docs(skills,docs): 三处 hook 文档不再教「批量写的行级谓词在 ctx.input.ast」(#5670)

引擎从未在写路径的 `HookContext` 上放过 AST。`packages/objectql/src/engine.ts`
的 5 个 `HookContext` 生产点里,`input: { ast }` 只出现在两条读路径(喂
`driver.find` / `findOne`);写路径构造的是 `{ id, data, options }` 与
`{ id, options }`,批量写的行级谓词走引擎内部的 `OperationContext.ast`(#2982),
从不进 `hookContext.input`。#5273 已把 spec 侧那份契约表改成引擎真正构造的形状
(PR #5668),真值由新增的 `packages/objectql/src/hook-input-shape-contract.test.ts`
钉死 —— 其中 `beforeUpdate` / `beforeDelete` 在 `multi: true` 上都断言
`'ast' in input === false`,并以 `beforeFind` 作阳性对照证明引擎并非到处不发 `ast`。

同一句从未兑现的陈述还活在三个面上,本单按已落地措辞逐处对齐:

- `skills/objectstack-data/rules/hooks.md`
- `skills/objectstack-data/references/data-hooks.md`
- `content/docs/api/data-flow.mdx`

三处改法一致:删掉「the row-scoping predicate is in `ctx.input.ast`」半句,替换为
两句实情 —— 批量写不向 hook 暴露谓词(谓词在引擎内部 `OperationContext.ast`,
composed 的 RLS / sharing 过滤器由此绑定 driver 调用本身,handler 无法放宽;要
限定批量就在调用方用 `options.where`),以及 #5038 之后 `after*` 按匹配行派发、
每行单记录形状、`input.id` 在那里绑定。主句(批量写触发同名事件、没有 `*Many`
事件)本身成立,原样保留。

前两个文件属已发布 skill 目录,沿其既有惯例带 issue 编号;`data-flow.mdx` 全文
零 issue 引用,故那处不引编号。三处都不写仓内 `packages/**` 测试路径 —— 这三份
文件现有引用里没有一条指向仓内源码路径,而 skill 经 `npx skills add` 装到第三方
项目后那种路径也不存在。

⛔ `content/docs/releases/v16.mdx:185` 带同句但未动:发布态记录按 CLAUDE.md 不在
代码 PR 里改。`packages/spec/**` 未动(#5668 已修)。

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GX3sL71LFq8m2usg6VqTSE

Author: claude

content/docs/api/data-flow.mdx

@@ -310,7 +310,7 @@ flowchart TD
     end
 ```
 
-The lifecycle events are defined by the `HookEvent` enum (8 events). Reads fire `beforeFind`/`afterFind` — for **both** `find` and `findOne`, so one subscription covers every read shape. Bulk writes (`multi: true`) fire the **same** `beforeUpdate`/`beforeDelete`/`afterUpdate`/`afterDelete` events as single-id writes, with the row-scoping predicate in `ctx.input.ast`. There are deliberately no per-method (`findOne`/`count`/`aggregate`) or `*Many` events: read authorization and row filtering are RLS/permission-rule concerns, and field masking is field-level metadata.
+The lifecycle events are defined by the `HookEvent` enum (8 events). Reads fire `beforeFind`/`afterFind` — for **both** `find` and `findOne`, so one subscription covers every read shape. Bulk writes (`multi: true`) fire the **same** `beforeUpdate`/`beforeDelete`/`afterUpdate`/`afterDelete` events as single-id writes. A bulk write hands hooks **no** row-scoping predicate: it lives on the engine-internal `OperationContext.ast`, so the RLS / sharing filters composed onto it bind the driver call itself, where no handler can widen them — scope a batch through `options.where` at the caller. The `after*` events instead dispatch **once per matched row**, each on a single-record-shaped context whose `input.id` names that row. There are deliberately no per-method (`findOne`/`count`/`aggregate`) or `*Many` events: read authorization and row filtering are RLS/permission-rule concerns, and field masking is field-level metadata.
 
 | Hook | Phase | Can Modify? | Can Abort? |
 |:---|:---|:---|:---|

skills/objectstack-data/references/data-hooks.md

@@ -58,11 +58,16 @@ ObjectStack provides **8 lifecycle events** organized by operation type:
 > **Why only 8?** The read events fire for `findOne` as well as `find` (the event
 > attaches to record materialization, not the engine method), so one subscription
 > covers every read shape — there is no `beforeFindOne`/`afterFindOne`. Likewise the
-> write events fire on bulk `multi:true` operations (the row-scoping predicate is in
-> `ctx.input.ast`), so there is no `*Many` event. And there is no `beforeCount`/
-> `beforeAggregate`: read authorization and row filtering belong to **RLS / permission
-> rules**, and field masking to **field-level metadata** — declarative mechanisms that
-> apply everywhere, rather than a hook every author must remember to re-attach.
+> write events fire on bulk `multi:true` operations, so there is no `*Many` event. A
+> bulk write hands hooks **no** row-scoping predicate: it lives on the engine-internal
+> `OperationContext.ast` (#2982), so the RLS / sharing filters composed onto it bind
+> the driver call itself, where no handler can widen them — scope a batch through
+> `options.where` at the caller. The `after*` events instead dispatch **once per
+> matched row**, each on a single-record-shaped context whose `input.id` names that
+> row (#5038). And there is no `beforeCount`/`beforeAggregate`: read authorization and
+> row filtering belong to **RLS / permission rules**, and field masking to
+> **field-level metadata** — declarative mechanisms that apply everywhere, rather than
+> a hook every author must remember to re-attach.
 
 ### Before vs After Hooks
 

skills/objectstack-data/rules/hooks.md

@@ -112,9 +112,14 @@ Sandbox essentials (full contract in
 
 > **One read event, one write event per kind.** `beforeFind`/`afterFind` fire for
 > `findOne` too (the event attaches to record materialization, not the method), and
-> the write events fire on bulk `multi:true` operations as well — the row-scoping
-> predicate is in `ctx.input.ast`. There is no `beforeFindOne`, `beforeCount`,
-> `beforeAggregate`, or `*Many` event.
+> the write events fire on bulk `multi:true` operations as well. A bulk write hands
+> hooks **no** row-scoping predicate: it lives on the engine-internal
+> `OperationContext.ast` (#2982), so the RLS / sharing filters composed onto it bind
+> the driver call itself, where no handler can widen them — scope a batch through
+> `options.where` at the caller. The `after*` events instead dispatch **once per
+> matched row**, each on a single-record-shaped context whose `input.id` names that
+> row (#5038). There is no `beforeFindOne`, `beforeCount`, `beforeAggregate`, or
+> `*Many` event.
 >
 > **Don't reach for a hook when a declarative mechanism already fits:**
 > - Read authorization / row filtering → **RLS / permission rules**, not a `beforeFind` hook.