Repository navigation
Commit 83048c6
pm-dispatch: the spec floor rides its first consumer's PR; bump cards read the hold index (#24315)
Fixes #24276
Clause-②: no
Two maintainer-ordered rules land in
`.claude/skills/pm-dispatch/references/**`. Each is paid for in its own
file, so the net line change is zero per file. There is no new gate, no
ceiling raise, no `CROSS_FILE_MOVES` entry and no edit to
`scripts/pm/check-skill-line-ratchet.mjs`. The PR stays a draft. This is
Tier S: the seat lands it after its in-seat contract review.
The maintainer's words, verbatim from the card: 「下限提到 next.3」, then
「下限提到 next.3我记得之前有一个升级任务,为什么没升级完整,以后如何避免。」, then 「立卡吧,A 和 B 都要」.
## What lands
### A: `dispatch-runbook.md` 〈派发词构造细则〉, right after the file-surface
lines
```text
- 卡的代码首读 spec 发布 R 首发的键、类型成员或运行时行为,即 R 的首个消费者:
- 同 PR 把消费包 `@objectstack/spec` 下限抬到 R,认领文件面点名 manifest 与 lockfile specifier。
- 「不动 pin」只禁 lockfile 解析版本与 `.objectui-sha`/console pin 的移动,⛔ 永不禁下限。
```
How the lines map to the ruling text:
- **Trigger.** "A card whose code first reads a key, type member or
runtime behaviour that spec release R first publishes" is line 1.
- **Duty.** "raises the consuming package's `@objectstack/spec` floor to
R in the same PR" is line 2, first clause.
- **File surface.** "The claim's file surface names the manifest and the
lockfile specifier when this applies" is line 2, second clause. The
condition is line 1.
- **Boundary.** "'No pin bump' bans only the lockfile move and the
`.objectui-sha` / console pin move — ⛔ it never bans the floor" is line
3.
- **One word made explicit.** Line 3 says the lockfile *resolution*
(解析版本). Raising a floor also rewrites the lockfile's `specifier:` line,
and line 2 tells the claim to name that line. If line 3 said only
"lockfile", the two lines would seem to contradict each other. This is
the distinction the ruling already draws, not a new one.
### B: `triage-duties.md`, right after the bump-card cap (bump 单单张封顶)
```text
- bump 单的等待卡清单读自 hold 索引,查询式与命中数落正文;⛔ 凭记忆写的不是该清单。
- hold 索引 = `Restart-when:` 或 `Blocked-by:` 行点名该发布、版本或上游 commit 的全部 open 卡。
```
The second line is the ruling's own definition of the hold index ("every
open card whose `Restart-when:` or `Blocked-by:` line names that
release, version or upstream commit"). The first line carries the rest:
"read from" the index, "the query and its hit count go in the card
body", and "⛔ a list written from memory is not that list".
## Where each rule sits, and why
**A goes in the runbook, not beside `cross-repo-coordination.md` L33.**
I measured both placements against the 120-byte cap.
- **Size.** A is 329 bytes across three lines. Its tightest one-line
trigger-plus-duty form is still 116 to 129 bytes before it names the
file surface. So one line beside L33 can hold only the boundary (line 3,
110 B), never the duty.
- **Splitting.** Splitting A would put the duty and its boundary in two
files that are read at different moments. `cross-repo-coordination.md`
is read at routing and at cross-repo claim or transfer time. The runbook
section is read while an order is composed: `execution-duties.md` 〈派发〉
sends the composer there for 构造细则与条款原文.
- **Where the miss happened.** The incident's miss was in an order: its
fence said "⛔ Do not bump a pin in this PR", and that kept the floor out
of the PR. The runbook section already holds the file-surface sentences
the order copies (文件面两句原文照抄). Putting A beside them means the seat
writes the file surface and the fence in one read.
- **Budget.** `cross-repo-coordination.md` (51/51) had the fewest
payable lines of the candidates.
**B goes in `triage-duties.md`.** The per-`next.N` consumer bump card is
filed by the triage seat. The incident's own card
(objectstack-ai/objectui#12873) says "Filed by the triage seat" in its
first paragraph. So `triage-duties.md` is the file that card's writer
reads, and B sits beside the bump-card cap as the rule for every bump
card's held-card list.
## Payment ledger: every removed or merged clause, and where it survives
All line numbers are at base `21070a2`.
| File | Base line(s) | Change | Clause removed | Where it survives |
|---|---|---|---|---|
| `dispatch-runbook.md` | L202 | line deleted (−1 line, −73 B) |
点名单当天现取仍会漏,dev 对实际改动路径重取补跑 | `execution-duties.md` L133 「点名单是线索不是规格,dev
对实际改动重取补跑」. That file is read by the same composing seat on every fire.
The dev side is in `os-dev.md` 〈本地验证范围〉: derive with no paths, and the
script takes the change set from the merge base. This is a dev duty
restated in a section whose first line (L188) says dev duties are
single-sourced in the os-dev definition and ⛔ not restated. |
| `dispatch-runbook.md` | L232 + L233 | folded into one line (−1 line,
−27 B) | 「⛔ 静默略过」 | The mandated sentence two lines up (L231, 「每一面都必须在
PR 正文有一个结论」) and `execution-duties.md` L142 「编译面清单逐面申报,⛔ 静默略过」. The
failure-mode text stays in the merged line: 评审把没提到的面读作漏掉. |
| `dispatch-runbook.md` | L235 | line deleted (−1 line, −108 B) | 只读读数
dev 不建闭包、不跑测试、不开 PR;只读树与记录,读数贴决策卡或立卡 | `os-dev.md` L355 「派发词标只读读数 ⇒
不建闭包、不跑测试、不开 PR;报 `status: "reading"`,读数贴决策卡」, which is the dev's own
binding text. 「或立卡」 survives as `os-dev.md` L63 (three-class findings go
into the report and the seat files them). This is a second os-dev duty
restated under L188. The PM-side line it followed (L234, the
measurement-first dispatch choice) stays. |
| `triage-duties.md` | L39 + L40 | folded into one line (−1 line, −88 B)
| 落卡必带四棱卡面块…「维护者速读」,⛔ 不留待有人接手再补 | `filing-gate.md` L17 「②
只有维护者能做的决定,落卡即带「维护者速读」与四棱块」. The triage seat reads that file at filing
time (`triage-duties.md` L41 points there). The inbox backfill is L52
(决策箱勤务:入箱校验/补全四棱块、速读). Kept in the merged line: 速读 sits above the block
(其上的 → 居四棱块之上), 同题, business language, and the one-word closing
question. |
| `triage-duties.md` | L55 + L57 | folded into one line (−1 line, −74 B)
| 是分诊固定产出,分诊席唯一权威生产 | `cross-repo-coordination.md` L37
「分诊座位唯一生产:定级/路由/type/查重/shadow/`duplicate_of`」 and L7 (中央分诊是
`domain:*`/type/定级的唯一生产者). Both are read by the triage seat before
routing on every fire. The mandatory-output sense survives as 新卡即时打.
Kept: 立单者可预填、分诊校正, 新卡即时打, 存量卡 ⛔ 不回填、下次碰到补. |
No payment is a re-wrap. Each one deletes bytes that are stated
elsewhere. Three are whole-line deletions; two are folds that drop a
restated clause. Moving whitespace pays for nothing here.
## line_budget
| File | Lines before → after | Ceiling | Headroom after | Bytes before
→ after |
|---|---|---|---|---|
| `references/dispatch-runbook.md` | 241 → 241 | 241 | 0 | 17,950 →
18,071 (+121: +329 rule, −208 payment) |
| `references/triage-duties.md` | 120 → 120 | 120 | 0 | 12,261 → 12,329
(+68: +230 rule, −162 payment) |
Every touched line is ≤ 120 bytes. The widest is 120 B, on the folded
`type` line.
## Worked example for B (an example only, not rule text)
The triage seat's reading on the card (comment 6248125841) names three
objectui cards that wait on the next release:
objectstack-ai/objectui#12932, objectstack-ai/objectui#12929 and
objectstack-ai/objectui#12484. Each is `pm:blocked`. I read each card on
its own through REST on 2026-10-18 at about 11:05Z. Each body carries a
line-anchored `Blocked-by: #24168`, the
Version Packages PR that publishes `next.4`, plus the upstream commit
the card waits for (`586705d85d`, `1e001c7df`, `e15897c0f1`).
So the next objectui consumer bump card should get its held-card list
from one query over objectui's open cards. That query matches
`Restart-when:` / `Blocked-by:` lines that name `next.4`,
`#24168` or one of those commits. Today it
should find at least these three. The bump card's body records that
query and its hit count. I did not run the enumeration myself, because
the os-dev definition keeps the dev off open-issue scans. The bump-card
writer runs it.
## Verification (at head `6cfc7839`, after the merge of `origin/main`
`0b44840d`)
The change set is 2 paths, +8/−8. `node scripts/pm/dispatch-gates.mjs
--commands --repo objectstack-ai/objectstack` derives 19 commands, the
same 19 the seat derived. I ran all 19 locally. I also ran `pnpm
check:pm-settings-deny-roster`, which the derivation flags because its
roster lives under `.claude`, and the local-only `pnpm
check:skill-frame-freshness`. Each exit code was captured before any
pipe. `--ran` reconciles 19 derived / 19 run / 0 NOT-MEASURED / 0 UNRUN.
```text
node scripts/check-closing-keyword-parity.mjs :: exit 0
node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0
node scripts/check-comment-mask-corpus.mjs :: exit 0
node scripts/pm/check-harness-current.mjs --self-test :: exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0
pnpm check:agent-test-spelling :: exit 0
pnpm check:cross-package-test-inputs :: exit 0
pnpm check:doc-authoring :: exit 0
pnpm check:driver-memory-census :: exit 0
pnpm check:gitlink-declared :: exit 0
pnpm check:nul-bytes :: exit 0
pnpm check:pm-governed-merges :: exit 0
pnpm check:pm-half-states :: exit 0
pnpm check:pm-label-write :: exit 0
pnpm check:pm-skill-id-lint :: exit 0
pnpm check:pm-skill-ratchet :: exit 0
pnpm check:refd-timer-probe :: exit 0
pnpm check:skill-frame-sync :: exit 0
pnpm check:watch-hint-literal :: exit 0
pnpm check:pm-settings-deny-roster :: exit 0
pnpm check:skill-frame-freshness :: exit 0
```
`pnpm --filter @objectstack/lint run check:doc-formula-expressions`
first exited 3 (PREREQUISITE NOT MET: `@objectstack/formula` and
`@objectstack/lint` were not built). I built that closure under
`scripts/pm/os-verify-lock.sh` (`VERDICT command-exit 0`) and re-ran the
gate. After the merge I rebuilt the closure again. `origin/main` moved
once more while the gates ran, to `52ff308b` (a REST route refactor).
That commit touches `scripts/check-route-envelope.mjs` and one
`scripts/adr-anchors/` file, which are none of this card's 19 families
and no path under `.claude/**`. I did not merge it again; the merge
queue rebuilds onto the current `main`.
The families CI runs (Lint & Repo Gates, TypeScript Type Check, Test
Core, Build Core, Governed Surface Queue Guard and the rest) are left to
this PR's check-runs. I did not re-run them locally. No package is
touched, so there is no build closure, package test or typecheck to owe.
## Acceptance notes
- **Not filed: the objectui consumer bump card has no line of its own.**
`triage-duties.md` names the console bump card and the cloud
`.objectstack-sha` bump card as cases, but not the per-`next.N` objectui
consumer bump card that the triage seat files by standing practice. B is
written to cover every bump card, so it covers that card too. This is an
observation, not a defect, a contract violation or a metadata trap, so
it is not filed. Carrier: none (承接者:无).
- **The fence's source sentences are unchanged.**
`cross-repo-coordination.md` L33 (⛔ pin bump 不做 rider) and `lanes/ui.md`
L20 (console pin bump 不做 rider) still read as before. A's boundary sits
where the order is composed, not beside them (see placement above).
## 维护者速读(草稿)
**改了什么**:PM 技能参考文件新增两条规则。A:哪张卡的代码第一次用到 spec 某个新版本才有的东西,就在同一个 PR 里把该包对
`@objectstack/spec` 的版本下限抬到那个版本;派发令里的「不动 pin」只管 lockfile 解析版本和 console
pin,不管下限。B:每张版本 bump 单列「等这个版本的卡」时,必须按 `Restart-when:`/`Blocked-by:`
行查出来,并把查询和命中数写进正文,不许凭记忆列。
**为什么改**:next.3 那次,一张卡已经用上 next.3 才有的类型,包的下限却还停在 next.2(派发令写了「不动
pin」,把抬下限也挡在外面);同时 bump 单凭记忆列等待卡,漏了一张正在等 next.3 的卡。这两条规则把这两个漏洞都补上。
**风险与代价(含回滚)**:只改给 PM
席位读的规则文本,不碰代码、门禁和发布物;两个文件行数都不增加,新增内容靠删掉别处已写明的重复句子付账,每一处删除及其存活位置都列在正文的表里。回滚即
revert 本 PR。
**席位意见**:
**你要做的**:无需操作。席位复核 PASS 后由席位入队落地(Tier S)。
---
_Generated by [Claude
Code](https://claude.ai/code/session_01V1upgJbYCZ5xed9pEQF4Tz)_
Co-authored-by: Claude <noreply@anthropic.com>1 parent 0d64aa7 commit 83048c6
2 files changed
Lines changed: 8 additions & 8 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
199 | 199 | | |
200 | 200 | | |
201 | 201 | | |
202 | | - | |
203 | 202 | | |
204 | 203 | | |
205 | 204 | | |
| |||
209 | 208 | | |
210 | 209 | | |
211 | 210 | | |
| 211 | + | |
| 212 | + | |
| 213 | + | |
212 | 214 | | |
213 | 215 | | |
214 | 216 | | |
| |||
229 | 231 | | |
230 | 232 | | |
231 | 233 | | |
232 | | - | |
233 | | - | |
| 234 | + | |
234 | 235 | | |
235 | | - | |
236 | 236 | | |
237 | 237 | | |
238 | 238 | | |
| |||
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
27 | 27 | | |
28 | 28 | | |
29 | 29 | | |
| 30 | + | |
| 31 | + | |
30 | 32 | | |
31 | 33 | | |
32 | 34 | | |
| |||
36 | 38 | | |
37 | 39 | | |
38 | 40 | | |
39 | | - | |
40 | | - | |
| 41 | + | |
41 | 42 | | |
42 | 43 | | |
43 | 44 | | |
| |||
52 | 53 | | |
53 | 54 | | |
54 | 55 | | |
55 | | - | |
| 56 | + | |
56 | 57 | | |
57 | | - | |
58 | 58 | | |
59 | 59 | | |
60 | 60 | | |
| |||
0 commit comments