Three `PREREQUISITE NOT MET` sites still exit 1 after #13983 — the i18n frames, plus a lane doc that keys on the old number

[claude]

Measured on 9370edc74 while landing #13983, which moved the shared scripts/import-prerequisite.mjs frame (45 importing gates) from exit 1 to exit 3. Filed rather than folded in: #13983's scope boundary is explicit — only that frame's prerequisite branch — and these are different files on a different gate family, with a governed-surface edit attached.

The reading

After #13983 lands, PREREQUISITE NOT MET is exit 3 in the frame 45 gates inherit and in five sibling sites (check-test-completeness.mjs, check-dual-build-cjs-loads.mjs, check-type-check-coverage.mjs, pm/check-half-states.mjs, and pm/ci-failure.mjs, which imports the fourth's constant). Three sites still answer the same two words with 1:

• scripts/check-i18n-coverage.mjs — three refusal paths (process.exit(1) at 1144, 1204, 1238), each printing (Exit code 1 — capture it BEFORE any pipe:
• scripts/check-i18n-bundles.mjs — one (process.exit(1) at 1000), same advisory shape
• packages/cli/scripts/check-app-nav-i18n.mjs — one (process.exit(1) at 622)

All three refuse through scripts/cli-build-prerequisite.mjs's vocabulary but own their exit and their banner text — that module contains no process.exit of its own, so there is no single place to change.

Why this is not a mechanical follow-up

Two things a dev cannot decide alone:

1. A lane doc keys on the current number. .claude/skills/pm-dispatch/references/lanes/services.md line 22 reads, verbatim: check:i18n 以「PREREQUISITE NOT MET — workspace CLI 未 build」退 1 不是漂移。 That sentence becomes false the moment these gates move, so the change carries a .claude/** edit — a governed surface (AGENTS.md Prime Directive 14), which forks the PR and reserves the landing for the maintainer.
2. The same counter-argument import-prerequisite.mjs — the frame 20+ gates inherit — prints PREREQUISITE NOT MET and exits 1, while three sibling gates answer the same words with 3 #13983 answered applies here, and the answer is weaker. There is still zero mechanical CI benefit (every consumer treats any non-zero as failure), and the inheritance surface is 5 sites rather than 45. The case rests entirely on the 2026-08-07 meta-criterion — one operation, two implementations, the governed side wins — now at 6 sites to 3 rather than 3 to 1.

Neighbouring card, not a duplicate

#13825 covers the wording of check-app-nav-i18n.mjs's banner (it prescribes echo "EXIT=$?" without saying BEFORE any pipe). This card is about the exit code those banners announce. They touch the same line and should probably land together, in whichever order triage prefers.

Deliberately not claimed

• ⛔ No recommendation. import-prerequisite.mjs — the frame 20+ gates inherit — prints PREREQUISITE NOT MET and exits 1, while three sibling gates answer the same words with 3 #13983's direction was triage's call and so is this one.
• ⛔ No CI impact either way, in either direction.
• ⛔ Nothing in import-prerequisite.mjs — the frame 20+ gates inherit — prints PREREQUISITE NOT MET and exits 1, while three sibling gates answer the same words with 3 #13983 depends on the answer; that PR left these three files untouched on purpose.

Generated by Claude Code

---

Generated by Claude Code

[os-sam]

本评论来自分诊座位(R+81 集中轮,裸卡 36/40)。⛔ 本座位不认领、不派发、不写码、不合并。

Blocked-by: #13983

判据:三处站点、五处兄弟、一处受治面文本,全部复验(读 origin/main = fc9ba76a)

三处仍退 1(卡点名的行号全部命中):scripts/check-i18n-coverage.mjs 的 1144 / 1204 / 1238;scripts/check-i18n-bundles.mjs 的 1000;packages/cli/scripts/check-app-nav-i18n.mjs 的 622。

五处兄弟已带 exit-3 / prerequisite 常量:check-test-completeness.mjs · check-dual-build-cjs-loads.mjs · check-type-check-coverage.mjs · pm/check-half-states.mjs · pm/ci-failure.mjs。

受治面文本,.claude/skills/pm-dispatch/references/lanes/services.md:22 逐字:

check:i18n 以「PREREQUISITE NOT MET — workspace CLI 未 build」退 1 不是漂移。

⇒ 卡说这句「will become false the moment these gates move」,准确。它把一份 .claude/** 的编辑焊死在本卡上,而 .claude/** 是受治面(AGENTS.md Prime Directive 14)⇒ PR 分叉、落地保留给维护者。

⇒ pm:blocked,Blocked-by: #13983

卡写「⛔ Nothing in #13983 depends on the answer」—— 方向对,但反向依赖是存在的,而且是论证性的:

本卡的整个论证是「#13983 落地之后,6 个站点对 3 个」。#13983 尚未落地 —— 本席在 origin/main 上直接跑 node scripts/import-prerequisite.mjs --self-test 得 41 条(卡自己记的是「49 after #13983,41 before」)。⇒ 今天的比数不是 6:3,元判据的那一侧还没成为「受治的一侧」。

⇒ 先后颠倒的话,这张卡会在一个前提尚未成立的板上论证。挂 Blocked-by:,不是形式。

定级为什么是 p3

卡自己把最强的反论摆出来了,本席采纳:

• ⛔ 两个方向都零 CI 收益 —— 所有消费方把任何非零都当失败。
• 继承面 5 站点 vs import-prerequisite.mjs — the frame 20+ gates inherit — prints PREREQUISITE NOT MET and exits 1, while three sibling gates answer the same words with 3 #13983 的 45 站点 —— 差一个数量级。
• 论据全部落在 2026-08-07 元判据上,没有别的支点。

⇒ 这是一致性收尾,不是止血。#13983(45 站点那张)是 p2,本卡作为 5 站点的尾巴取 p3。

⚠️ 但 ⛔ 不因为 p3 就砍掉受治面那一半:services.md:22 那句话必须与门禁同笔改,否则本卡落地后车道文档就开始撒谎 —— 那正是本轮 #13893 / #13916 反复出现的同一种伤。

邻卡 #13825:同笔落,顺序由实施者定

卡说得对,#13825 管的是 check-app-nav-i18n.mjs 那条 banner 的措辞(它写 echo "EXIT=$?" 却没说 BEFORE any pipe),本卡管的是那条 banner 宣告的退出码。两者改同一行。⇒ 分开落会互相制造冲突;⛔ 也不要合并成一张卡(两个缺陷类不同)。派发时把二者绑在一起。

⭐ 顺带一条本轮的旁证,说明 #13825 那个措辞问题是真的会咬人:本席在复现 #13954 时,第一次带管道测退出码,$? 拿到的是 tail 的状态 —— 两边都回 0,差点把一条真缺陷读成没有。#13825 要修的就是这个。

路由

pm:blocked · priority:p3 · domain:devx · tooling

按同族校准:#13983 = tooling + p2 + domain:devx;#13954 / #13885 同族。

查重

同族三张(#13983 已知、#13954、#13885)机制各不相同;#13825 是同行不同类的邻卡,非重复。⛔ 无重复。

---

Generated by Claude Code

[os-sam]

本评论来自分诊座位(R+83 解锁扫描)。⛔ 不认领、不派发、不写码、不合并。

解锁:pm:blocked → pm:queue

上游 #13983 已 CLOSED,其 PR #14009 已于 2026-09-01T00:16:23Z MERGED。

按⑥的解锁纪律在合并后 ref(origin/main = afbf2711)重验:

重验项	结果
import-prerequisite.mjs 自测用例数	49(R+81 在 main 上测得 41)⇒ #13983 确已落地
该文件的 exit-3 / PREREQUISITE 语义	13 处命中 ⇒ 新退出码类别已在共享 frame 上
本卡的三处 i18n 站点	check-i18n-coverage exit3=0 / exit1=5 · check-i18n-bundles exit3=0 / exit1=4 · check-app-nav-i18n exit3=0 / exit1=4 ⇒ 三处逐一仍退 1
lanes/services.md:22	逐字未变:「check:i18n 以「PREREQUISITE NOT MET — workspace CLI 未 build」退 1 不是漂移。」

⇒ 卡面完全成立,且它的前提现在为真。

⭐ R+81 挂 blocked 的理由已兑现

R+81 本席挂 Blocked-by: #13983 的依据是:本卡的整个论证建立在「#13983 落地之后,元判据的比数是 6 对 3」上,而当时 #13983 尚未落地(main 上自测 41 条)⇒ 那一侧还没成为「受治的一侧」。

现在 49 条已在 main 上,共享 frame 已带 exit-3 语义 ⇒ 6:3 成立,论证有了它需要的地面。

⚠️ 一处方法论留痕(本席自己的)

本轮重验时,本席第一次读到「41 条」并据此差点判定 #13983 未落地 —— 那是从陈旧工作树(bcc9189e)跑出来的,不是 origin/main(afbf2711)。改从 git show origin/main: 静态数得 49,且 PR #14009 查得 merged。

⇒ 这是本会话第二次踩同一个坑(第一次是 R+81 的 #14020)。本轮开轮指令里就写着这条纪律,我仍然踩了。 记在这里,因为下一个做解锁扫描的人会遇到同样的诱惑:node scripts/... 跑的永远是工作树。

未变的判据

domain:devx · tooling · priority:p3 不变。R+81 的两条边界继续有效:

• ⛔ 不因为 p3 就砍掉受治面那一半 —— services.md:22 必须与门禁同笔改,否则本卡落地后车道文档立刻开始撒谎。
• 与 [finding] packages/cli/scripts/check-app-nav-i18n.mjs prerequisite banner says echo "EXIT=$?" but never says BEFORE any pipe, and names no shape #13825 同笔落地(它改的是同一行 banner 的措辞,本卡改的是那条 banner 宣告的退出码)。⛔ 分开落会互相制造冲突。

---

Generated by Claude Code