pm-dispatch: three rules-layer lines outside the charter-revision PR still carry pre-ruling text (os-dev.md 先搜再立, SKILL.md state-table rows, lanes/ui.md :25 + objectui AGENTS §9)

Filed by the charter-revision dev (session session_01DAcomhvR9kKizeYgg89Vo8, branch claude/issue-17942-charter-revision) as a bare card: ⛔ not routed, not graded — triage grades (the 2026-09-13 ruling on the chain head, comment 5652544529, moves grading, routing and dedupe to the triage seat; the filer attaches keywords). Derived in-scope from the charter-revision family whose chain head is issue 17942 — a candidate class-1 sub-issue for the seat that owns that family.

Dedupe keywords: 先搜再立, os-dev.md, 停在 draft 等人合, 跨车道移交, charter residue.

What

The rules-layer PR of the charter-revision family (#18018) lands the 2026-09-13 rulings in SKILL.md, core-rules.md, AGENTS.md and three lane files. Three lines OUTSIDE that PR's declared file surface still carry the pre-ruling text and now contradict the landed rules:

.claude/agents/os-dev.md (rule 3, the 「先搜再立」 line): the dev is still told to dedupe before filing (targeted REST or one search_issues). Ruling ③ (5652544529, point 3, verbatim 「立卡者不查重,只在卡面附 3–5 个查重词」) moves dedupe to the triage seat; the filer attaches keywords. os-dev.md has its own ratchet ceiling — pay by density there.
.claude/skills/pm-dispatch/SKILL.md 〈状态模型〉 table rows (widest-row pin 342 B): the assignee row still lists 「跨车道移交」 as a release cause and the pm:retriage row still lists 「跨域 PR 指定车道、改路由」 without the pre-dispatch qualifier. The bullet 「pm:retriage 改路由只对未派发卡」 landed by the PR governs; the rows lag it.
.claude/skills/pm-dispatch/references/lanes/ui.md :25 (「命中即整 PR 停在 draft 等人合,⛔ 不 ready、不入队、不 auto-merge、不自合、不留批准」) and objectui AGENTS.md §9's governed-surface paragraph: ruling C (issue 17971, comment 5652306063, verbatim 「C. approve 后不管后续改动都由席位落地:」) makes the owning seat land a governed PR once an authorized approval exists — objectui PRs included (the card names objectui 9374 / 9377). Both lines still read as unconditional draft-until-human-merge. Seam: the objectstack half lands here, the objectui half in objectui's own AGENTS.md (objectstack side leads per 〈多仓协调〉).

Executable criterion

Each of the three sites either states the ruled rule or points at the line that does, so git grep -n '先搜再立' .claude/agents/os-dev.md exits 1, git grep -n '跨车道移交' .claude/skills/pm-dispatch/SKILL.md exits 1 (or the row carries 未派发), and git grep -n '停在 draft 等人合' .claude/skills/pm-dispatch/references/lanes/ui.md exits 1 — with the ratchets green (pnpm check:pm-skill-ratchet) and no ceiling moved.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01DAcomhvR9kKizeYgg89Vo8 (GitHub os-project-manager, skills seat), claimed at 2026-09-13T15:33Z
Branch: claude/issue-18019-charter-residue-three-lines
Worktree: objectstack-issue-18019
Domain: domain:skills (graded p2 pm:queue by triage; class-1 derived of the #17942 chain; GOVERNED rules layer ⇒ four-piece at the end — ACCEPT on the card, draft kept, needs-user-decision on the PR, 速读 on the PR, reviews from os-zhuang + hotlong — then ruling C landing by this seat once an authorized approval exists)
File surface: .claude/agents/os-dev.md (rule 3, the 先搜再立 line → the filer attaches 3–5 dedupe keywords and never pulls the board; ruling ③ 5652544529 point 3 verbatim), .claude/skills/pm-dispatch/SKILL.md 〈状态模型〉 two table rows (the assignee row's 跨车道移交 release cause and the pm:retriage row's 跨域 PR 指定车道、改路由 both qualified to 未派发 or dropped, so the rows agree with the landed bullet 「pm:retriage 改路由只对未派发卡」; widest-row pin 342 B, 812/812), .claude/skills/pm-dispatch/references/lanes/ui.md :25 (ruling C wording — a governed hit stays draft until an authorized approval, then the owning seat lands it; 38/38). Every file is at its ceiling ⇒ net 0 lines per file, density payments inside the same file named in the commit, 120-byte line cap, frame block md5 3327d02c56f8a0eca88569dad2270f32 unchanged. references/core-rules.md only if a mirrored line exists (〈优先级〉: a rule changed in both places changes in the same PR — the dev measures). ⛔ objectui AGENTS.md §9 is a separate objectui card (this seat's round-end batch). ⛔ nothing else in .claude/**.
Container & model: M, mode:subagent, model: opus — tier exit recorded: dispatch-gates --tier mandates claude-fable-5-1 for SKILL.md; this card is a one-line-class mechanical governed edit (three lines lagging rules already landed by PR #18018, no new rule), so execution drops to opus with the skill-face review at claude-fable-5-1 by this seat — the ruling's own stated exit, judged from the card content.
Clause-②: no
Thread-read: the card shows 1 comment and the API lists 0 — the grading comment's author is the suspended account; the labels it wrote (priority:p2, pm:queue, domain:skills, documentation) stand and are the grading of record
Serial constraints cleared at 2026-09-13T15:33Z: the three files read FREE on the hot-file queue (#7623) and no claude/issue-18019-* / 17800-* / 17497-* branch exists; the dev re-reads open PR file lists before editing; H17 index names only dispatch-gates.mjs (#14290), not this face; verify lock free; rate ≥14.7k/15k; origin/main at 226970bb (carries PR #18018 9489e2c0 and PR #18038 c185d087, the rules these lines lag). Devs in flight after this one: 2 (the cap).

Generated by Claude Code

os-dev-report

{
  "issue": 18019,
  "status": "done",
  "branch": "claude/issue-18019-charter-residue-three-lines",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18051",
  "premise_still_valid": true,
  "summary": "Three governed lines that lagged the charter rulings now state them, one commit per file quoting its ruling verbatim. os-dev.md :50 drops the 先搜再立 board scan for 立卡者不查重 plus 3-5 keywords on the card (ruling 3). SKILL.md :106 drops 跨车道移交 from the release causes (three now) and qualifies 改路由 with 限未派发, adding 在飞卡跟到 MERGED; :113 adds 定车道与改路由限未派发卡 - both rows now agree with the landed bullet pm:retriage 改路由只对未派发卡 and with 认领即跟到 MERGED / 借复核不移卡. lanes/ui.md :25 states ruling C: a governed hit stays draft until an authorized approval exists, then the owning seat lands it, and the seat never approves. Every file was at its ceiling, so every edit is an equal-line replacement: net 0 lines in all three, no ceiling moved, no row wider than the 342 B pin. GOVERNED surface - the PR is DRAFT and stays draft; no ready flip, no queue, no auto-merge, no approval.",
  "tests": "dispatch-gates --commands --repo objectstack-ai/objectstack at HEAD ac73575c derived 18 families; all 18 run in the foreground with $? captured before any pipe (cmd > log 2>&1; EXIT=$?), all exit 0. dispatch-gates --ran with per-command exit codes reconciles: '18 derived famil(ies) accounted for - 18 run, 0 NOT-MEASURED (a DERIVED zero), 0 UNRUN'. check:doc-formula-expressions first answered exit 3 PREREQUISITE NOT MET (unbuilt @objectstack/lint - NOT MEASURED, not a red) and was re-run to exit 0 after 'turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2' via scripts/pm/os-verify-lock.sh (OS_VERIFY_LOCK_SLOT=issue-18019; its own line: 'VERDICT command-exit 0 - held the lock 158s - waited 0s'). Ratchet readings at HEAD: os-dev.md 403 lines (ceiling 403, headroom 0); SKILL.md 812 lines (ceiling 812, headroom 0) and widest table row 342 B (pin 342, headroom 0); lanes/ui.md 38 lines (ceiling 38, headroom 0); core-rules.md 151/151 untouched. SKILL.md frame block L733-754 md5 3327d02c56f8a0eca88569dad2270f32 byte-unchanged (measured after the edit). Line bytes: os-dev.md :50 117 -> 115 B, ui.md :25 117 -> 118 B (both under the 120 B cap); SKILL.md rows 216 -> 235 B and 232 -> 269 B, widest row still L244 at 342 B. Executable criterion, git grep -c over origin/main 226970bb and HEAD: 先搜再立 1->0, 跨车道移交 1->0, 停在 draft 等人合 1->0. Lit controls on neighbouring surviving phrases, same greps and same two refs: 'noted, not filed' 3->3, 离手恒走释放 1->1, 异议评论写明所求 1->1, 未命中的 PR 按同节走合并队列落地 1->1. Control-byte self-scan over the three files: grep -naP over the non-printable class returns exit 1 (no hits). No ablation is owed: this diff authors no gate and no runtime behaviour - the three grep pairs above are the whole falsifiable claim. Repo-wide pnpm lint is not owed and can yield no verdict on this diff: eslint --no-inline-config --format json over the three files reports 3 files / 0 errors / 3 warnings, each 'File ignored because no matching configuration was supplied' - the universe is read from eslint's own config, the count from its JSON output, and no type-aware linting reaches markdown, so no untouched file's verdict can move.",
  "files_changed": [
    ".claude/agents/os-dev.md (1 line, :50)",
    ".claude/skills/pm-dispatch/SKILL.md (2 lines, :106 and :113)",
    ".claude/skills/pm-dispatch/references/lanes/ui.md (1 line, :25)"
  ],
  "premise_checks": {
    "P1": "CONFIRMED - os-dev.md :50 read 先搜再立 on origin/main 226970bb",
    "P2": "CONFIRMED - SKILL.md :106 listed 跨车道移交 among the four release causes; :113 read 跨域 PR 指定车道、改路由 with no 未派发 qualifier. The bullets they lagged: :227 认领即跟到 MERGED ... 不移卡, :228 pm:retriage 改路由只对未派发卡, :231 已派发卡 不因触 spec 转席, and contract-review.md :25 借复核不移卡",
    "P3": "CONFIRMED - lanes/ui.md :25 read 命中即整 PR 停在 draft 等人合, ... 不留批准",
    "P4": "CONFIRMED - all 12 open PRs (18046 18041 18039 18035 18022 18016 18014 17982 17835 17811 17635 17076) fetched into refs/remotes/pr18019probe/* and diffed against their OWN merge-base; none touches the three files nor core-rules.md. 17835 is the only open PR whose two-dot diff vs origin/main names .claude/ paths, and that is main having moved ahead of its base, not the PR's diff (its merge-base diff is 20 files, zero under .claude/)",
    "P5": "MEASURED NO - references/core-rules.md mirrors none of the three pre-ruling lines. It already carries the POST-ruling statements: :51 立卡者只附查重词,分诊查重零命中须控制词, :8 受管面由维护者人合或授权批准后席位落地, :123 受管面 PR 留 draft 并向两个授权批准账户请审. So core-rules.md is not in this diff and the 〈优先级〉 same-PR rule is not engaged"
  },
  "deviations": [
    "维护者速读 shape: the dispatch asked for four sub-sections (改了什么 / 为什么改 / 风险与回滚 / 一句问); os-dev.md's standing clause fixes five (改了什么 / 为什么改 / 风险与代价(含回滚) / 席位意见 / 你要做的, 席位意见 left blank for the seat). Standard text wins over the dispatch word, so the PR carries the five-section form, which is a superset of what was asked.",
    "The lock-held build exceeded the harness's 120s foreground cap and was moved to the background by the host. It was not left there: the turn blocked in the foreground on tail --pid until the holder exited, and the verdict was read from the wrapper's own VERDICT command-exit 0 line, never from the background notification.",
    "The first labels POST answered HTTP 415 (no Content-Type). Retried with the header on the same additive endpoint, then read back: size/xs and skip-changeset both present, nothing stripped.",
    "This checkout is SHALLOW (git rev-parse --is-shallow-repository = true). No --is-ancestor negative reading was relied on anywhere; P4 used merge-base plus a diff against that base, and every grep reading is a positive count on a ref present in this checkout.",
    "Channel: card body and timeline read through the zero-quota single-card payload channel; the open-PR file surface through repo-scoped REST plus a local git fetch of the PR head refs. MCP was used for writes only."
  ],
  "open_questions": [
    {
      "question": "SKILL.md :106 keeps 改路由 as a release cause, qualified 限未派发. Under the landed bullets a card whose assignee is set is 已认领/在飞, and 可派发 is defined as 无 assignee - so the qualified cause can only fire on a card a seat self-claimed and has not yet dispatched. Is that the reading you want the row to carry?",
      "options": [
        "A. keep 改路由(限未派发) - the cause survives for the self-claimed-but-undispatched window, which is the only window the landed bullet leaves open",
        "B. drop 改路由 too, leaving two causes (前提证伪、弃飞无接管) - re-routing then lives only in the pm:retriage row",
        "C. keep 改路由 unqualified in this row and let the :113 row carry the whole qualifier"
      ],
      "recommendation": "A - it is the narrowest reading that leaves the row true, keeps the cause discoverable where a seat looks for it, and states the pre-dispatch limit at the place the mistake is actually made. B deletes a legitimate act; C leaves this row readable as a licence to re-route an in-flight card, which is the defect the card was filed about."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: .claude/agents/os-dev.md :53, :54 and :58 still describe the dedupe channel the filer no longer uses - the one targeted MCP search_issues fallback on 403, the control-word rule for a zero result, and 那一次查重 in the MCP-budget line. Same residue class as :50, but outside this card's declared one-line surface, and the file is at 403/403 so paying for a rewrite needs its own density budget. Carrier: the seat that owns the charter-revision family (#17942 chain).",
    "noted, not filed: lanes/ui.md :24 points at objectui AGENTS.md section 9 for the governed-surface list and prohibitions, and that paragraph still reads draft-until-human-merge. The claim already routes it to a separate objectui card, so it is recorded here as a seam, not a new finding. Carrier: the objectui half of this family."
  ],
  "mcp_calls": "2 - create_pull_request and this add_issue_comment; every read went through git, the zero-quota payload channel, or repo-scoped REST"
}

Generated by Claude Code

ACCEPT — skills seat (session session_01DAcomhvR9kKizeYgg89Vo8), 2026-09-13T15:55Z. The os-dev-report is read in full; PR #18051 head ac73575c reviewed in-seat at claude-fable-5-1: ## Contract review PASS with 速读终稿 on the PR. The one open question is answered there: option A (改路由 stays, qualified 限未派发). GOVERNED rules layer ⇒ four-piece: needs-user-decision on the PR, reviews requested from os-zhuang and hotlong, draft kept; ruling C landing by this seat once an authorized approval exists. Fixes #18019 closes this card on landing.

Generated by Claude Code

Landed — skills seat (session session_01DAcomhvR9kKizeYgg89Vo8), 2026-09-13T23:01Z. PR #18051 merged by the queue as 137eb00e7 (single parent fb29f62ce); two readings at 2026-09-13T23:01Z: the queue ref gh-readonly-queue/main/pr-18051-* is gone, and git log origin/main carries (#18051) with (#18073) as the lit control. Ruling C landing: approval os-zhuang 5192546519 on ac73575c, review of record 5654347747, step record 5656672056. The three charter residue lines now state the landed rules (os-dev.md :50 立卡者不查重; SKILL.md :106/:113 改路由限未派发; lanes/ui.md :25 ruling C). Residue (pm:dispatched, assignee) stripped in this pass and read back. The objectui half (AGENTS.md §9) is objectui#9418.

Generated by Claude Code

skills(pm-dispatch): charter references — clause-② applicability, the…

docs(pm,agents): three rules-layer lines catch up with the charter ru…

docs(pm): landing-operations names the two landing tiers and lanes/ui…

Navigation Menu

Repository navigation

pm-dispatch: three rules-layer lines outside the charter-revision PR still carry pre-ruling text (os-dev.md 先搜再立, SKILL.md state-table rows, lanes/ui.md :25 + objectui AGENTS §9) #18019

What

Executable criterion

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

pm-dispatch: three rules-layer lines outside the charter-revision PR still carry pre-ruling text (os-dev.md 先搜再立, SKILL.md state-table rows, lanes/ui.md :25 + objectui AGENTS §9) #18019

Description

What

Executable criterion

Activity

claude commented on Sep 13, 2026

os-project-manager commented on Sep 13, 2026

claude commented on Sep 13, 2026

claude commented on Sep 13, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions