Skip to content

[spec] HookContext.api 声明为 z.unknown() —— 按 (ctx: HookContext) 标类型的 hook 无法调用 ctx.api.object(…),而文档/技能全在这么教 #5945

New issue
New issue
Closed
Closed
[spec] HookContext.api 声明为 z.unknown() —— 按 (ctx: HookContext) 标类型的 hook 无法调用 ctx.api.object(…),而文档/技能全在这么教#5945
Labels
domain:specDomain lane — seat card indexed by label:pm:seatpm:queueReady for the PM dispatch looptarget:v17Release blocker for v17 — on the board until fixed, dropped as no longer valid, or accepted for GA

Description

@os-zhuang
os-zhuang
opened on Aug 6, 2026

事实(对 origin/main 核实)

packages/spec/src/data/hook.zod.ts 里,hook 的主数据通道是这样声明的:

  /**
   * Cross-Object API
   * …
   * Usage in hooks:
   *   const users = ctx.api.object('user');
   *   const admin = await users.findOne({ where: { role: 'admin' } });
   */
  api: z.unknown().optional().describe('Cross-object data access (ScopedContext)'),

于是 HookContext['api'] 推断为 unknown,JSDoc 里那两行示例本身编译不过。实测(用 check:skill-examples 同一套 tsconfig,对已构建的 @objectstack/spec 声明编译):

error TS18046: 'ctx.api' is of type 'unknown'.      // ctx.api.object('x')

为什么是缺陷而不是风格问题

  1. 整个语料库都在教这个写法。 skills/objectstack-data/references/data-hooks.md:732-745(handler: async (ctx: HookContext) => { const users = ctx.api?.object('user'); … })、content/docs/automation/hooks.mdx:173-224、content/docs/api/error-handling-server.mdx:289/382、content/docs/automation/hook-bodies.mdx 的能力表 …… 都是 (ctx: HookContext) + ctx.api.object(…)。这些块都没有 os:check 标记,所以从没被编译过 —— 一旦标记就会全红。也就是说:AI 与人照抄的首选写法,在类型层面是不成立的,只是没有任何门看见。
  2. 它把「消费者兜底」写成了唯一出路。 [docs] runtime-services 的 hook 示例在教 ctx.services —— hook 上下文从来没有这个键,照抄的 hook 会拒掉每一次写入 #5720 的 PR docs(kernel): runtime-services 的 hook 示例改教真实通道 —— 去掉不存在的 ctx.services (#5720) #5938 需要一个真编译的 hook 示例,只能在示例里写 const api = ctx.api as CrossObjectApi(自建局部结构类型)。这正是 contract-first 反对的方向:每个消费者各自 cast 一遍,cast 的形状彼此不一致,而真值(ObjectQL 的 ScopedContext:object(name) → repo,transaction(cb))在契约里一个字都没有。
  3. 对照组说明这不是「Zod 表达不了就只能 unknown」。 同仓 packages/spec/src/contracts/ 已经用普通 TS interface 声明了一堆运行时服务面(IDataEngine、ISharingService、IObjectQLEngine …),transaction / ql 这些同样"引擎侧对象"的键也是 z.unknown(),但它们不是文档教读者天天调的东西 —— api 是。

待定(不预设结论,交维护者/分诊)

  • A. 在 packages/spec/src/contracts/ 声明作用域跨对象 API 的接口(如 IScopedContext:object(name) 返回一个 repo 接口,transaction(cb)),HookContext.api 的 TS 类型指向它(Zod 侧可继续 z.custom/z.unknown,运行时不校验活对象),让 ctx.api.object('x').findOne(…) 直接编译过。成本:要把 repo 的方法面(find/findOne/count/aggregate/insert/update/upsert/delete)定成公共契约,并与 ScopedContext 实现对齐(此后二者漂移会被 check:api-surface 之类的门看见 —— 这正是收益)。
  • B. 什么都不改,把语料库改成显式窄化(每处示例自带 cast)。成本:教 AI 写 cast,且 cast 形状无人校验;api 的真实面继续只存在于 objectql 的实现里。
  • C. 折中:只声明 object(name) 与 transaction(cb) 两个成员(文档实际教的全集),其余留待有消费者时再加 —— 与仓内「有证据才声明」的惯例一致(IDataEngine 的 #4251 注释就是这个论证)。

我倾向 C:它让文档现在教的写法真的成立(收益立刻兑现),声明面又不超过已被消费的部分;A 的完整 repo 契约可以在有第二个消费者时按同一论证增补。但这是公共契约拍板,故不自行实现。

关联

未加标签、未认领,交 PM 分诊定级。

Activity

  1. claude
    added
    domain:specDomain lane — seat card indexed by label:pm:seat
    needs-user-decisionBlocked on a maintainer decision — do not dispatch; decision cards only, never on a PR
    on Aug 6, 2026

  2. claude commented on Aug 6, 2026

    @claude
    claudeBot
    on Aug 6, 2026 – with Claude
    Contributor

    分诊:needs-user-decision + domain:spec(路由标签,方向未定)。

    本评论来自分诊座位 Routine(#5474 试点),不构成认领。

    Generated by Claude Code

  3. claude
    added
    target:v17Release blocker for v17 — on the board until fixed, dropped as no longer valid, or accepted for GA
    on Aug 6, 2026

  4. claude commented on Aug 7, 2026

    @claude
    claudeBot
    on Aug 7, 2026 – with Claude
    Contributor

    Ruling (maintainer, 2026-08-07): option C — declare the minimum IScopedContext that the corpus actually teaches.

    HookContext.api stops being z.unknown() and gains an interface covering object(name) (returning the repo face the docs actually use — at minimum findOne / find) and transaction(cb). Everything else waits for a second consumer, which is the same "declare what has evidence" discipline IDataEngine was argued into being under (#4251).

    Option B (teach explicit casts everywhere) is the wrong direction outright: it teaches AI authors to cast and leaves the shape unchecked by anyone. Option A (the full eight-method contract) can grow later under the same rule.

    This is a v17 first-hour-experience blocker: a new user copying the documented first hook currently gets TS18046, and the JSDoc example in the spec file itself does not compile. Landing C also makes #5943's gate able to bring those corpus blocks under os:check.

    Implementation note: align the declared face with objectql's actual ScopedContext and re-run check:api-surface.

    Operator: PM session session_01GcjbQLUQKysMU9uXB34iyv; maintainer ruling 2026-08-07 (decision-inbox round 2). Veto window open — comment or reopen to overturn.

    Generated by Claude Code

  5. claude
    added
    pm:queueReady for the PM dispatch loop
    and removed
    needs-user-decisionBlocked on a maintainer decision — do not dispatch; decision cards only, never on a PR
    on Aug 7, 2026
  6. os-zhuang
    added a commit that references this issue on Aug 20, 2026

    fix(tooling): derive check-engine-double-contract's own corpus figure…

    62a4eb6
  7. akarma-synetal
    added a commit that references this issue on Aug 23, 2026

    fix(tooling): read a defaulted mock initializer, name the inheritance…

    57e0059
  8. akarma-synetal
    added a commit that references this issue on Sep 17, 2026

    fix(engine): `ObjectRepository` declares the `findOne` / `update` sha…

    5c8f5af
  9. os-steve
    added 3 commits that reference this issue on Sep 18, 2026

    fix(spec): make the hook face's structurally-referenced types nameabl…

    0532dae

    fix(spec): take the two names the finding owes, and decline the third…

    3d2de76

    docs(spec): correct the declined-leg reason — the churn claim was cac…

    6d8fc4f
  10. akarma-synetal
    added a commit that references this issue on Sep 28, 2026

    feat(spec): publish the typed hook `ctx.api` face — HookApi / HookObj…

    9a910c4
  11. akarma-synetal
    added 2 commits that reference this issue on Oct 7, 2026

    test(spec): contracts and conversions test titles state each cited de…

    8256a4b

    test(spec): the fourth data/ file group's test titles state each cite…

    124533b
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    domain:specDomain lane — seat card indexed by label:pm:seatpm:queueReady for the PM dispatch looptarget:v17Release blocker for v17 — on the board until fixed, dropped as no longer valid, or accepted for GA

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions