skills(pm-dispatch): dedupe at grading by file/mechanism over the open board of both repos — same finding ⇒ `duplicate_of`, never a second dispatch by os-elon-musk · Pull Request #18981 · objectstack-ai/objectstack

os-elon-musk · 2026-09-18T09:30:04Z

Fixes #18963
Clause-②: no

Governed rules layer (.claude/skills/pm-dispatch/SKILL.md) — draft PR, four-piece terminal; the skills seat's contract-tier review, then the maintainer's approval lands it. skip-changeset: nothing published moves (.claude/** is internal agent tooling, in no package's files[]).

What changed — three rows of the 分诊 band, 812 / 812 held, no line added, no re-wrap

The triage band's shadow rows read as a cross-repo keyword search only, and the dedupe step that would have caught two cards on one file and one defect — graded two days apart into two lanes by the same seat, both dispatched, the second PR superseded — did not exist in the rules text: a dedupe by FILE / MECHANISM against the OPEN board, own repo included, before keywords, with the finding (not the card) as the unit. Governing triage reading, verbatim: 「缺的不是勤勉,是一条按『文件/机制』而不是按『关键词』去重的步骤」 · 「去重的单位是『发现』,⛔ 不是『卡』」.

line	before (bytes)	after (bytes)
:366	「跨仓查重(shadow 检查):跟跨仓引用 + 关键词搜各姊妹仓。」 (75 B)	「查重/shadow 检查先按文件/机制查本仓与姊妹仓 open 卡(含 `pm:dispatched`),再跟引用与关键词。」 (120 B)
:367	「shadow 命中在飞 ⇒ `Blocked-by:` 不派;open 未认领 ⇒ 先收敛成一个派发入口。」 (99 B)	「同文件同缺陷 = 同一发现,不分车道:证据搬到先卡,后卡关 `duplicate_of`,⛔ 不并排派发。」 (115 B)
:368	「shadow 命中已完成 ⇒ 卡可能过期。」 (47 B)	「其余在飞 ⇒ `Blocked-by:` 不派;open 未认领 ⇒ 先并成一个派发入口;已完成 ⇒ 卡可能过期。」 (119 B)

Band total 221 B → 354 B (+133 B) at +0 lines; every row is at or under the 120-byte cap (120 / 115 / 119). Lines :223 (「分诊座位唯一生产:定级/路由/type/查重/shadow/duplicate_of…」) and every line outside :366–:368 are byte-identical to origin/main — the diff is 3 insertions / 3 deletions on one file. The parallel-draft bands of PR #18903 (:509–:670) and PR #18921 (:162–:183) are untouched.

What the new rows say (measured against the card's requirements)

By file / mechanism, before keywords, over the OPEN board of both repos, pm:dispatched included — :366: 「先按文件/机制查本仓与姊妹仓 open 卡(含 pm:dispatched),再跟引用与关键词」. The population is stated once, in the same row (the PM's mechanism assumption 2, confirmed: 「跨仓」 became 「本仓与姊妹仓」 in place).
The unit is the finding, not the card — :367: 「同文件同缺陷 = 同一发现,不分车道」; the later card's evidence moves to the earlier card (「证据搬到先卡」), the later card shuts as duplicate_of (「后卡关 duplicate_of」, the state the :136 row already binds to the duplicate reason), and it is never dispatched beside the earlier one (「⛔ 不并排派发」).
Every hit that is NOT the same finding keeps the old trichotomy — :368: in flight ⇒ Blocked-by:, not dispatched; open and unclaimed ⇒ fold into ONE dispatch entry first; completed ⇒ the card may be stale. 「其余」 scopes the row to what :367 did not already settle, so a same-finding sibling in flight lands on duplicate_of, not on Blocked-by:.

Retired clauses and their survivors (nothing retired without one)

retired spelling	survivor
:366 「跨仓查重(shadow 检查)」	:366 「查重/shadow 检查 … 本仓与姊妹仓」 — same name (the :310 spelling 「查重/shadow 检查」), population widened to both repos
:366 「跟跨仓引用 + 关键词搜各姊妹仓」	:366 「再跟引用与关键词」 — same two methods, now second to file / mechanism; 「各姊妹仓」 is carried by 「本仓与姊妹仓」 earlier in the row
:367 「shadow 命中在飞 ⇒ `Blocked-by:` 不派」	:368 「其余在飞 ⇒ `Blocked-by:` 不派」 — same mechanism, scoped to hits that are not the same finding
:367 「open 未认领 ⇒ 先收敛成一个派发入口」	:368 「open 未认领 ⇒ 先并成一个派发入口」 — 「收敛成」 → 「并成」 (3 B) is the only compression; 「先」, 「open 未认领」 and 「一个派发入口」 all kept
:368 「shadow 命中已完成 ⇒ 卡可能过期」	:368 「已完成 ⇒ 卡可能过期」 — folded into the trichotomy row

The 「shadow 命中」 subject of the old :367/:368 is now carried by :366 naming the check and :368's 「其余」 reading against :367; no row outside the band was touched to make that binding.

references/core-rules.md (151 / 151): its only shadow mention is :56 「跨仓查重与 shadow 检查恒归中央 ⛔ 不下放」 — a rule about WHO runs the check (central triage), not about its population or order, and it stays true under the new rows (:55 already makes same-repo dedupe central too). No rule moved there, so no twin is owed (the PM's mechanism assumption 3, confirmed by git grep -i shadow on that file: 1 hit, :56).

Reader test — the second grading, replayed under the new rows

At 2026-09-18T00:06Z the triage seat grades #18844 (check-single-claim-paths.mjs 401s because node fetch ignores HTTPS_PROXY; remedy: the proxy re-exec the sibling scripts already carry). Platform state of #18314 at that instant, from its label events: open, pm:queue + domain:spec + priority:p2 since 2026-09-17T10:10Z, unclaimed (pm:dispatched and the assignee arrived at 05:49Z, five hours later — so at grading it was a queued sibling, not yet an in-flight one; the dispatch text's 「pm:dispatched when #18844 was graded」 describes the later claim-time instant, which is the seat's own half, #18964).

:366 — search the open board of this repo by FILE: check-single-claim-paths.mjs ⇒ [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314's title carries the file name verbatim; it is on the open board (pm:queue; had it already been pm:dispatched, 「含 pm:dispatched」 keeps it in the population).
:367 — same file, same defect (fetch vs HTTPS_PROXY, the re-exec remedy), a different lane (domain:spec vs the skills seat's filing) ⇒ 「同一发现,不分车道」 ⇒ [finding] scripts/check-single-claim-paths.mjs reads GitHub through node's global fetch with no proxy re-exec, so with real PR context it exits 1 on GitHub API 401 inside an agent container — the gate can be read only in CI, while its sibling post-stamped already re-execs with --use-env-proxy #18844's evidence (the seven-file census) moves onto [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314; [finding] scripts/check-single-claim-paths.mjs reads GitHub through node's global fetch with no proxy re-exec, so with real PR context it exits 1 on GitHub API 401 inside an agent container — the gate can be read only in CI, while its sibling post-stamped already re-execs with --use-env-proxy #18844 shuts as duplicate_of [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314; ⛔ no second dispatch, no PR fix(scripts): re-exec check:single-claim-paths through the session proxy so its verdict is readable outside CI #18945.
:368 — not reached: the hit was settled as the same finding, so neither Blocked-by: nor 「先并成一个派发入口」 applies.

Outcome: one card (#18314, p2, domain:spec), one dispatch, one PR. The old :366 could not reach this: its population was 「各姊妹仓」 only, and its method was keywords.

Gates — all at `ea2521bd9` (the final commit; the tree did not move after these runs)

Derived from the worktree's own changeset with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (stderr: 「gate list derived from the tree of 'objectstack-ai/objectstack' at commit ea2521b」; change set: 1 path vs merge base 26c73fb4e): 20 commands, 20 run, every exit code captured by redirect-then-$?, never through a pipe. --ran reconciliation: 「20 derived famil(ies) accounted for — 20 run, 0 NOT-MEASURED, 0 UNRUN」 (exit 0).

command	exit	verdict line
`pnpm check:pm-skill-ratchet`	0	✓ `.claude/skills/pm-dispatch/SKILL.md is 812 lines (ceiling 812; headroom 0)` · ✓ `references/core-rules.md is 151 lines (ceiling 151; headroom 0)` · widest table row 342 (pin 342)
`pnpm check:pm-skill-id-lint`	0	✓ 27 file(s) clean (pattern `/#[0-9]{3,}/g`)
`pnpm check:skill-frame-sync`	0	✓ the one declared copy is internally coherent; 4 axes; 74 markdown files scanned — four-axis block :733–:754 md5 `3327d02c56f8a0eca88569dad2270f32` before and after
`pnpm check:pm-governed-prose`	0	✓ 2 instruction surface(s) name all 5 registered governed surfaces and claim no others
`pnpm check:nul-bytes`	0	OK (8891 text files, no raw ASCII control bytes); control-byte grep of the edited file: 0 hits
`pnpm --filter @objectstack/lint run check:doc-formula-expressions`	0	✓ self-test 58 cases · 22 record-scoped examples across 438 files / 1378 TS blocks clean · 9 TSDoc examples clean · 14 `*When` predicates clean — after `pnpm --filter "@objectstack/lint..." build` under `scripts/pm/os-verify-lock.sh` (VERDICT command-exit 0, 30 s held; first attempt without the lint build exited 3 PREREQUISITE NOT MET, which is a refusal, not a measurement)
`pnpm check:agent-test-spelling`	0	✓ 0 violations — 515 files
`pnpm check:cross-package-test-inputs`	0	OK: 29 package(s) read outside themselves, all declared; 255 self-test cases
`pnpm check:doc-authoring`	0	✓ 821 pinned sites across 231 files hold the baseline
`pnpm check:driver-memory-census`	0	OK — every declaration ledgered, every entry live
`pnpm check:pm-expected-skips`	0	✓ self-test 99 cases
`pnpm check:pm-governed-merges`	0	✓ 328 assertions
`pnpm check:pm-half-states`	0	✓ self-test 4963 cases
`pnpm check:refd-timer-probe`	0	✓ 11 cases, negative controls included
`pnpm check:watch-hint-literal`	0	✓ 71 declarations across 4 rostered names
`node scripts/check-closing-keyword-parity.mjs`	0	OK (3 parsers agree on all 9 keywords; 5 files carrying the grammar, all registered)
`node scripts/check-closing-keyword-parity.mjs --self-test`	0	✓ 40 assertions, 5 mutations driven to red
`node scripts/check-comment-mask-corpus.mjs`	0	✓ 6857 files, 0 disagree, 0 unparseable
`node scripts/pm/check-governed-queue-guard.mjs --self-test`	0	✓ 253 cases
`node scripts/pm/check-harness-current.mjs --self-test`	0	all 26 cases passed

Builds were run with OS_SKIP_DTS=1 (the formula gate imports compiled JS, not declarations); no gate that reads .d.ts is in this derivation. Repo-wide scans (pnpm lint) are CI's run, not owed here — this diff touches no lintable source.

Line budget

SKILL.md: 812 before / 812 after / ceiling 812 — paid in-band by rewriting the three rows (:366 75 → 120 B, :367 99 → 115 B, :368 47 → 119 B); the one compression is 「收敛成」 → 「并成」 in the :368 survivor.
references/core-rules.md: 151 / 151 / 151 — untouched, no twin owed.

Acceptance notes

origin/main moved from 26c73fb4e (BASE) to 873e0e8e2 while this ran; git log BASE..origin/main -- .claude/skills/pm-dispatch/ is empty, and neither PR skills(pm-dispatch): key the clause-② contract review by lane — spec and skills owe it on every round, other lanes owe none #18903 nor PR skills(pm-dispatch): a passing control certifies the instrument, not the question — a zero-hit reading names the instrument's reach and one known target outside it #18921 has landed, so no merge of origin/main was performed — the branch is one commit on BASE and its three-dot diff against main is exactly the table above. Declared here rather than done: a merge commit would carry no content for this file.
noted, not filed: the seat's own half (the claim-time stem scan in references/dispatch-runbook.md) is finding(skills): the seat's claim-time serial check reads open PRs' file lists only, so a sibling card already claimed and pushed but not yet opened as a PR is invisible to it — one git ls-remote --heads origin | grep <file stem> would have shown #18314's branch before #18844 was claimed #18964 and was not touched; the runbook carries no spelling of the shadow check to drift against these rows (git grep -i shadow on references/: only core-rules.md:56). 承接者: finding(skills): the seat's claim-time serial check reads open PRs' file lists only, so a sibling card already claimed and pushed but not yet opened as a PR is invisible to it — one git ls-remote --heads origin | grep <file stem> would have shown #18314's branch before #18844 was claimed #18964.
noted, not filed: check-closing-keyword-parity reports it skipped packages/spec/CHANGELOG.md (6,080,453 B over its 2 MiB cutoff for UNREGISTERED files) — the gate's declared, printed behaviour, not a defect; 承接者:无.
noted, not filed: dispatch-gates --ran accepts an exit code per recorded line and flags a bare record as 「CLAIMED」; the record here was re-run in that form (see the report). 承接者:无.

维护者速读(草稿)

改了什么 — 分诊规则里的三行「shadow 检查」被改写为一条真正的去重步骤:定级时先按文件/机制、再按关键词,对本仓与姊妹仓全部 open 卡(含已派发的)查重;同文件同缺陷视为同一个发现,后卡的证据搬到先卡、后卡以 duplicate_of 关闭,不再并排派发。行数 812/812 不变,只在原三行内改写,没有折行凑数。

为什么改 — 同一个文件、同一个缺陷的两张卡,被同一个分诊席在两天里定进两个车道,两边都派了 dev,第二个 PR 作废:浪费了一轮 dev。原规则只说「跨仓 + 关键词」,本仓已有的卡根本不在检查范围里。

风险与代价(含回滚) — 纯规则文本,不碰代码与发布包;风险是分诊多做一次按文件名的板面检索(成本很小)。回滚 = revert 这一个 commit,三行恢复原文。

席位意见 — (留空,席位定稿成评论)

你要做的 — 一个动作:APPROVE(受管面,需你的批准后由席位落地)。

Generated by Claude Code

…n board of both repos before keywords; same finding ⇒ duplicate_of, never a second dispatch The triage band's shadow rows (:366–:368) read as a cross-repo keyword search only. Two cards on one file and one defect were graded two days apart into two lanes by the same seat, both dispatched, and the second PR closed as superseded — the step that was missing is dedupe by FILE / MECHANISM against the OPEN board (own repo and sibling repos, `pm:dispatched` included) before keywords, with the finding — not the card — as the unit: the later card's evidence moves to the earlier card and the later card closes `duplicate_of`. Paid in-band at 812 / 812: the three rows are rewritten, no line added, no re-wrap; every retired clause survives in the same band (see the PR body). Claude-Session: https://claude.ai/code/session_01BTeBejoPUvRHN8WdAJC6oF Co-authored-by: Claude <noreply@anthropic.com>

os-elon-musk · 2026-09-18T09:36:17Z

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: ea2521bd92e63f3542333c5f5546084dfa51026b

① Derived judgments

Surface: .claude/skills/pm-dispatch/SKILL.md (+3 / −3, 812 / 812), one commit, the 分诊 shadow rows :366–:368 only; everything outside the band byte-identical, the four-axis block md5 unchanged. Governed rules layer (check-governed-merges.mjs --test: GOVERNED). No published package moves; no contract's accept/reject set moves. Hunks at :363–:371 — disjoint from PR skills(pm-dispatch): key the clause-② contract review by lane — spec and skills owe it on every round, other lanes owe none #18903 (:509–:670) and PR skills(pm-dispatch): a passing control certifies the instrument, not the question — a zero-hit reading names the instrument's reach and one known target outside it #18921 (:162–:183); no landing on .claude/skills/pm-dispatch since the claim; origin/main moved on unrelated paths only (the dev declared no merge; the queue rebuild validates).
The three rows, read on the head with bytes: :366 「查重/shadow 检查先按文件/机制查本仓与姊妹仓 open 卡(含 pm:dispatched),再跟引用与关键词。」 (120 B, at the cap) — the population (both repos' open cards, dispatched included) and the order (file / mechanism first, references and keywords second); :367 「同文件同缺陷 = 同一发现,不分车道:证据搬到先卡,后卡关 duplicate_of,⛔ 不并排派发。」 (115 B) — the unit is the finding regardless of lane, the action is evidence-to-the-earlier-card and duplicate_of on the later, never a second dispatch — exactly the two readings the triage 5727155157 asked the taker to carry; :368 「其余在飞 ⇒ Blocked-by: 不派;open 未认领 ⇒ 先并成一个派发入口;已完成 ⇒ 卡可能过期。」 (119 B) — the old trichotomy kept for hits that are not the same finding. Payment in-band, no line added: old :366's 「跟跨仓引用 + 关键词搜各姊妹仓」 survives as 「本仓与姊妹仓 … 再跟引用与关键词」; old :367 and :368 fold into the new :368 with one compression (收敛成 → 并成); the seat finds no clause without a survivor. core-rules :55–:56 (dedupe is the central triage's production; cross-repo dedupe and shadow stay central) bind WHO runs the check, not its population — no twin owed; correct.
Reader test (the dev's, re-read by the seat against [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314's label events): at the 2026-09-18T00:06Z grading of [finding] scripts/check-single-claim-paths.mjs reads GitHub through node's global fetch with no proxy re-exec, so with real PR context it exits 1 on GitHub API 401 inside an agent container — the gate can be read only in CI, while its sibling post-stamped already re-execs with --use-env-proxy #18844, :366 searches the open board by file and lands on [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314 (title carries the file name; pm:queue then, pm:dispatched from 05:49Z — both in the population), :367 makes it the same finding across lanes ⇒ duplicate_of, no second dispatch. One card, one dispatch, one PR.
Seat correction recorded: the dispatch text said [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314 was pm:dispatched when [finding] scripts/check-single-claim-paths.mjs reads GitHub through node's global fetch with no proxy re-exec, so with real PR context it exits 1 on GitHub API 401 inside an agent container — the gate can be read only in CI, while its sibling post-stamped already re-execs with --use-env-proxy #18844 was graded; [finding] scripts/check-single-claim-paths.mjs cannot be run from an agent container — it 401s because node's fetch ignores HTTPS_PROXY, while three sibling scripts already carry the re-exec fix #18314's events read pm:queue at 00:06Z and pm:dispatched at 05:49Z — the card's own timeline table had it right, the dispatch narrative did not. The landed :366 covers both states, so no text moved on it.
Seat spot-check on a scratch worktree at this head: check:pm-skill-ratchet exit 0 (812 / 812), check:pm-skill-id-lint exit 0 (27 files clean), check:skill-frame-sync exit 0, check:pm-governed-prose exit 0; lines over 120 B: 23, as on main (the pinned table rows).

② Semver level

skip-changeset is correct: .claude/** ships in no package's files[]. Clause-②: no on the claim (5727820576) and in the body; PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 18981 at 2026-09-18T09:36Z: exit 0, one live claim, no widening tell.

③ Boundary flags

Dev flags: no open questions; four noted, not filed items (core-rules :56 unaffected; finding(skills): the seat's claim-time serial check reads open PRs' file lists only, so a sibling card already claimed and pushed but not yet opened as a PR is invisible to it — one git ls-remote --heads origin | grep <file stem> would have shown #18314's branch before #18844 was claimed #18964 untouched; the closing-keyword gate's declared CHANGELOG skip; the dispatch's timeline detail — recorded above) — agreed, nothing owed. Deviations accepted, all declared: no origin/main merge (the merge condition did not fire); OS_SKIP_DTS=1 builds under the lock; two pushes (probe + commit); the --ran record re-written with exit codes. Gates: 20 derived / 20 run / 0 NOT-MEASURED at this head (check:doc-formula-expressions exit 3 first as PREREQUISITE NOT MET, then 0 after the lint build under the verify lock). 0 MCP calls; 5 REST / git writes.
CI on this head at 2026-09-18T09:36Z: 13 success · 12 skipped · 4 in_progress — nothing red; the landing act reads it again.

Implemented-by: claude/issue-18963-dedupe-by-file-at-grading
Reviewed-by: session_01BTeBejoPUvRHN8WdAJC6oF

VERDICT: PASS

Generated by Claude Code

os-elon-musk · 2026-09-18T09:36:20Z

维护者速读(终稿)— PR #18981 · 席位定稿 2026-09-18T09:36Z(草稿在正文,以本评论为准)

改了什么:分诊规则里的三行「shadow 检查」改写成一条真正的查重步骤:定级时先按文件/机制查本仓与姊妹仓的全部 open 卡(含已派发的),再看引用与关键词;同文件同缺陷视为同一个发现、不分车道——证据搬到先卡,后卡以 duplicate_of 关闭,不再并排派发;其余命中沿用原来的三分法(在飞不派、未认领先并、已完成可能过期)。行数不变(812/812),只在原三行内改写,没有折行凑数。

为什么改:今天早上同一个文件、同一个缺陷的两张卡(#18314 与 #18844)被同一个分诊席隔两天定进两条车道,两边都派了 dev,第二个 PR 作废——浪费一轮 dev。原规则只说「跨仓 + 关键词」,本仓已有的卡根本不在检查范围里。分诊席自己在 #18963 上认了这一步是缺的。

风险与代价(含回滚):纯规则文本,不碰代码与发布物;代价是分诊每张卡多做一次按文件名的板面检索。与在等你批准的 PR #18903、#18921 同文件不同行段,谁后落谁先合 main。回滚 = revert 一个提交。

席位意见:建议批准。三行各在 120 B 之内,退掉的每个子句都有去处;「不分车道」四个字是关键——正是它让两条车道各自的分诊看见同一张卡。席位这边的另一半(认领前扫远端分支)是 #18964,另走 runbook。

你要做的:确认这条查重规则该进分诊纪律,是则 APPROVE(GOVERNED_APPROVERS 账户);之后由本席按裁决 C 落地。

Generated by Claude Code

os-elon-musk · 2026-09-18T11:27:18Z

Landing record — ruling C, on an authorized APPROVED (skills seat, session_01BTeBejoPUvRHN8WdAJC6oF) · 2026-09-18T11:27Z

os-zhuang (GOVERNED_APPROVERS) approved this PR (review 5247231818 at 2026-09-18T11:25Z), marked it ready and added it to the merge queue in the same minute — the two CCR calls this seat's channel could not make, made by the approver's own hand. The seat's pre-landing reads on head ea2521bd92: the review of record 5728142814 names this head; PM_SWEEP_REPO=objectstack-ai/objectstack node scripts/pm/check-clause2-carriers.mjs --pair 18981 exit 0 (card #18963); checks at 2026-09-18T11:25Z: 18 success · 16 skipped, none red. SKILL.md 812 / 812, hunks :363–:371, disjoint from PR #18903 and PR #18921 (the latter now queued ahead of it) — the queue rebuilds on the current main. needs-user-decision cleared by the seat now (the decision it marked has been made). The merge is the queue's; #18963 closes on it, and the seat clears its pm:dispatched / assignee residue after the merge lands.

Generated by Claude Code

… tool corpus half and charter :339 in one batch (objectstack-ai#19021) Fixes objectstack-ai#18993 Clause-②: no `scripts/pm/check-prior-rulings.mjs` answered the `Prior rulings read:` line out of three tree corpora (`docs/adr/**`, `AGENTS.md`, `packages/spec/src/**`) while `SKILL.md:339` required it to cover 「已裁卡」 — a card's own thread, where the cheapest ruling lives, was in no corpus. Measured consequence on objectstack-ai#17518: a director-seat ruling (5651572469, 2026-09-13) was followed by a retriage that re-presented the card as unruled, a second maintainer ruling with a different prescription (5716259259, 2026-09-17), and a dispatch round that ended with zero diff. This PR makes the card's own thread a corpus half of the tool and rewrites the charter line in the same batch, so the requirement and the tool never diverge again. ## What changed **`scripts/pm/check-prior-rulings.mjs`** (report-only, no write path, unchanged exit register) - Every `--card` run now reads the card's comment thread over the board route the file already uses (its own proxy re-exec guard, taken once, never on `--self-test`), one paged `GET …/issues/{n}/comments?per_page=100&page=P` walked by page number to a short page, bounded at `MAX_THREAD_PAGES` (50), reconciled against the card's `comments` field. - A ruling comment is recognised by the **shape of its first content line**, three measured shapes: the charter's `Ruling: batch #n item k …` (`ruling-line`), the pre-charter `## Ruling recorded — …` heading (`ruling-recorded`, the shape 5651572469 is written in), and `Ruling addendum …` (`ruling-addendum`). Every hit prints its id, author, timestamp and shape. - Two measured near-misses are pinned OUT and listed uncounted (`**Ruling C landing step** — …`, `## Ruling A's patch is falsified …`). - The paste line gains a `thread:` clause with three spellings kept apart: `thread: 2 ruling(s) (id, id)`; `thread: none` (read, no ruling); `thread: not read (no --card)` (a `--terms`-only run). A thread that could not be read is the WHOLE line `unresolved`, exit 3, exactly as an unreadable corpus — never `none`. - The report prints the read's cost (pages, bytes, ms) and, when a ruling stands, the 〈升级与决策〉③ warning; with two or more rulings it says to read every one and to return a disagreement to the director seat. - Self-test: 99 → 144 cases (the three shapes, the near-misses, first-content-line anchoring, the fixture thread with two rulings oldest-first, `none` / `not read` / `unresolved`, and an offline paged board: two pages, one page, a 403, no token, the page bound, a network failure, a malformed page; plus a structural pin that no `Link` header is followed). `check-half-states.mjs`'s source-text pin on the line's writer (H62) still holds: 4963 cases pass. **`.claude/skills/pm-dispatch/SKILL.md` :339** — rewritten in place, net 0, 120 bytes → 120 bytes, 812 / 812 lines before and after: > before: `- 方向落在 ADR/已裁卡/不可重裁规则:贴 `check-prior-rulings.mjs` 的 `Prior rulings read:` 行,或不点。` > after: `- 方向在 ADR/本卡裁决/不可重裁规则:贴 `check-prior-rulings.mjs` 的 `Prior rulings read:` 行,或不点。` The three places a direction may already be ruled now read exactly as the tool's line covers them: an ADR, a ruling on this card, a non-re-rulable rule. **`.claude/skills/pm-dispatch/references/decision-analysis.md` :45** — the human-facing statement of the same line shape (the tool's `formatPasteLine` docblock names it as the twin), rewritten in place, 113 → 120 bytes, 50 / 50 lines: ``` before: - 四棱后一行:`Prior rulings read: TERMS → N hits; ADR-xxxx Dn|none`;未搜成整行写 unresolved。 after: - 四棱后:`Prior rulings read: TERMS → N hits; ADR Dn|none; thread: K|none`;未搜成整行写 unresolved。 ``` (The file spells the placeholders in angle brackets; they are written as capitals here because the platform mutates angle-bracket fragments in a body.) This file was not in the dispatch's named surface; it is included because it states the line's shape and would otherwise be the next 「a requirement and a tool that cannot answer it」 pair. No open PR touches it (file lists of the 14 skills-lane PRs read 2026-09-18T11:3xZ). ## Hypotheses measured 1. **Ruling shape is stable; author is not.** Census over 1,100 comments (REST, reads only): 2026-09-11T06Z–09-13T21Z (600): `## Ruling recorded — …` 22 (os-tesla 19, claude[bot] 3), `Ruling: …` 4 (os-tesla 2, claude[bot] 2); 2026-09-18T04Z–11Z (500): `Ruling: …` 26 (hotlong 26), `Ruling addendum` 2 (hotlong). The charter line (`director.md`: 「裁决评论首行固定一条:`Ruling: batch #N item K · letter X · maintainer 「VERBATIM」 UTC`」 — placeholders capitalised here, the file spells them in angle brackets) landed 32a3214 2026-09-13T11:27Z; the heading shape is the same seat's earlier spelling. The fallback the dispatch named — 「every comment by a `GOVERNED_APPROVERS` login」 — is falsified: it would have missed every 2026-09-13 ruling, 5651572469 included. A quoted 「同意」/「agreed」 predicate hits 63 of 500 recent comments (claims, ledgers, dev reports) and names nothing. 2. **Cost, the tool's own meter** (node `fetch` on the re-exec'd route, pages serial, 2026-09-18T11:3xZ): | thread | comments | pages | bytes | thread read | whole run | rulings / near-misses | |---|---:|---:|---:|---:|---:|---| | objectstack-ai#17518 (the incident) | 17 | 1 | 141,623 | 347–369 ms | 2,353 ms (before: 2,001 ms) | 2 / 0 | | objectstack-ai#12708 (director seat post) | 372 | 4 | 2,212,674 | 1,984 ms | 3,669 ms | 0 / 0 | | objectstack-ai#6015 (longest thread on the board) | 962 | 10 | 7,365,556 | 4,431 ms | 6,254 ms | 0 / 0 | `curl` on the same pages reads ~10% more bytes (156,874 on objectstack-ai#17518): GitHub pretty-prints for a `curl/*` User-Agent and minifies for any other, measured directly; the tool reports what its own route pays. The fixed cost per dedupe on the incident card is one page and about a third of a second. 3. **The line, before / after, on `--card 17518` through the proxy** (exit 0 both): ``` before: Prior rulings read: spec,assembledpackagebodyschema,declares,callable,custom,branches,inert-json,artifact,hold,schema,embedding,loses (+1 more) → 117 hits; ADR-0087 D4, ADR-0130 D4, ADR-0029 D6, ADR-0039 D7, ADR-0056 D5, ADR-0076 D3, ADR-0076 D9, ADR-0087 D1, ADR-0087 D3, ADR-0087 D7 after: Prior rulings read: spec,assembledpackagebodyschema,declares,callable,custom,branches,inert-json,artifact,hold,schema,embedding,loses (+1 more) → 117 hits; ADR-0087 D4, ADR-0130 D4, ADR-0029 D6, ADR-0039 D7, ADR-0056 D5, ADR-0076 D3, ADR-0076 D9, ADR-0087 D1, ADR-0087 D3, ADR-0087 D7; thread: 2 ruling(s) (5651572469, 5716259259) ``` and the report's thread block: ``` the card's own comment thread (ruling comments, by first-line shape): 2 ruling(s) of 17 comment(s) — read in 1 page(s), 141,623 bytes, 347 ms 5651572469 os-tesla 2026-09-13T06:10:22Z [ruling-recorded] Ruling recorded — B: `packages/spec` declares the inert-JSON artifact stage; the read A… 5716259259 hotlong 2026-09-17T14:41:18Z [ruling-line] Ruling: batch objectstack-ai#149 item 1 · letter B (artifact-stage JSON-only variants of `FlowFunctio… ``` 4. `--terms single,posture,tenant` (no card): exit 0, line ends `; thread: not read (no --card)`. ## Gates Derived with `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` from the worktree (three-dot change set: the three files above): 40 families, every one run, exit codes captured redirect-then-`$?`, reconciled with `--ran` (`40 derived, 40 run` after the one PREREQUISITE NOT MET was cleared by its declared build: `check:doc-formula-expressions` exited 3 for an unbuilt `@objectstack/formula`; built under the verify lock, re-run, exit 0). `check:pm-dispatch-gates` exceeds the foreground cap and ran detached (`tail --pid`): 779 s, `1849 cases pass`, exit 0. Also: `node scripts/pm/check-prior-rulings.mjs --self-test` 144 pass; `check:pm-skill-ratchet` ✓ (812 / 812, 50 / 50, widest rows unchanged); `check:pm-skill-id-lint` ✓ 27 files clean; `check:nul-bytes` ✓; `check-governed-merges.mjs --test` over the three paths: GOVERNED (2 of 3 on the register — expected, `.claude/**`); eslint on the edited script: 1 file, 0 errors, 0 warnings (`--no-inline-config`, the root flat config's `**/*.mjs` block; no type-aware rules, so the diff moves no untouched file's verdict). Readings taken at `cba931995`, the branch head. The same derivation from a throwaway worktree at `origin/main` 64de4c5 (five commits past this branch's base, four of them touching family-defining files) yields the identical 40 families — none new, none gone. This branch is deliberately not merged with `main` here: the incoming range carries four `merge=os-regen` artifacts whose driver debt would hold the push on a PR that touches no spec file; the seat's `update-branch` is the right instrument if a refresh is wanted before landing. No published package moves (`scripts/pm/**` and `.claude/**` ship in no `files[]`): `skip-changeset`. ## Acceptance notes - noted, not filed: the file's self-test keeps a flat case list with the verdict handshake and no battery-name floor (the AGENTS.md `--self-test` section's floor rule is stated for writing a new one); the 2026-09 self-test shape census owns that class — carrier: whoever works that census; not this card. - noted, not filed: `readCard` still uses the global `fetch` while `readThread` takes an injectable one; symmetry only, no behaviour — carrier: none. - Boundary kept: no merge with objectstack-ai#18963 / PR objectstack-ai#18981 (grading-time dedupe by file) — same class, different mechanism; objectstack-ai#18963 is not addressed here. ## 维护者速读(草稿) **改了什么**:查重工具 `check-prior-rulings.mjs` 多读一处 —— 被查那张卡自己评论线上的裁决评论;贴进卡面的那一行 `Prior rulings read:` 多出一段 `thread: …`,写明本卡已有几条裁决、评论 id 是哪几条;读不到线程时整行仍写 `unresolved`,绝不写成「没有」。章程 `SKILL.md:339` 与 `decision-analysis.md:45` 各原位改一行,行数不变,同一批落地。 **为什么改**:objectstack-ai#17518 上四天内出现两条互相矛盾的维护者裁决,一整轮派发白跑 —— 原因是重新分诊的席位没看到卡上四天前的那条裁决,而工具即使跑了也查不到:卡的评论线不在它任何语料里。这次把「已裁卡」这半边做成机械面,并把章程那句话改成工具真能答的范围。 **风险与代价(含回滚)**:每次查重固定多一次线程读取 —— 事故卡 1 页、约 0.35 秒;全板最长线程(962 条评论)10 页、约 4.4 秒,已实测。识别靠裁决评论首行的固定写法(三种实测形态),不靠作者身份;写法漂移会以「近似项」在报告里显形而不是静默漏掉。回滚 = revert 本 PR 三个文件,无发布物、无数据迁移。 **席位意见**:(留空) **你要做的**:确认三种裁决写法的识别边界可接受;若同意,按受管面流程合并。 --- _Generated by [Claude Code](https://claude.ai/code/session_01BTeBejoPUvRHN8WdAJC6oF)_ --------- Co-authored-by: Claude <noreply@anthropic.com>

github-actions Bot added size/xs documentation Improvements or additions to documentation labels Sep 18, 2026

os-elon-musk added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Sep 18, 2026 — with Claude

os-elon-musk added the needs-user-decision label Sep 18, 2026 — with Claude

os-elon-musk requested review from hotlong and os-zhuang September 18, 2026 09:36

hotlong mentioned this pull request Sep 18, 2026

[PM seat] director(项目总监) — 🟢 os-zhuang · 第 35 场 session_01VYToj6PQehTEKNrjGM9akg · 开轮 2026-10-06T14:15Z · 第 31–34 场台账在评论 #12708

Open

os-zhuang approved these changes Sep 18, 2026

View reviewed changes

os-zhuang marked this pull request as ready for review September 18, 2026 11:25

os-zhuang added this pull request to the merge queue Sep 18, 2026

os-elon-musk removed the needs-user-decision label Sep 18, 2026

os-elon-musk mentioned this pull request Sep 18, 2026

feat(pm): the Prior rulings read: line covers the card’s own thread — tool corpus half and charter :339 in one batch #19021

Merged

Merged via the queue into main with commit 9ee8e35 Sep 18, 2026
39 checks passed

os-zhuang deleted the claude/issue-18963-dedupe-by-file-at-grading branch September 18, 2026 12:02

os-elon-musk mentioned this pull request Sep 18, 2026

feat(pm): a PR over 5,000 changed lines lands only by a human merge — size predicate in check-governed-merges --test, the same reading in dispatch-gates, one rule line in SKILL.md and landing-operations #19033

Merged

Navigation Menu

Repository navigation

skills(pm-dispatch): dedupe at grading by file/mechanism over the open board of both repos — same finding ⇒ `duplicate_of`, never a second dispatch - #18981

skills(pm-dispatch): dedupe at grading by file/mechanism over the open board of both repos — same finding ⇒ `duplicate_of`, never a second dispatch#18981
os-zhuang merged 1 commit into
mainfrom
claude/issue-18963-dedupe-by-file-at-grading

os-elon-musk commented Sep 18, 2026

Uh oh!

os-elon-musk commented Sep 18, 2026

Uh oh!

os-elon-musk commented Sep 18, 2026

Uh oh!

os-elon-musk commented Sep 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

os-elon-musk commented Sep 18, 2026

What changed — three rows of the 分诊 band, 812 / 812 held, no line added, no re-wrap

What the new rows say (measured against the card's requirements)

Retired clauses and their survivors (nothing retired without one)

Reader test — the second grading, replayed under the new rows

Gates — all at ea2521bd9 (the final commit; the tree did not move after these runs)

Line budget

Acceptance notes

维护者速读(草稿)

Uh oh!

os-elon-musk commented Sep 18, 2026

Contract review

① Derived judgments

② Semver level

③ Boundary flags

Uh oh!

os-elon-musk commented Sep 18, 2026

维护者速读(终稿)— PR #18981 · 席位定稿 2026-09-18T09:36Z(草稿在正文,以本评论为准)

Uh oh!

os-elon-musk commented Sep 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Gates — all at `ea2521bd9` (the final commit; the tree did not move after these runs)