docs(pm-dispatch): 过滤/谓词语义裁决派发必须枚举编译面清单并逐面申报 (#6410)

.claude/skills/pm-dispatch/SKILL.md

@@ -1383,6 +1383,56 @@ prompt:
 
 一句话:**一个在缺信封的实现上无法转红的拒收用例,读起来是覆盖,实际不是。**
 
+**过滤 / 谓词语义裁决:派发令枚举完整的编译面清单,PR 逐面申报 —— 派发令的标准
+条款(#5930 裁决的流程半边)。** 适用判据:本单会**改变一条过滤 / 谓词语义**(算子
+的 NULL 处理、组合子恒等、比较数形状、算子词表……)。满足时派发令**把下面那张表逐面
+抄进去**,并带这一句(原话):
+
+> 本单改的这条语义由**多个互相独立的编译器 / 求值器**各自实现。派发令列出的**每一
+> 面**都必须在你的 PR 正文里有一个结论:**已改** / **本就合规**(给出证据)/
+> **明确不在范围**(给出理由)。⛔ 不许静默略过 —— 评审把「没提到的面」一律读作
+> 「漏掉的面」,不读作「不需要改」。
+
+**这条防的不是「做错」,是「做对了一部分然后以为做完了」。** 一个 `FilterCondition`
+语义由 **5 个互相独立的实现**承载(下表)⇒ 每条语义裁决的成本 ×5,而漏面**反复
+复发**,三次都留在代码注释里:
+
+- **#5146 → #5903**:裁决只落到面 1,面 2 是**不继承面 1 的独立编译器**,于是同一个
+  驱动的两种连接模式对同一条过滤给出两种答案。现场记录在
+  `driver-turso/src/remote-transport.ts:1731`:「LOCAL mode inherits that fix
+  (`TursoDriver extends SqlDriver`), this independent compiler inherited none of
+  it」。
+- **#5326 / #5335**:面 3 与面 4 各**又花一圈**才对齐,记录在
+  `spec/src/data/filter.zod.ts:370`。
+- **#5905**:#5298 的裁决由 PR #5962 落到 driver-sql / formula / service-analytics
+  与 conformance 表,**唯独漏了 HAVING 面** —— `objectql/src/having-filter.ts:37` 的
+  原话是「was not in that PR's inventory, which left this file as the lone
+  holdout」。**「inventory」这个词本身就是本条款的缺席证明**:那次派发确实有一份清
+  单,只是它不完整,而没有任何机制要求它完整。
+
+三次都不是难度问题,是**没有一份清单在问「还有几面」**。
+
+编译面清单(逐面实测 @ `main` `48f98b0`,2026-08-07):
+
+| # | 面 | 落点(file:line) | 备注 |
+| --- | --- | --- | --- |
+| 1 | `driver-sql` | `packages/drivers/driver-sql/src/sql-driver.ts:7083`(`applyFilterCondition`) | `driver-sqlite-wasm`(`sqlite-wasm-driver.ts:67`)与 **local 模式**的 `driver-turso`(`turso-driver.ts:174`)都 `extends SqlDriver`,**靠继承共用这一面**,不单独算面 |
+| 2 | turso RemoteTransport | `packages/drivers/driver-turso/src/remote-transport.ts:1526`(`private buildWhereSQL`) | **独立编译器,不继承面 1** —— 一个驱动的两面,由连接模式选中哪面 |
+| 3 | service-analytics read-scope-sql | `packages/services/service-analytics/src/read-scope-sql.ts:259`(`compileScopedFilterToSql`) | RLS 读侧 |
+| 4 | service-analytics filter-normalizer | `packages/services/service-analytics/src/strategies/filter-normalizer.ts:1235`(`lowerAnalyticsWhere`) | analytics / cube 侧 |
+| 5 | `formula` | `packages/formula/src/matches-filter.ts:73`(`matchesFilterCondition`) | RLS 写侧 `check` 与公式求值;JS 两值语义的基准面 |
+| 半面 | objectql `having-filter` | `packages/objectql/src/having-filter.ts:92` / `:98`(`applyHaving` / `matchesHaving`) | 聚合**后**过滤。算半面是因为词表是子集,**但申报义务不打折** —— 它是**唯一没有 conformance 表覆盖的面**(`FILTER_LOGIC_CASES` 不驱动 HAVING 路径),所以漏了它连门禁都不会红 |
+| 冻结 | `driver-memory` / `driver-mongodb` | — | #5499 冻结投入:**pin-annotate,不翻转**。冻结面仍要申报,结论是「不在范围 + #5499」。现场注释见 `read-scope-sql.ts:176`、`having-filter.ts:41` |
+
+**这张表本身由 PR 维护 —— 与域表同一纪律。** 增删一面(新驱动、新求值器、某面被合并
+或退役、冻结状态变化)的那个 PR 顺手改这里,不留给下一次裁决重新数。清单**会**过期是
+必然的,清单**没有维护者**才是缺陷。
+
+⚠️ 派发前复核一遍再抄,⛔ 不要凭这张表的记忆填派发令:本仓的包路径搬过家(驱动进
+`packages/drivers/`、服务进 `packages/services/`),行号更是每天在动。一条够用的复核
+串:`grep -rn 'matchesFilterCondition\|buildWhereSQL\|compileScopedFilterToSql'
+packages --include=*.ts | grep -v node_modules`。
+
 **Issue 正文是线索,不是规格 —— and the dispatch wording is what makes an
 honest "the premise is dead" cheap to return.** Step 1's stale-premise check
 is the PM's sample; the dev's verification is the real thing, so the prompt

skills/objectstack-pm-dispatch/SKILL.md

@@ -414,6 +414,43 @@ Non-negotiables for this dispatch:
 Return ONLY the JSON report defined in the operating procedure.
 ```
 
+#### One semantics, N independent implementations — enumerate them in the prompt
+
+**Applicability:** the issue changes a *semantic rule* — how an operator, a
+predicate, a comparison, or an absent/empty value is interpreted — and that rule
+is implemented **more than once**, by compilers or evaluators that share no code.
+Query filters, expression languages, permission predicates and serialization
+formats all tend to grow this shape as a project adds backends.
+
+When it applies, the dispatch prompt carries an **explicit inventory of every
+implementing surface**, and requires the agent to give a verdict **for each one**
+in its PR body: **changed** / **already conformant** (with evidence) /
+**explicitly out of scope** (with a reason). A surface the PR never mentions is
+reviewed as one that was *missed*, not as one that needed no change.
+
+Why this is worth a standing clause instead of case-by-case judgment: the failure
+it prevents is not "implemented it wrong", it is **"implemented part of it and
+believed the work was finished"**. That failure is invisible at review time — the
+diff is correct and the tests are green, while the untouched surfaces keep
+answering the old way until a user hits the divergence. The cost scales with the
+count: a semantics carried by N implementations makes every ruling an N-part
+task, and the parts that get skipped are exactly the ones nobody wrote down.
+
+Two disciplines keep the inventory trustworthy:
+
+- **The inventory is maintained by PR.** Whichever change adds, retires or merges
+  an implementing surface updates the list in that same PR. An inventory going
+  stale is inevitable; an inventory with **no owner** is the defect.
+- **Re-verify before pasting.** Paths move and surfaces get added between
+  rulings, so re-derive the list from the code at dispatch time rather than
+  copying the previous prompt. An inventory that was right last month and is
+  pasted unchecked reintroduces the very miss it exists to prevent.
+
+Surfaces that are **deliberately frozen** (deprecated backends, formats kept only
+for compatibility) stay in the inventory. Their verdict is "out of scope —
+frozen", recorded rather than silently absent: a reader cannot otherwise tell a
+frozen surface from a forgotten one.
+
 #### Dispatch backends
 
 **`mode:subagent` (default).** Sub-agents inside the PM's own session. Reports