Repository navigation
Commit eb833b5
docs(pm-dispatch): triage round reads H14 and writes pm:blocking through label-write (#29862)
<!-- claude-projects-attribution: {"github_login":"os-support-ai"} -->
_Requested by **Yibo Wang** · [project
thread](https://claude.ai/code/project/chan_01BVUyyPie8gHwoPM18iABwB?thread=cmsg_01BVUyyPie8gHwoPM18iABwBAgUoQmc6qtKPen78izJ97A)_
Fixes #29646
Clause-②: no
Metadata-protocol: none
**Rules text only.** Two lines are added to
`.claude/skills/pm-dispatch/references/triage-duties.md`; nothing
published moves (`skip-changeset`). No new gate, no new
`check-half-states` row, no new label or state.
## 维护者速读
- **改了什么**:分诊座位职责里补两行步骤:每日对账 fire 读 `check-half-states` 的 H14 行,「NOT
carrying」行(在飞卡也一样)用 `label-write.mjs --add pm:blocking` 挂上;「stale」与
UNJUDGED(seam 卡)行先手核跨仓依赖再 `--remove`,留下的写明理由;增删的卡号写进轮次记录。
- **为什么改**:`pm:blocking` 是车道取卡序的第二位,运行细则说它「写入与摘除都在分诊轮
sweep」,但分诊的轮次步骤从未写这一步,H14 也只报告不写;结果板上没有一张 open 卡带它,第二位形同虚设(本卡 Ask 1 已由分诊
2026-11-03 回填 12 张)。
- **风险与回滚**:只加两行规则文字,无代码无发布面;回滚即还原本 PR。stale 方向保留了 H14 自己的保护(先核跨仓依赖、seam
卡不判),不会误摘。
- **先看**:`triage-duties.md` 第 34–35 行,紧跟「停 `pm:blocked` 同笔写
`Blocked-by:`」那条;diff 只有这两行。
- **两分钟验证**:`pnpm check:pm-skill-ratchet && pnpm
check:pm-skill-id-lint`,预期两条都 ✓(文件 97→99 行,天花板 99;两行各 114 字节)。
- **需要你做的**:无,席位落地(Tier S,`.claude/**` 内)。
## What changes
One file, two lines added, 97 → 99 lines against the ratchet ceiling of
99 (the file's whole headroom). Both lines sit under the 120-byte line
cap (114 B each) and carry no issue number (`check-skill-id-lint`). They
are placed directly after the existing rule that writes the
`Blocked-by:` line, because that line is the index H14 derives the label
from.
```diff
@@ triage-duties.md, after line 33 (停 `pm:blocked` 同笔在卡正文写行首 `Blocked-by: #N` …) @@
+- 日对账 fire 读 `check-half-states` H14,在飞卡亦然:NOT carrying ⇒ `label-write.mjs --add pm:blocking`;
+- stale/UNJUDGED 行手核跨仓依赖后 `--remove pm:blocking`,留则写明理由;增删卡号进轮次记录。
```
What the two lines say, read against H14's own output:
- **Add direction.** The daily two-repo reconciliation fire (line 8,
「每日一 fire 跑两仓全量对账」) reads the H14 rows from the seat's own
`check-half-states.mjs` run, which is how the 2026-11-03T00:12Z round
did it; tying it to that fire rather than 「每轮」 keeps a whole-repo sweep
off the hourly round. A row reading `targeted by N open card(s)'
Blocked-by: line … but NOT carrying pm:blocking` names the card number
and the direction; the seat writes `label-write.mjs --add pm:blocking`
for it through the relay like every other label write. 在飞卡亦然: the write
applies to a dispatched or assigned card too, as an explicit exception
to line 77's 「在飞卡不动」, because a claimed blocker is exactly the card
`execution-duties.md` line 48 needs labelled.
- **Remove direction.** A row reading `pm:blocking carried while no open
card in THIS REPO targets it … a stale derived cache`, and a row H14
marks UNJUDGED (a seam card carrying `repo:*`), take the same path: the
seat reads the cross-repo dependents by hand (H14's index is repo-local
by construction; a dependent in objectui or cloud is invisible to it),
then `--remove pm:blocking`, or keeps the label and writes why in the
round record. Folding UNJUDGED into this branch is H14's own remedy for
a seam card; parking those rows would leave a standing finding every
sweep.
- **Round record.** The added and removed card numbers, and the reason
for any kept stale/UNJUDGED label, go into the round record, which is
what the triage comment on the card already promised from that round on.
Nothing else moves: `dispatch-runbook.md` line 8 (「`pm:blocking`
的写入与摘除都在分诊轮 sweep,依赖者全关即摘」) stays as the rule statement,
`execution-duties.md`'s pull order stays the reader,
`check-half-states.mjs` stays report-only, and `label-write.mjs` is the
existing four-step write — the lines name the spelling it already has.
## Why
**Why the sweep never wrote the label (measured on `origin/main`
55dca46, the branch's base).** The write existed only as a delegation
with no step at the other end:
- `dispatch-runbook.md` line 8 says the write and the removal happen in
the triage round's sweep — a statement of *where*, not a step.
- `references/triage-duties.md` (97 lines, the only text that lists what
a triage round does: the five-disjunction backlog sweep at lines 12–21,
the grading rules, the `Blocked-by:` line write at line 33) never names
H14, `check-half-states.mjs`, or `pm:blocking`. Its sweep reads labels
only; nothing in it builds or reads a `Blocked-by:` reverse index.
- `references/lanes/triage.md` (17 lines) lists what the seat pushes to
the maintainer and says the seat 「只写标签与审计评论」, but names no `pm:blocking`
write either.
- `scripts/pm/check-half-states.mjs` H14 (`h14BlockingCacheIncoherent`,
~line 4124) is the only code that compares the cache with the index, and
every one of its sentences ends in 「Report-only: the remedy is the
triage sweep's derivation pass … never a label written from this script」
— it hands the write back to a 「derivation pass」 that no procedure text
defines.
- No script writes the label: `ensure-pm-labels.sh` creates it,
`close-cards.mjs` strips it as residue on close, `label-write.mjs` is
the generic four-step write, and nothing else in `scripts/pm/**` or
`.github/workflows/**` names `pm:blocking` (`grep -rn pm:blocking`,
excluding `check-half-states.mjs`: those three files only).
So the runbook pointed at the sweep, the sweep's script pointed at 「the
derivation pass」, and the triage round steps contained neither. The
label was written only when someone remembered (the backfill of 12 cards
on 2026-11-03T00:12Z), which is exactly the dependence the card's Ask 2
asks to remove. `.github/workflows/half-state-patrol.yml` declares no
schedule since ruling 208 (on-demand only), so the seat's own round run
is the read this step depends on, not a workflow.
**Why route (a) and not an H14 output change.** The order file's smaller
fix first: H14's printed row already carries the card number (as a link
in `--format=markdown`, as `#N` in the default output) and the direction
in its first words (`NOT carrying` vs `carried while no open card`),
with UNJUDGED rows marked by the `UNJUDGED` prefix. A reader can act on
it as printed, so the procedure line suffices and
`check-half-states.mjs` is untouched (no self-test pin owed).
## Tests
None owed: prose in a `.claude/**` reference file, no code or test
moved. The two skill gates that judge this file are in Verification.
## Acceptance notes
- **Line budget.** `triage-duties.md` 97 → 99 against ceiling 99;
headroom now 0. A later rule in this file will have to pay by compaction
or a ruled raise.
- **Wording.** 「NOT carrying」 and 「stale」/「UNJUDGED」 are H14's own row
words, kept in English so the seat matches the printed row rather than a
translation; 「日对账 fire」 is line 8's own term. 「手核跨仓依赖」 restates H14's
own remedy condition rather than adding a new one. Patch round 1
(74dc29a, from the at-tier review's should-fixes): 「每轮」 → 「日对账 fire」,
「在飞卡亦然」 added, 「UNJUDGED 行不动」 folded into the stale branch with
「留则写明理由」; net 0 lines, line 77 untouched.
- **Attribution thread link.** The order file spells the skills-lane
thread in the attribution line; this body carries this card thread's
link because the session's own attribution rule names it. Same project,
same requester.
- **Shallow clone.** `git log -S pm:blocking` over the skill tree is NOT
MEASURED here (shallow checkout, `--is-shallow-repository` true), so
whether an earlier compaction once removed such a step is not claimed
either way; the Why section rests on the current tree only.
- Out-of-scope findings: none.
## Verification
- `pnpm check:pm-skill-ratchet` on 4e45c01 and 74dc29a: ✓
(`triage-duties.md` 99 lines at ceiling 99; line-byte check clean).
- `pnpm check:pm-skill-id-lint` on 4e45c01 and 74dc29a: ✓ 109 file(s)
clean.
- `node scripts/pm/dispatch-gates.mjs --commands --repo
objectstack-ai/objectstack` derived 19 commands for the one changed path
(merge base 55dca46); each run serially with the exit code captured
before any pipe, then `--ran`: 19 derived, 19 run, 0 NOT-MEASURED, 0
UNRUN, exit 0. `check:doc-formula-expressions` first exited 3 (missing
`@objectstack/formula` / `@objectstack/lint` dist); after `turbo run
build --filter=@objectstack/formula --filter=@objectstack/lint` under
`os-verify-lock.sh` (VERDICT command-exit 0, 136 s) it exited 0.
- `check-half-states.mjs` not touched, so no `--self-test` owed; `pnpm
check:pm-half-states` (its self-test gate) was in the derived 19 and
exited 0.
- Live `check-half-states.mjs` run from the card thread on 55dca46
(2026-11-03 ~08:45Z, exit 0): H14 rows 0 (0 missing, 0 stale, 0
UNJUDGED) over 216 open issues, `Blocked-by:` comment fallback read 28
of 28; H15: 11 open `pm:blocking` cards. The card's done-when half 1
holds on a fresh reading; this PR is half 2.
- CI check-runs: in_progress at PR open; the seat reads them at ACCEPT.
## Human Minimum
Each item is one rule from
[CONTRIBUTING.md › Human
Minimum](https://github.com/objectstack-ai/objectstack/blob/main/CONTRIBUTING.md#human-minimum),
which links the AGENTS.md text that states it.
- [x] 1. Schemas start as Zod; `packages/spec` holds no runtime logic.
(not touched)
- [x] 2. Every name I added uses the casing for its kind. (no names
added)
- [x] 3. No formatter ran; `pnpm lint` is CI's run (repo-level scan).
- [x] 4. One issue, one branch off an up-to-date `main`, one pull
request.
- [x] 5. A change that publishes carries a changeset, and both
declaration lines are filled in here and in the changeset. (`.claude/**`
is not published: `skip-changeset`)
- [x] 6. `packages/spec` touched: its generated artifacts are
regenerated and committed. (not touched)
- [x] 7. No hand edits to `content/docs/references/`,
`packages/*/CHANGELOG.md` or `content/docs/releases/`.
- [x] 8. Title, body and comments are in English. (the 维护者速读 block and
the rule text itself are Chinese by the dispatch contract)
- [x] The *Before Submitting* list in CONTRIBUTING.md is done.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_01WGufzihBUhzn1w5tDRoFZT
<!-- signed: domain:skills#3 · os-support-ai ·
session_01WGufzhd2Ad9ANGTyAMW7pH -->
---------
Co-authored-by: Claude <noreply@anthropic.com>1 parent dbe2a93 commit eb833b5
1 file changed
Lines changed: 2 additions & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
31 | 31 | | |
32 | 32 | | |
33 | 33 | | |
| 34 | + | |
| 35 | + | |
34 | 36 | | |
35 | 37 | | |
36 | 38 | | |
| |||
0 commit comments