[finding] Four `@objectstack/` packages this repo does not build are named in published docs with no declared roster — including one the CLI tells users to `pnpm add`

[os-zhuang]

Found while closing the member-existence blind spot in check:published-readme-exports (#10893). Filed unassigned, no pm:queue — triage's call. This is #10893's second shape, which that card said was not decidable offline and deliberately left open.

The observation

The #10893 sweep measured every @objectstack/ token in the 60 published documents against the workspace member map. After the repairs, four names remain that this repo does not build, and all four are legitimate as written — they are prose, not runnable imports:

name	named in	what the tree says it is
@objectstack/security-enterprise	packages/plugins/plugin-audit/README.md:365	the enterprise hierarchy resolver. packages/cli prints pnpm add @objectstack/security-enterprise as a capability hint, and packages/cli/test/capability-preflight.test.ts:74 pins that message
@objectstack/service-tenant	(removed from READMEs by #10893, still in source)	the cloud control-plane runtime — packages/spec/src/system/constants/platform-object-names.ts:132-134: "Objects contributed by the CLOUD runtime (@objectstack/service-tenant, defined in the separate cloud repository)"
@objectstack/framework	packages/console/README.md:3, 13, 25	the umbrella install name — pnpm add @objectstack/framework, also in packages/console/package.json's description and packages/cli/src/utils/console.ts:17
@objectstack/organizations	packages/plugins/plugin-security/README.md (added by #10893)	the enterprise multi-org runtime — packages/cli/src/commands/serve.ts:2423 tells the operator to add it to their app

Two more are historical names, correct because the sentence is about the past: @objectstack/service-external-datasource and @objectstack/service-datasource-admin, which packages/services/service-datasource/README.md:11-12 names as the packages it used to be.

Why it is worth a card

Nothing declares any of this. A reader cannot tell an out-of-repo package from a fabricated one, and neither can a gate — which is exactly why #10893's new assertion is fenced to import statements only: measured on 409077e93c, zero published documents import an out-of-repo @objectstack/ package from a fence, so the fence costs nothing today and an allowlist would have been a population of zero. That fence holds only while the count stays zero.

The sharper half is @objectstack/framework. It is not a passing mention — it is an install command in a published README, and it names a package that is in no workspace directory of this repo. Either it is published from somewhere this repo does not record, or the canonical install instruction for the framework is wrong. Nothing here can tell which, and that is the finding.

The question that has to be answered before anything is built

May a published README name an @objectstack/ package this repo does not build, and if so, where is that declared? Three shapes, and picking one is a ruling rather than a scan:

• a declared roster (scripts/*.json, one row per out-of-repo package with where it ships) — makes the gate able to admit a fenced example for an enterprise package, at the cost of a file that must be kept true;
• prose-only, enforced — the status quo after [finding] Three published READMEs tell readers to install packages that are in no directory of this repo — and check:published-readme-exports is blind to it by construction #10893, which needs no roster but permanently forbids a runnable example for a package that really does ship elsewhere;
• a registry probe — decides it from npm rather than from a declaration, and takes this family of gates online, which every other assertion in it deliberately is not.

⛔ Not a rider on #10893: that card's PR closes the offline-decidable half and says in its body that this half wants its own card.

Refs: #10893 / its PR (the offline half, and the measured fence) · #10920 (knowledge-turso, the same question on a runtime surface) · #9532 · #10368

[huangyiirene]

First-touch grading (triage seat, session session_013ELWDeai2Y6Wqrh7DmkUV6): promoted out of finding into the decision inbox. domain:devx kept — the roster/gate half lands in scripts/. This card is made the parent for the shared prior question; #10920 (the runtime-surface instance) is being marked pm:blocked against it rather than entering the inbox separately, per the same-family meta-criterion.

Why it is not queueable as filed: the body's three shapes each widen what a gate accepts (a roster admits fenced examples the current fence forbids; a probe decides admission from the network). That is the mechanical boundary test for the manual floor — accepted-set expansion — so no seat rules on it.

[facets-block] 四棱分析(业务视角)

一句话问题:我们公开的文档和产品运行时提示里,可以出现「这个仓库不构建的 @objectstack/ 包」吗?可以的话,它从哪发布这件事记在哪?

客户可感知的代价:今天 packages/console/README.md 的第一条安装指令是 pnpm add @objectstack/framework,而这个包不在本仓任何目录里。它要么真的在别处发布(那我们只是没记录),要么我们的头号安装指令是坏的。读者分不出来,gate 也分不出来。同类还有三个:CLI 会打印 pnpm add @objectstack/security-enterprise(且有测试 pin 住这句话)、serve.ts 让运维加 @objectstack/organizations、spec 常量注释指向 cloud 仓的 @objectstack/service-tenant。

选项	客户端效果	我们付的代价
A 建声明式名册(一行一个仓外包 + 它从哪发布)	gate 能放行真实仓外包的可运行示例,读者能查到出处	多一个必须保持真实的文件
B 只允许散文提及、机械强制(#10893 后的现状)	永远不能为一个真实在别处发布的包写可运行示例	零成本,但把一条合法用例永久关死
C 联网查 registry 判定	自动、永远准	把这族刻意离线的 gate 拉上网(限流、离线 CI 失效)

行业类比:像文档里写「请安装我们的企业版插件」,但公司内部没有任何一张表说明企业版插件由哪个团队发布、发布到哪 —— 客服和文档作者只能互相打听。

① 实际业务需求:这四个名字里至少一个(@objectstack/framework)是安装指令,不是顺带提及,而且被 CLI 的一条测试 pin 住。⚠️ 但这一半严格说不是决策题,是没人查过的事实题 —— 见下面的推荐。
② 项目长远合理性:A 把「仓外包」从口头知识变成声明,与 contract-first 同向;B 永久禁止一类合法用例;C 破坏这族 gate 一致的离线设计,为一个问题引入一整类新故障。
③ 防 AI 写错:今天读者和 gate 都区分不了「仓外包」与「凭空捏造的包」—— AI 会照抄 README 里的 pnpm add,这正是批量犯错被掩盖的温床。A 的「声明即强制」最对症;B 至少保住 gate 的响亮拒绝;C 把判断权交给网络。
④ 创业阶段不扩散:B 零成本;A 一个文件加一份保持真实的义务;C 最贵。

推荐:先答事实,再选形状。 四个包到底发布没发布、发布在哪,这个读数一旦有了,选项就塌缩了 —— 如果它们真在别处发布,A 几乎是唯一诚实的选择;如果一个都没有,那这些根本不是治理问题,是四处文档/运行时错误,按 B 修文本即可,连 ruling 都不需要。⛔ 不推荐 C。

本分析看不见什么:① 本席无法从仓内查 npm registry,四个包的发布状态是零读数,而它决定 A 与 B 谁对;② cloud 仓不在本会话的 GitHub 授权面,@objectstack/service-tenant 的归属无法自查;③ @objectstack/framework 是否是伞包(umbrella)发布名,本仓无任何记录。

裁决格式(低摩擦):你如果知道这四个包发布在哪,回一句话就够了 —— 我据此直接路由(有出处 ⇒ A 建名册;没有 ⇒ B 修四处文本)。你如果也不确定,回「查」,我路由一张 S 级测量卡(查这四个名字的发布状态与出处)进 domain:devx 队列,本卡挂 Blocked-by: 等它。也可以直接回 A / B / C 覆盖以上。

裁后执行段:本卡裁定落地时,同笔解除 #10920 的 pm:blocked(它等的是同一个前置事实,不会自己解锁 —— 本卡即使转成工作态仍是 open,解锁扫描看不见)。

---

Generated by Claude Code

[huangyiirene]

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, session_01ApyDuQY2fkunMCqXiqvBhR)

The full decision inbox (46 cards) was presented with per-card four-axis recommendations; the maintainer accepted the batch, verbatim: 「接受所有」.

Ruled: Measure first — the triage recommendation ("先答事实,再选形状") is adopted as-is. This card's first step is the measurement described in the recommendation: establish the publication status and provenance of the four out-of-repo names (@objectstack/framework, @objectstack/security-enterprise, @objectstack/organizations, @objectstack/service-tenant). The A/B fork is decided by the measured result, not now: packages that really are published elsewhere collapse the choice toward A (a declared roster with provenance); if none are published, this is four documentation/runtime text errors to be fixed under B's shape, with no governance ruling needed. Option C (a registry probe at gate time) is not taken. The post-ruling execution note stands: when this card's fork resolves, the same change unblocks #10920, which waits on the same prior fact.

State: needs-user-decision removed → pm:queue (ready for dispatch).

---

Generated by Claude Code