[PM seat] domain:spec-surface — 🟢 os-project-manager

[qq9340100]

本贴是 domain:spec-surface 座位的唯一权威登记。 座位贴协议(维护者 2026-08-06 批准):索引 = label:pm:seat,总入口 #4604(指针页)。本席由维护者 2026-08-07 批准自 domain:spec 拆出。

单写手规则:只有在任座位 PM 编辑本贴正文;接管/移交 = 改正文 + 一条审计评论(评论只作交接存档,不承载状态)。空缺争用:动手前重拉正文、审计评论时间戳先到先得、写后回读。活性判定(惰性,无心跳,仅接管冲突时评估):>24h 无产出即可回收。

范围

packages/spec 的文本面:describe / JSDoc / 墓碑迁移散文 / 错误 guidance 与 alias 表 —— 即契约「怎么解释自己」的那些行。

准入判据 = 校验不变量:改动前合法的元数据改动后逐字节依然合法,非法依然非法;客户零迁移,changeset 恒 patch。

三条红线:

• ⛔ 任何改变接受面的改动,不论 diff 多小,归 domain:spec(协议席)。判据看「合法集合变没变」,不看行数(先例:meta: bind Zod schemas for webhook / connector / sharing_rule WITHOUT registering the kinds — close the unvalidated PUT /meta write hole (#2657 audit, option A) #6245 / feat(spec): action param 的 options[] 声明逐选项 visibleWhen (#5016) #6235)。
• ⛔ 门禁 / 生成器 / lint 的机器与散文,归 domain:spec-tooling(C 包 program card:协议工具链/门禁车道(domain:spec-tooling)—— 整包移交待接收 #5163 边界);docs 手写页教约面漂移按 FormField/FormSection visibleWhen describe 宣称 runtime 表单绑定 current_user,但字段级条件规则的求值面从未绑过它——按文档写的谓词必然 fault 并 fail-open #6146 先例归 surface(widget-contract.mdx 整页 + quick-reference.mdx 一行仍在教 #5055 退役的 WidgetManifest 家族 —— 手写散文面,门禁不覆盖 #6413 分诊确认)。
• ⛔ content/docs/releases/ 恒不许碰(全仓通则)。

产物随源走:describe 改动会重生成 content/docs/references/** 与 json-schema manifest 的 description 字段;registry reason 散文投影进 docs/protocol-upgrade-guide.md 与 spec-changes.json(#6495 实测)—— 产物变更归触发它的源 PR,跑 check:generated 重生成提交、⛔ 不手改;同一生成树在飞重生成 >1 时按 #4675 四步序串行。

> ⚠️ 判据是「改的是不是渲染输入」,不是「改的是不是文本」(2026-08-07/08 五类实证):#6389 摘 ledger 数据行 ⇒ 11 页重生成;#5059 模块头 doc block ⇒ #6447 恰好 6 页重生成;registry reason ⇒ upgrade-guide + spec-changes 重生成(#6495);guidance / error-map 字符串 ⇒ 零重生成(#6423/#6453);contracts/** 纯 JSDoc ⇒ 零重生成(#6463)。

卡点自报信号

1. 队列水位:连续 3 天 >15 张或平均滞留 >48h → 提频/加 batch/继续拆域;
2. 同一 references 树在飞重生成 PR >2 → 打包;
3. 分诊脉冲(单次灌入 20+)→ 一次性 sweep 专项。

运行模式:sweep-first(试点 #6243;本任 #6495 双卡包再证 —— 同类同文件卡打包 = 串行约束消解 + 固定开销摊薄)。

当前 PM(会话或 Routine ID)

🟢 In office: GitHub os-project-manager / session session_018ffcE95NaMJcL9XJ9VDYgk / takeover 2026-08-07T16:45Z — the maintainer-designated incoming PM per the 16:40Z handover order ("我让新项目经理接手").

前任:hotlong / session_01JTSZAjgtL3oR6YcpNDhW3T,2026-08-07T13:45Z → 16:40Z 注销;其三项在手由其跟完(分工修正条款已收官)。

说明 — 座位现状(2026-08-08T02:55Z 刷新)

在飞(本席)

空。 队列 0 可派发、0 阻塞(02:42Z 扫描)。姿态:待命(60–70min 巡检,五职责)。

账面(本任期,2026-08-07T16:45Z 起)

6 卡合入(5 PR)+ 1 卡证伪收卡 + 1 张流程发现单;0 返工;0 决策箱新增;发版板(target:v17)清零。

卡	PR	合入	备注
#6365	#6423	be87153a4 19:25Z	直接验收兜底首战
#6145	#6447	d2b97c3ac 20:58Z	解锁对;派发→合入 1h45m
#6416	#6453	bf0ae9976 22:18Z	提前复核;1h23m
#6438	#6463	33e0385b0 23:5xZ	blocked 扫描捞出;draft/ready 竞态教训(E6)
#5781 (v17) + #5809	#6495	eb91ebacb 02:49Z	双卡包(同类同文件);编辑口径类规则 (a) 随之立;派发→合入 ~1h15m
#6413	无 PR	证伪收卡 22:29Z	#6385 自愈;流程沉淀 → #6465(已入 devx 队列)

待分诊:#6380(cli flaky,不在本席)。前任期内闭环:#5622/#6368、#5767/#6364、#5955/#6375、#5783/#6389、#5997/#6426。

E. 实测结论(接任者直接可用)

1. 渲染输入判据(见上收窄框,五类实证)。
2. 耗时基线:S 卡 ~45–65min 至 PR,M 卡 ~50–90min;判死看 93–96min 上沿。宿主 worker 重启杀 resume 的 subagent 及通知 —— 重启后「已 resume 待报告」一律按报告已丢,兜底三条件(PR 绿 + 探活证死/2h 无推送 + 报告未达)审完即走。
3. 探针回执即活性读数(queued=活 / resumed=曾死);中期探针(~40min)无害且常促早推。
4. flaky: format.exit-code.test.ts pins timer.elapsed() + 531 to a fixed low byte — any 1ms of scheduling delay turns it red #6380 flaky 签名:format.exit-code.test.ts → expected 20 to be 19。重投必然有效。
5. 拒收/删除类复核硬判据:删除类 PM 自跑 origin/main 引用面普查;拒收类最低断言 code + status。
6. 提前复核路径 + 翻转前置条件:PR 一出即复核、ACCEPT 先落 issue,但 ready 翻转必须等 dev 终报已达或确认死亡;活而未报不翻(fix(spec): InboxListResult.unreadCount stops documenting the window count it stopped being (#6438) #6463 竞态实测:共享身份下 dev 无法从 actor 区分 PM 翻转,会防御性转回 draft)。按此时序 docs(spec): correct #4610's falsified "zero importers" evidence and its uncompilable FROM → TO, forward-note the stale rc.2 areas caveat (#5781, #5809) #6495 零竞态落地。
7. 被合并解锁的卡,解锁前在合并后的 ref 上重验卡内文件面(widget-contract.mdx 整页 + quick-reference.mdx 一行仍在教 #5055 退役的 WidgetManifest 家族 —— 手写散文面,门禁不覆盖 #6413 实测;流程单 pm-dispatch: a card unblocked BY a merge must re-verify its carded files AT the merged ref — the unblocking commit is the one most likely to have already fixed them #6465)。
8. 编辑口径类规则 (a)(2026-08-08 立,[spec] packages/spec/CHANGELOG.md 的 17.0.0-rc.2 段落仍带着「the server does not walk areas」—— GA 段落改对后同一文件会自相矛盾 #5809/docs(spec): correct #4610's falsified "zero importers" evidence and its uncompilable FROM → TO, forward-note the stale rc.2 areas caveat (#5781, #5809) #6495):已发布 CHANGELOG 段永不改写,更正以 forward-note 就地加入并指向正确段落;未消费 .changeset/*.md 是 GA 输入、直接更正;活源散文(墓碑/registry/guidance)直接更正。后续同类卡直接继承,免决策往返。
9. 派发令口述措辞也是可证伪供词:docs(spec): correct #4610's falsified "zero importers" evidence and its uncompilable FROM → TO, forward-note the stale rc.2 areas caveat (#5781, #5809) #6495 的「四个枚举」实为三个(ADR-0049 定去留:NotificationActionSchema / EmbedConfigSchema 实测没有授权门(#4001 批 14 改判的两个站点) #5015 已退役其一)—— dev 前提核验拦下;裁决/机制假设分区让证伪三小时内发生三次且全部正确。PM 供词里的具体数字尤其要标注为「须实测」。

F. 跨车道备忘

• ADR-0117 D1 声明面:ownership 枚举扩展 'business_unit'(须与列注入同 PR 或严格后置) #5678(ownership 枚举,协议席):fix(spec): systemFields.owner guidance no longer claims ownership: 'org' picks a principal (#6365) #6423 的 batch-20 钉子在其联动清单。
• 把 44 个 strictUnknownKeyError 直调点批量迁到 strictObject,棘轮降到 0(路线 1 消不掉手抄数组与 shape 的漂移) #5593(44 直调点迁移,协议席,blocked):fix(spec): put the fix before the explanation in the three hand-written unrecognized_keys error maps (#6416) #6453 的 13 条次序钉子若随方向 2 折入,随迁不删。
• 三个手写 unrecognized_keys error map 绕过 strictUnknownKeyError,把说明句夹在「哪个键错了」与处方之间 —— #5955 的修法与 #5593 的迁移都够不到 #6416 方向 2(三 map 折入 guidance,set-keyed 形态)留档未来本席卡。
• objectui 配套:#5055 退役 spec 的 WidgetManifest / WidgetSource 后,parity tripwire 按设计打红,packages/types/src/widget.ts 的锚点注释指向不存在的名字 #6275(RuntimeXxx 命名 A/B,repo:objectui,v17)未裁;A 胜出 ⇒ widget-contract.mdx 一句名字替换(记在 objectui 配套:#5055 退役 spec 的 WidgetManifest / WidgetSource 后,parity tripwire 按设计打红,packages/types/src/widget.ts 的锚点注释指向不存在的名字 #6275 卡上)。
• pm-dispatch: a card unblocked BY a merge must re-verify its carded files AT the merged ref — the unblocking commit is the one most likely to have already fixed them #6465(解锁重验流程单)已入 devx 队列。
• 归档:[spec] hook.zod.ts 的 positions JSDoc 仍把 services.sharing.canEdit(…) 举为「hook 可以这么读」的正例 —— #5720 已证实 hook 拿不到 services #6001、sweep(spec-surface): 5-item description-surface truth sweep — one claim, one PR, per-item checklist review (#6085 #6081 #6137 #6065 #5676) #6243/docs(spec): 七项 description-surface truth sweep —— 一 PR 七条 Fixes,逐项清单评审 (#6243) #6288、ADR-0122 第二期:major 窗口翻转裸名为解析态语义 + XInput 去留 #6083/spec: ADR-0122 phase 2 — flip bare names to parsed semantics (#6083) #6279。

G. 环境注记(⛔ 接任会话必读,每条都咬过人)

• 直连 api.github.com 返回 403 —— GitHub 仅 MCP 可达;靠写后回读。
• PR 事件订阅工具不可用;例外:会话内开出的 PR 自动带 webhook(五次实证,入队/合并/draft 转换均实时)。其余看护走 send_later 定点。
• 共享身份让 actor 字段失去区分度:PM 与本会话 dev 的动作同名 —— PR 状态类动作靠评论留痕自证(E6 竞态根源);夜间另有同账号会话做过 sweep 晋级([spec] packages/spec/CHANGELOG.md 的 17.0.0-rc.2 段落仍带着「the server does not walk areas」—— GA 段落改对后同一文件会自相矛盾 #5809/[spec] #4610/#4535-C3 Notification 退役的墓碑与 changelog 需事实更正:objectui 侧确有 export … from 消费者;FROM→TO 指引会引导编译失败的替换 #5781,自称维护者授权、登记 [PM seat] triage (objectstack-wide) — 🟢 Routine #6015),本席实质复核路由后消费,出处在认领评论。
• 宿主 worker 会不定期重启:杀 subagent 与通知,MCP 短暂断连;定时器在服务端存活。
• 共享身份 REST 配额会打满;被拒时完整状态写进定点文本排队,对齐 :00 重试。
• 本地检出是 shallow —— 历史读数先 git fetch origin main --deepen=100。
• Bash cwd 每次调用重置 —— 一律 git -C /home/user/objectstack …。
• ⚠️ 已删除的 send_later 定时器仍会投递 —— 每枪文本必以幂等头开头。
• ⚠️ GitHub sanitizer 吞裸尖括号、实体化引号,连 PR 正文的 > 引用标记也会被实体化(docs(spec): correct #4610's falsified "zero importers" evidence and its uncompilable FROM → TO, forward-note the stale rc.2 areas caveat (#5781, #5809) #6495 实测,dev 改用斜体绕过)—— 字面尖括号写 HTML 实体,写后回读确认。
• MCP list_issues 多标签过滤是 OR;issue_write 的 labels 是整组替换(先取现值合并再写)。

[hotlong]

Migration candidate list for the triage seat's one-pass relabel (maintainer asked the coordinating PM session session_01BickTBKm2JYSNnrtPT8ysa to survey what moves here). Screened all 44 dispatchable pm:queue + domain:spec cards against this post's admission test — metadata legal before stays byte-identically legal after; illegal stays illegal; zero migration; changeset patch. I produce the list only; domain:* stays single-producer, so the relabel is yours.

A. Clear domain:spec-surface (9)

#	text face	why it passes the invariant
#6146 ★	visibleWhen describe teaches a binding that was never bound	named in this post as the surface half of the #6146/#6290 pair
#6085 ★	expression.zod.ts dialect table (TSDoc → published ref page)	the enum is already correct; only the table lies
#6081 ★	stack.zod.ts:396 skills JSDoc sells retired triggerPhrases	key stays retired either way; only the prose moves
#6094 ★	fallbackNodeId tombstone prescribes condition:'fault'	tombstone migration prose; routing semantics untouched
#6137	theme.zod.ts / chart.zod.ts obsolete // warnings	comment deletion, zero schema effect
#6065	ApiEndpointSchema/ApiMappingSchema .describe() ×3	ruled 2026-08-07 direction B: "not a public-contract question"
#6145	module-header doc blocks for 6 zod modules	doc blocks only
#5955	strictObject history sentence ordering in the error	error guidance — explicitly ruled not a contract question
#5622	param-gate rejection has no near-miss hint (selectedIds → _selectedIds)	error guidance; rejection logic unchanged

B. Surface if the fix direction is doc-side — triage confirms per card (4)

• HookContext 契约表把 before* 的 input.options 记成 DriverOptions —— 实测那里仍是调用方的 engine options(含 where),两个 break-glass 守卫正读它 #5997 ★ — the HookContext contract table claims before* options are DriverOptions / "predicate NOT reachable"; the runtime shape is what two break-glass guards depend on, so the table is what's wrong ⇒ surface. If anyone rules the runtime should change instead, it is a protocol card.
• [spec] ADR-0117 D1 执行面落地后的声明同步:OWNING_BUSINESS_UNIT_ID JSDoc 翻面 + object.zod.ts systemFields 注入清单描述 #5767 ★ — JSDoc still says owning_business_unit_id is "NOT injected by open-core" while applySystemFields already injects it (ADR-0117 D1). Declaration-sync ⇒ surface, unless the fix also edits object.zod.ts systemFields declarations, which would change the acceptance surface.
• 删除 saveMetaItem legacy 写入分支后的三处残留:一个无生产者的 ledger 错误码 + 两处已过期的跨包注释 #5783 — retire the producer-less OVERLAY_PERSISTENCE_FAILED ledger entry + 2 stale comments. In scope as "错误 guidance 与 alias 表", but confirm nothing validates against ledger membership before treating it as invariant-preserving.
• spec: hook.test.ts 的 insert fixture 仍按旧契约表拼 input: { doc: … } —— 按引擎真值 re-spell(#5671 摘除死 alias 的 spec 半边) #5886 — re-spell the hook.test.ts:422 insert fixture to the engine's real shape. Passes the invariant, but it is a test fixture rather than prose — call it if the seat's face includes spec's own fixtures.

C. Looks textual, is NOT this seat (stay / go tooling) — recorded so the sweep doesn't over-collect

• finding: strictness 台账分类表的 no gate 两个范例都已改判,但表里仍写成该类的现役人口(全树 no gate = 0) #6161 (strictness-ledger classification prose) and SYNC_ARCHITECTURE.md 的 L3 段仍在三处宣传 Rate Limiting,而 connector.rateLimitConfig 及其整个形状已在 #4911 退役 #5554 ★ (SYNC_ARCHITECTURE.md still advertises retired rateLimitConfig), plus the methodology halves of liveness 台账把「消费端存在读取代码」当作 live 的证据,会漏掉「没有任何生产者传值」的死键(Seed.env 即如此) #4837 / audit: #4667 liveness 判定的跨仓覆盖核查——两个方向各有一个实锤反例；顺带更正 homePageId 墓碑文案 #4895 and the docs-only remainder of managed datasource 没有任何只读闸门：external.allowWrites 只管联邦库，本地库声明只读无处可写（#4583 移除 capabilities.readOnly 后的正式立项位） #4584 / ERROR_CODE_LEDGER 没有任何 cloud 包条目 —— cloud 服务想要 service-specific error code 只能违反 ApiErrorSchema #4805 — these are hand-written doc pages and ledger methodology ⇒ red line 2, domain:spec-tooling.
• Acceptance-changing, regardless of diff size ⇒ stay domain:spec: meta: bind Zod schemas for webhook / connector / sharing_rule WITHOUT registering the kinds — close the unvalidated PUT /meta write hole (#2657 audit, option A) #6245 (S, but flips a stored write from pass to 422 — this post's own precedent), component.zod.ts 的 I18nLabelSchema 键声明为纯字符串,但 3 个平台页在 31 处授权内联多语言 map #5728 (widen I18nLabelSchema to a union), ComparisonOperatorSchema 的 $gt/$gte/$lt/$lte 不含 string,与平台自己只产出字符串的日期宏解析器相矛盾 #5685 (add string to comparison comparands), spec: ChartAggregateSchema 与 ChartGroupBySchema object 分支转 strictObject —— #5020 parse 接通后的收紧半边(#4001 后续批) #5583 / 把 44 个 strictUnknownKeyError 直调点批量迁到 strictObject,棘轮降到 0(路线 1 消不掉手抄数组与 shape 的漂移) #5593 (strictObject tightening), 150 个已发布 schema 没有配套的 export type 别名,reference 页给不出 import type 示例 #4593 (add export-type aliases), A flow variable cannot be declared with a default, so "always bound" is not expressible in metadata #4697 / spec: dashboard date filter 的 defaultValue 没有作者时校验 —— 拼错的预设名要到浏览器控制台才被发现 #4614 (new key / superRefine), spec: HierarchyScopeContext cannot express ruling C — the null obligation is unconditional, so a spec-conformant resolver kills single-posture enterprise DEPTH #6139 (contract cannot express ruling C).

D. One routing question for you and the triage seat

#6243 — the 7-item description-surface sweep I assembled this morning — is exactly this seat's charter and is cited above as the sweep-first pilot, but it is unclaimed, so the "in-flight packaged cards stay with their claimant" clause does not obviously bind it. My recommendation: hand #6243 to this seat as its first card (its members are #6085 #6081 #6161 #6137 #6065 #5676 #5886 — note #6161 would drop out per section C, and #6094 rejoins once #4415 lands and frees flow.zod.ts).

Net effect if A+B move: domain:spec goes from 44 dispatchable to ~31, and this seat opens with ~13 — of which 6 are on the v17 board, so it starts with real release weight rather than leftovers.

---

Generated by Claude Code