scripts/pm/ensure-pm-labels.sh: pm:retriage missing from the four-repo pm state-machine vocabulary loop

[os-warren]

Observation found while implementing the pm:retriage state-model row (backlink: implementation PR #10089, card #10081 — that card's scope was ruled SKILL.md-only, so this is deliberately not a rider there).

Blocked-by: #10081

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", and its first loop creates the full pm state-machine vocabulary across all four repos (objectstack, objectui, cloud, objectos). Once the pm:retriage row lands in the state model, that identity claim drifts: the new label is absent from the script's loop.

Functionally covered today: the SKILL.md row says the label is created on first application (triage's cross-repo duty), so nothing is broken for users — this is script/state-model drift, not a defect anyone hits. Cost of the drift: a first application on a repo where the label object does not yet exist depends on the applying seat remembering to create it, and the script's rerun can never repair the gap.

Proposed shape when graded: one gh label create pm:retriage line inside the existing four-repo loop, description kept at or under 100 characters (the script header documents the hard cap: longer descriptions 422 and the idempotency || true swallows the failure silently, so the label is never created and a rerun cannot repair it).

Duplicate check: searched open issues for ensure-pm-labels / label-vocabulary drift (zero hits; positive control on "pm:retriage" returns the parent card).

---

Generated by Claude Code

[os-warren]

Claim: session ae0d8bd1-5667-5f58-8ff8-fa22ec154325 · branch claude/issue-10090-ensure-labels-retriage

解锁:Blocked-by #10081 已随 PR #10089 合并(维护者手合)关闭,pm:retriage 条文生效。折叠检查:ensure-pm-labels.sh 无同文件竞争 → 单卡。模型档:路径无强制,S 级机械 → opus。注:objectstack 仓的标签已在首次应用时自动创建(#9756/#9705),本卡交付的是四仓词表的持久化(脚本是权威)。

---

Generated by Claude Code

[os-warren]

{
 "issue": 10090,
 "status": "done",
 "branch": "claude/issue-10090-ensure-labels-retriage",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/10100",
 "premise_still_valid": true,
 "summary": "Added one `gh label create pm:retriage` line to the four-repo loop in scripts/pm/ensure-pm-labels.sh, between pm:blocking and finding (the state-model table's own ordering), with a comment block in the pm:blocking / needs:contract-review idiom citing the maintainer ruling verbatim. Description is 96 chars, inside the header's documented 100-char hard cap; colour d4c5f9 is unused by every other label in the file and deliberately not grey, because the state-model row's stated purpose for coexistence is list visibility. Verified the premise on origin/main first: the header's identity claim and the four-repo loop are exactly as the card describes, and the pm:retriage row is live in SKILL.md. The objectstack label object already exists (ededed, empty description) so the script is a no-op there — the durable contribution is the other three repos, stated in the PR body along with the consequence that this script never calls `label edit` and so will not re-colour the existing object.",
 "tests": "All at final commit 88abe7588 (dispatch-gates confirmed the tree: 'change set derived from git — 1 path(s) vs merge base 907c11d2c of origin/main and HEAD', 'committed 1, working tree 0, untracked 0'). Union re-derived with `node scripts/pm/dispatch-gates.mjs`, no path args, EXIT=0 — it named `check:cross-package-test-inputs` (lint.yml + ci.yml, via 'scripts/**'), which the dispatch list did not; ran it plus the dispatch-named nul-bytes family. Exit codes captured by redirecting to a file BEFORE any pipe. Verdict lines: `pnpm check:cross-package-test-inputs` EXIT=0 → 'All 57 self-test cases passed.' / 'OK: 12 package(s) read outside themselves, all declared, and turbo.json hashes every declared glob.'; `pnpm check:nul-bytes` EXIT=0 → '✓ check-nul-bytes --self-test: 75 assertions over a temp git repo (real scan() path)' / 'check-nul-bytes: OK (scanned 6367 text file(s) -- 6367 tracked, 0 untracked-not-ignored; skipped 5 binary; no raw ASCII control bytes).'; `bash -n scripts/pm/ensure-pm-labels.sh` EXIT=0; manual control-byte scan of the edited file (grep -naP) returned no match. The script has NO self-test or dry-run mode, so verification used an out-of-tree harness putting a fake `gh` on PATH that records every `label create` call: with the fix, 48 create calls, `pm:retriage d4c5f9 96c` once per repo on all four, 'DESC-CAP: PASS (every -d ⩽ 100 chars)'. REVERSE VERIFICATION, direction predicted in writing before running — prediction: deleting the added line loses exactly 4 rows, TOTAL 48→44, DESC-CAP stays PASS, and NO gate reddens (the drift is silent, since nothing mechanically couples SKILL.md's state model to this script). Observed exactly that, diff showing only the four pm:retriage rows plus the total. Restore was from the commit (not from origin/main), confirmed byte-identical: sha256 376359c14aea43f55aa1c9b95d46d7c165491d1877c460b3a6ef12666e21e762 matching the pre-ablation hash, `git diff --exit-code` EXIT=0, and the restored listing diffed identical to the pre-ablation one. NOT VERIFIABLE FROM THIS CONTAINER, stated plainly: the script's live half — there is no `gh` CLI installed here and no label-write credentials, so no label object was created or edited by this run; the only live read was the existing objectstack object's colour/description via the GitHub API. Running the script stays a PM landing step.",
 "labels_read_back": ["size/s", "skip-changeset"],
 "open_questions": [],
 "out_of_scope_findings": [
 "filed as #10098: pm:seat and priority:p0 are state-model rows with no entry in the loop — same drift class, deliberately NOT fixed in-place because the correct shape (which loop each belongs in; whether the absence is deliberate) is pinned by no existing evidence",
 "filed as #10099: the 100-char label-description cap is a documented silent-failure mode with no mechanical guard behind it — live headroom is one character on repo:objectui, and the cap is measured in characters not bytes"
 ]
}

Notes for the PM, outside the JSON:

• Duplicate check for both findings could not use mcp__github__search_issues — its positive control (searching for pm:retriage, which this very open card contains) returned zero, so it was producing false negatives. Fell back to enumerating all 11 open domain:skills cards; no duplicate.
• gh is not installed in this container, so any protocol step that assumes the CLI has to be mirrored through the MCP tools.
• CI at report time: only the Vercel status had reported (success, "Canceled by Ignored Build Step"); the gate farm was still in_progress. Per the standing contract, no waiting.

---

Generated by Claude Code

[os-warren]

ACCEPT — PR #10100 复核通过,已 flip ready + auto-merge(显式 SQUASH)。注意:同 CI 的 Lint & Repo Gates 可能撞 #10071 flake —— 撞了不算本 PR 的红,静候即可。

复核记录(session ae0d8bd1-5667-5f58-8ff8-fa22ec154325):

• 亲取 diff:一行 + 注释块,措辞照 pm:blocking 先例,排序照状态模型表,颜色刻意非灰(并存可见性是本行的存在目的)、描述 96c 在 100c 硬帽内 ✓。
• 无自测脚本的验证方案得当:树外 fake-gh 采录 harness(48 次 create、四仓各一次 pm:retriage、DESC-CAP 全过),反向验证预测「删行丢恰 4 行且零门禁变红(漂移是静默的)」全中 —— 把「没有门禁护这个」如实测出来而不是宣称覆盖 ✓。
• 活半边如实不可验(本容器无 gh 无标签写凭证):脚本实跑留作维护者落地步;运维上无空窗 —— 姐妹仓标签会在首次 API 应用时自动创建(objectstack 已循此例),脚本是持久词表非前置条件。
• Out-of-scope 两卡([finding] ensure-pm-labels.sh: pm:seat and priority:p0 are state-model rows with no entry in the four-repo vocabulary loop #10098 pm:seat/priority:p0 同类漂移待证、[finding] ensure-pm-labels.sh documents a silent 100-char label-description failure mode with no mechanical guard behind it #10099 100c 帽无机械守卫)已收,下轮首触。

---

Generated by Claude Code