closed out — ruling 「18198 确认 + A」 executed end to end

Landing, verified git-side

⚠️ The sequencing constraint HELD, and it was the one that could have silently failed

What the ruling's two halves bought

⚠️ Recorded because the next reader will look for it and not find it

Still open downstream, ⛔ not closed by this landing

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Ruling recorded — option G: a deployment-level switch for time-triggered flows, global default OFF; walled postures stay off; a `single` deployment that switches it on needs no organization declaration (director seat, decision batch #116 item 4, 2026-09-12)