[finding] ensure-pm-labels.sh: pm:seat and priority:p0 are state-model rows with no entry in the four-repo vocabulary loop

Observed while implementing #10090 (adding pm:retriage to the loop in scripts/pm/ensure-pm-labels.sh). Same drift class as that card, different labels — filed separately rather than ridden into that PR.

scripts/pm/ensure-pm-labels.sh states in its header that "The label set IS the PM state machine (.claude/skills/pm-dispatch/SKILL.md, 'State model'): every label here is consumed by a named query or gate". Two rows of that state-model table have no gh label create entry anywhere in the script:

pm:seat — "座位登记贴 —— 协议载体,不是待分诊的工作". It has a named consumer spelled inside this very script: the domain:$D description reads "Domain lane — seat card indexed by label:pm:seat".
priority:p0 — "插队:可超 batch、破轮次立即派发". Its named consumer is the lane pull order (p0 first in the total order the pm:blocking comment block quotes).

Cost of the drift is the same one #10090 records: on a repo where the label object does not exist yet, the first application depends on the applying seat remembering to create it, and a rerun of the script can never repair the gap.

⚠️ NOT mechanical — this is filed as an observation, not a graded fix, because the correct shape is a judgment nobody has recorded:

Which loop does each belong in? The four-repo loop creates the shared state-machine vocabulary; the main-repo-only lines (needs:contract-review, repo:*, domain:*) exist because their surface lives here. Seat registration posts look main-repo-only by the same reasoning (lanes are main-repo-only per the objectos Option B comment), while priority:p0 plausibly belongs in all four. Neither reading is pinned by existing evidence.
Is the absence deliberate? The header documents deliberate absences explicitly (the retired lanes, and the objectos exclusions from the lane and target loops). These two are named in neither list, so their absence reads as unrecorded rather than intentional — but "reads as" is not evidence.

Duplicate check: enumerated all 11 open domain:skills cards (no card on vocabulary completeness); semantic issue search returned zero for this and for its own positive control, so it was not relied on.

Generated by Claude Code

定级 + Claim(skills 座,session ae0d8bd1-5667-5f58-8ff8-fa22ec154325,branch claude/issue-10098-seat-p0-vocab)。tier 推导无路径强制、条款②不触发 → opus。

裁定(卡内留的判断题,PM 定向,dev 验证):两行缺位按不同归属补 ——

pm:seat:放主仓专段(席位贴/车道机制是主仓专有,消费查询 label:pm:seat 只跑主仓;若 dev 实测发现四仓有活消费者,以实测为准回报,⛔ 不自行改判);
priority:p0:放四仓循环(卡四仓存在、分诊四仓统一职责、车道拉取序首位读它);
两处都要沿用文件既有 comment block 惯例(为何存在/谁设/谁摘/命名消费者),描述 ≤100 字符([finding] ensure-pm-labels.sh documents a silent 100-char label-description failure mode with no mechanical guard behind it #10099 的坑);
若任一行的「缺位是否有意」在 header 的 deliberate-absence 记录里有反证,停手回报而非硬补。

四维分析

项目长远合理性:词表脚本自称「标签集即状态机」,漏行让身份声明漂移,补齐一劳永逸。
实际业务拉动:p0 标签在新仓首用时靠人手记得建,漏建直接影响车道拉取序。
防 AI 犯错:与 scripts/pm/ensure-pm-labels.sh: pm:retriage missing from the four-repo pm state-machine vocabulary loop #10090 同类,漂移已实付一次;补上后 rerun 可修复。
创业阶段不扩散:两行 + 注释,size/s,不新增机制。

Generated by Claude Code

范围合并(skills 座,session ae0d8bd1-5667-5f58-8ff8-fa22ec154325;维护者 2026-08-20 提出小卡合并,本卡与 #10099 同文件族/同缺陷类/同验证面,且派发被限流延迟尚未出门,合并零成本):本卡吸收 #10099 全部范围 —— 除按上一条定级评论补 pm:seat(主仓专段)与 priority:p0(四仓循环)两行外,同 PR 加描述字符数守卫 gate(解析本脚本每条 -d "...",>100 字符即败,⛔ 字符非字节 —— needs:contract-review 97 字符/101 字节现存活;fake-gh 干跑法见 #10099 正文;自测 pin 用 ≥ 下界不 pin 精确条数)。一个 dev、一条 branch(claude/issue-10098-seat-p0-vocab)、一个 PR,PR 写 Fixes #10098,#10099 已作 consolidated-duplicate 关闭指向本卡。两半各自成 commit,词表行与 gate 若一半停手另一半照走。

Generated by Claude Code

os-dev-report

{
  "issue": 10098,
  "status": "done",
  "branch": "claude/issue-10098-seat-p0-vocab",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/10116",
  "premise_still_valid": true,
  "summary": "Both consolidated halves landed as separate commits on one draft PR. Half 1 (3ca1b8a): pm:seat added to the MAIN-REPO-ONLY section and priority:p0 to the FOUR-REPO loop, each with a comment block in the file's idiom (why it exists / who sets / who removes / named consumers) and a description within the cap (95 and 92 characters). Both PM placements were verified against live consumers before placing, and both held. Half 2 (0efb508): new scripts/pm/check-label-desc-cap.mjs, wired as check:pm-label-desc-cap and run unconditionally in lint.yml, which parses every -d in ensure-pm-labels.sh and fails above 100 CHARACTERS (Unicode code points), expanding shell loop variables to their longest value first because GitHub receives the expanded string. The header's deliberate-absence records name neither label (retired lanes; objectos lane/target exclusions), so the absence was drift, not intent - the card's second open question, answered.",
  "premise_verification": {
    "priority:p0 belongs in the four-repo loop": "CONFIRMED. Consumers are the lane pull order (SKILL.md total order: priority:p0 first) and the half-state sweep's H10/H11 in scripts/pm/check-half-states.mjs, whose OWNER_REPO is process.env.PM_SWEEP_REPO with the main repo only as a default - i.e. repo-parameterized, not main-only. Live: objectui carries priority:p0 card objectui#5304, and its label OBJECT was auto-created by that first application with GitHub's default colour ededed and an EMPTY description - the exact drift the card describes.",
    "pm:seat belongs in the main-repo-only section": "CONFIRMED. There is exactly one seat post per domain:* / repo:* lane (SKILL.md), and both label families are created main-repo-only in this same script, so the label:pm:seat index can only be read on the main repo. Live: zero pm:seat cards in objectui and zero in objectos.",
    "boundary": "objectstack-ai/cloud is not in this session's allowed repo set, so the cloud half of both card counts is unverified. Neither placement rests on it: priority:p0 goes into the loop that already includes cloud, and pm:seat's argument rests on the lane families being main-repo-only."
  },
  "tests": "All at 0efb508 (git rev-parse --short HEAD after the final commit); exit codes captured before any pipe, never through tail. Gate union derived with `node scripts/pm/dispatch-gates.mjs` with NO paths (script takes its own change set off merge base 1800ffac2; it reported 'change set derived from git - 4 path(s)'). GREEN: check:pm-label-desc-cap (verdict line: 'check:pm-label-desc-cap: 16 label descriptions in scripts/pm/ensure-pm-labels.sh, all at most 100 characters (longest: 100, repo:objectui)'), plus its --self-test ('check-label-desc-cap --self-test: all cases passed'), check:node-version, check:required-contexts, check:workflow-status-functions, check:shard-attestation, check:cross-package-test-inputs, check:type-check-coverage, check:pm-dispatch-gates, check:nul-bytes - EXIT[...]=0 each. bash -n on ensure-pm-labels.sh: exit 0. Control-byte scan (grep -naP) over all four edited files: no match. ONE NON-GREEN, NOT CAUSED BY THIS DIFF: check:type-check-debt exits 1 on a build-state precondition - 'Error: --re-measure cannot run: 55 workspace dependenc(ies) of the ledgered packages have no built type entry point on disk'. Control: reverting lint.yml and package.json to origin/main in place reproduces the identical error, so it is the unbuilt worktree, not this change. Note: the first run of four gates failed with ERR_MODULE_NOT_FOUND for 'yaml' because I had not yet run pnpm install in the worktree; after install (under flock /tmp/os-heavy-verify.lock) all four pass.",
  "reverse_verification": "Three legs, each with the direction predicted in writing BEFORE running. (1) Over-cap plant, predicted RED: appended one character to repo:objectui's description (100 -> 101). Gate exit 1, message 'scripts/pm/ensure-pm-labels.sh:102: repo:objectui - description is 101 characters, over GitHub's 100-character cap'; the --self-test went red independently on TWO cases, including the fake-gh runtime leg 'no description GitHub would 422 reaches the fake gh', so the dry-run cross-check is live and not decorative. Restored byte-identical via git checkout of the committed blob: git hash-object = ec076de35b768d434c7e5c5efb1c225dce11e8c5 = git rev-parse HEAD:scripts/pm/ensure-pm-labels.sh, and the gate returned to exit 0. (2) Byte-vs-character control, predicted a byte guard reds a live label: measuring the same parse in bytes reds exactly needs:contract-review (97 chars / 101 bytes), which GitHub accepts; the shipped character guard reds nothing. This is the #10099 trap, demonstrated rather than asserted. (3) Vocabulary lines, predicted 4 and 1: fake gh on PATH running the REAL script invoked priority:p0 4 times (objectstack, objectui, cloud, objectos) and pm:seat exactly once (objectstack). Baseline leg on the origin/main copy of the same script: 48 label create calls, priority:p0 0, pm:seat 0; after: 53. No ablation here involves compiled artifacts, so no dist rebuild is implicated in any leg. Also worth recording: the self-test caught one WRONG expectation of mine on its first run (I expected the longest-value tie-break to pick v18 over v17 - same length, so it keeps the first); the case now pins the real tie-break behaviour.",
  "labels_read_back": "PR #10116 labels read back AFTER the write: ci/cd, dependencies, size/m, skip-changeset. The three bot labels were read first and written back as a union, so none was clobbered; skip-changeset is correct here because the diff is scripts/**, .github/workflows/** and one root package.json script entry, which publish nothing.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT FILED - duplicate search blocked by API rate limit, reporting to PM instead of filing blind. Finding: the LIVE label objects diverge from the vocabulary this script defines, and create-if-missing can never reconcile them. Measured 2026-08-20: objectstack's priority:p0 object reads 'Critical: blocker, must ship before MVP' (colour B60205), which contradicts the state-model meaning (queue-jump ORDERING, explicitly not an exemption and not a severity), and objectui's priority:p0 object has an EMPTY description and GitHub's default colour ededed. The script's own comment idiom already calls aligning existing objects 'a separate, deliberate PM action', but no mechanism or card carries that action, so the drift is permanent by construction. Suggest an unassigned finding card in the main repo (domain:skills) proposing either a one-off reconcile pass or a `gh label edit` reconcile mode for the script; I can file it once the rate limit clears if PM prefers."
  ]
}

Generated by Claude Code

PM 验收:ACCEPT(合并版,PR #10116 @ `0efb508`)

亲取 diff 复核(4 文件,两半各自成 commit):① 词表两行 —— priority:p0 落四仓循环、pm:seat 落主仓专段,与定级裁定一致,且 dev 把裁定验证后才落位(p0:H10/H11 扫描 repo-参数化 + objectui 有活体 p0 卡且其标签对象正是灰色空描述的漂移标本;pm:seat:lane 家族主仓专有 + objectui/objectos 零 pm:seat 卡),注释块按文件惯例写足四要素,描述 95/92 字符;② desc-cap gate —— 字符(码点)非字节(活体 needs:contract-review 97 字符/101 字节的正反钉都在自测里)、shell 变量按最长值展开计入、不可解析变量判红不跳过、干跑交叉验证防解析漂移、地板 12 防空解析装绿、lint.yml 无条件跑。反验三腿方向全中,恢复以 blob hash 证字节同一;自测还抓住 dev 自己一处错误预期(等长 tie-break),按实测钉住 —— 这是自测在干活的证据。check:type-check-debt 非绿经 origin/main 回退对照证明为未建 worktree 前置条件,非本 diff。

边界诚实:cloud 仓不在会话可达集,两处落位论证都不依赖它 —— 接受。

非治理面 → 转 ready + auto-merge(SQUASH)。#10099 的规格随本 PR 落地,合并后两卡同关(#10099 已 duplicate 关闭)。Dev 报告的 out-of-scope 发现(活体标签对象与词表脚本的永久漂移,无对账机制)由本座代立卡。

Generated by Claude Code

fix(pm): seed the `tracking` label in ensure-pm-labels.sh's five-repo…

Navigation Menu

Repository navigation

[finding] ensure-pm-labels.sh: pm:seat and priority:p0 are state-model rows with no entry in the four-repo vocabulary loop #10098

四维分析

PM 验收:ACCEPT(合并版,PR #10116 @ `0efb508`)

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] ensure-pm-labels.sh: pm:seat and priority:p0 are state-model rows with no entry in the four-repo vocabulary loop #10098

Description

Activity

os-warren commented on Aug 20, 2026

四维分析

os-warren commented on Aug 20, 2026

os-warren commented on Aug 20, 2026

os-warren commented on Aug 20, 2026

PM 验收:ACCEPT(合并版,PR #10116 @ 0efb508)

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

PM 验收:ACCEPT(合并版,PR #10116 @ `0efb508`)