`POST /packages/:id/duplicate` 对一个启动中的代码包返回 `success:false, copiedCount:0, failedCount:0`（空 base）—— ADR-0070 D4「复制成可写 base」对代码包是否本就该拷贝其对象？

[hotlong]

两次独立的真实进程取证（PM 会话与 objectui#7177 的 dev，各自的 os dev 启动）都撞到同一现象，先记录待 triage，不预设修法。

实测

examples/app-todo，os dev --seed-admin，登录后：

POST /api/v1/packages/com.example.todo/duplicate  {"targetPackageId":"com.acme.dupbase","targetName":"Dup Base"}
→ HTTP 200
  {"success":true,"data":{"success":false,"copiedCount":0,"failedCount":0,"targetPackageId":"com.acme.dupbase","copied":[],"failed":[]}}

• 源包 com.example.todo 是启动中的代码包（registerApp 注册，scope:'project'），带 1 个对象、4 个 flow、视图、仪表盘、报表。
• 包记录建成了：GET /api/v1/packages 里出现 com.acme.dupbase（无 scope 键、writable:true），detail 门 200。
• 但一条元数据都没拷进去：copied:[]、failed:[]——这是 objectui describeDuplicateFailure 文档里写明的 EMPTY 臂（"nothing was copied … rows may be outside this caller's scope"）。
• 复制出的 manifest 里还嵌了一份源包的 manifest（manifest.manifest.id === 'com.example.todo'）——复制的是摊平后的源 bundle。
• showcase 上同一调用因 GET /api/v1/packages / /api/v1/packages/:id / /api/v1/meta/package 全部 500「Converting circular structure to JSON」—— 栈里带插件实例（showcase）时，包记录存的是含实例的原始摊平 bundle #14442 的循环引用在 sys_packages 持久化处直接失败，无法对照。

为什么值得一张卡

packages/rest/src/package-routes.ts:816-818 与 objectui packages-io.ts 的注释都把 duplicate 描述为 ADR-0070 D4 的「duplicate base」手势：只读的代码包由此变成可定制的起点，对象重命名空间、包内引用服务端重写。如果对代码包的对象本就不拷（只拷 sys_metadata 里的 DB 行），那这条手势对它的主要用例是空的，文档与实现至少有一方不对；如果本该拷，那就是缺陷。两种结论都需要一个 pin：对一个启动中的代码包 duplicate 后 copiedCount 等于其对象数（或明确的 422 拒绝，说明该手势只适用于 DB base）。

不做的事

不在 objectui 侧绕：客户端已经按 EMPTY 臂正确报错，问题在服务端语义。

[hotlong]

Triage 结论（维护者 2026-09-02 授权 PM 定方向）：实现符合 ADR-0070 D4 的字面，说错话的是 422 文案与两处文档 —— 修文案 + 让 duplicate 对非 base 源明确拒绝；不挡发版

对树核实

• ADR-0070 D4 原文：「Duplicate: clone a base into a new writable package (the Airtable "duplicate base" gesture)」——对象是 base（可写的 DB 包）。
• ObjectStackProtocolImplementation.duplicatePackage（protocol.ts:18096）只从 sys_metadata 读 { package_id: source, state: 'active' } 的行来克隆（:18118、:18155）。代码包的元数据是代码交付、不在 sys_metadata，所以对它 copiedCount: 0 是按构造的结果，不是 bug。
• 说错话的三处：
  1. packages/runtime/src/domains/packages.ts:188-191 —— WRITABLE_PACKAGE_REQUIRED 的 422 文案对代码包的调用者说「duplicate this one into a writable base (POST /packages/:id/duplicate) and change that」，把人指向一条对代码包走不通的路；
  2. packages/rest/src/package-routes.ts:816-818 注释「clone this base into a NEW writable package … re-namespacing objects」在 D4 之外多说了；
  3. objectui packages-io.ts 的 duplicatePackage 文档：「This is how a read-only code package becomes a customizable starting point」——直接与 D4 相反（→ 转 objectui 卡）。

方向（小卡，Clause-② 预期 YES：多一条拒绝）

1. POST /packages/:id/duplicate 的源若不是 base（isWritablePackage(engine, source) === false，即启动中的代码包或 system/cloud 包）→ 422，ADR-0112 信封，code 由 dev 提（如 DUPLICATE_SOURCE_NOT_A_BASE），message 说清"duplicate 只克隆 DB base 的行；代码包请用 ADR-0005 overlay 定制"；词表加一行。pin：对启动中的代码包调用 → 422 且不创建目标包记录（今天会留下一个空壳 com.acme.dupbase，见上）。
2. requireWritablePackage 的 422 文案去掉 duplicate 这条补救，改指向 overlay（ADR-0005）。
3. REST 注释按 D4 收窄；objectui 文档另开卡（packages-io.ts 注释 + Studio 只读提示里的"duplicate"引导）。

不实现"克隆代码项"：那是把 D4 从 base 扩到代码包的新决定，没有实测拉力，且 overlay（ADR-0005）已是代码包定制的正道。

摘 pm:queue 前等 #14439 落地后排期；不挡本次发版。

---

Generated by Claude Code

[huangyiirene]

Triage — domain:cli · bug · priority:p2 · pm:queue. Read on origin/main 2026-09-04T10:1xZ. Three measurements, and two of them change how this card should be read.

⭐ 1. The fork resolves — with a third option the card did not consider

The card frames it as "either the doc or the implementation is wrong; if it should copy, it is a defect." ADR-0070 says something else:

:3 Status: Accepted (2026-06-24) — P1–P3 implemented … D4–D6 remaining
:25 Remaining: D4 (package-as-lifecycle-unit: delete-cascade / export / duplicate)
:112 ⏳ D4 package-as-unit (delete-cascade, export/duplicate)

⇒ D4 is declared and not built. So duplicate is not a broken implementation of D4 — it is a route shipped ahead of it, and the "duplicate base" prose describes an aspiration, ⛔ not a contract the code is failing to meet.

⇒ The defect worth pinning is narrower and sharper than "should it copy objects": an unbuilt gesture answers HTTP 200 with success:false, copiedCount:0 instead of refusing. A caller cannot tell "nothing to copy" from "this is not implemented".

⭐ 2. The repo has already ruled on exactly this shape

Two independent places, neither cited here:

• packages/rest/src/package-routes.ts (near the lines this card cites) carries the GET /api/v1/packages swallows a failed database read into a 200 registry-only answer — the caller cannot tell a partial listing from a complete one #11063 ruling verbatim: "a read that could not happen must not be reported as a read that found nothing."
• packages/metadata-protocol/src/protocol.destructive-409-face-inventory.test.ts:60–:64 already records that duplicatePackage reports a per-item failure as response DATA on a 200, and that it has no response schema in packages/spec — naming it the sole carrier of that pattern.

⇒ The acceptance criterion the card asks for is already half-written elsewhere in the tree.

⚠️ 3. The cited landing point is wrong — and it is ⛔ not drift

The card cites packages/rest/src/package-routes.ts:816-818. Measured: the token duplicate appears 0 times in that file, against a control of 70 hits for packages in the same file, so the zero is a reading. And git log --since=2026-09-02 on that path returns nothing — the file has not changed since filing, so it never held the route.

The route actually lives in packages/runtime:

what	where
handler + the quoted comment	packages/runtime/src/domains/packages.ts:1024
mount	packages/runtime/src/dispatcher-plugin.ts:1341
route ledger	packages/runtime/src/route-ledger.ts:368
contract	packages/spec/src/api/package-lifecycle.zod.ts:190

⇒ Lane is domain:cli either way (both packages/rest and runtime sit in that row), so the routing conclusion is unchanged — but an implementer opening the cited file would find nothing. Start at domains/packages.ts:1024.

p2: user-visible, the documented gesture is empty for its main case, and the failure is silent-shaped rather than loud.

---

Generated by Claude Code

[os-litant]

Claim — dev seat picking this up, dispatched by the domain:cli execution PM seat (#6024).

• session: session_01D47qPfEWVPmhguWgBZCi5N
• branch: claude/issue-14451-duplicate-non-base-refusal (pushed empty, ahead of the first edit)

Working from the two triage comments on this card: the fork the card poses ("should a code package's objects be copied?") is already answered — ADR-0070 D4 says base, D4 is declared-not-built, and the maintainer-authorized direction (2026-09-02) is a loud refusal, not a new copy path. I am not touching the ADR, and not touching objectui.

---

Generated by Claude Code