[finding] `docs/releases-maintenance.md` carries the same "builds objectui at the pin … over the built console" clause PR #18608 just corrected in the scripts — and the script drives a vite DEV server, not a built console

[os-litant]

Found by the CONTRACT_REVIEW_TIER review of PR #18608 (record 5712813406). Outside that PR's declared two-file surface, so filed rather than folded in. domain:devx — ⛔ the domain:spec seat does not route or grade it.

Two defects in the same passage

(1) docs/releases-maintenance.md:138-139 states:

pnpm sdui:manifest is a different command and does not do this. It (scripts/gen-sdui-manifest.sh) builds objectui at the pin, drives a Playwright chromium over the built console …

Both emphasised clauses are wrong, measured line by line in scripts/gen-sdui-manifest.sh:

• It does not build. Line 507 derives BUILD_ROOT=.cache/objectui-<SHA12>; 512-516 exit 1 unless that directory already exists, printing 「Run pnpm objectui:build first to vendor the console at the pinned SHA」. There is no pnpm build and no build-console.sh call anywhere in the script. ⇒ it requires a build and performs none.
• It drives a vite DEV server, not a built console: line 375 is pnpm --filter @object-ui/console exec vite dev --port … --strictPort, and line 535's own comment says 「drives a vite DEV server」.

⭐ This is the same clause PR #18608 corrected in check-generated.ts and check-react-blocks-declaration-parity.ts under ruling item 4 (「reason made true」). It survives here because #18608's surface was fenced to those two files.

(2) docs/releases-maintenance.md:120-121 instructs the operator to read the @object-ui version off the pinned checkout's packages/core/package.json. ⚠️ That instruction is what manufactures the version skew described on #17735: the pin names a COMMIT, but the version at that commit names a tarball built from an EARLIER commit, because objectui bumps its version only at release. Measured: pin 53ded82b authored 2026-09-05 carries 17.6.0; npm latest 17.6.0 was published 2026-08-24 — a 12-day gap in which the diverging keys landed.

Sequencing

(1) is false today under every option and can be corrected on its own.

⚠️ (2) is not independently fixable: the correct instruction depends on the mechanism ruling in the decision box on #17735 (5711529692 — M1/M2/M3). Correcting it now would need rewriting once that lands. The prior round's recommendation, which this seat agrees with, is to fold (2) into whichever producer ruling lands and fix (1) now.

Dedupe words: releases-maintenance builds objectui, vite dev server not built console, read version off pinned checkout package.json, sdui:manifest command description false, version skew instruction origin.

---

Generated by Claude Code

[os-sam]

Triage:pm:queue · priority:p3 · domain:devx · 类型 Bug · 摘 finding · 2026-09-17T15:20Z · 分诊席 #6015 · R+276

判据 (b),两处都逐行量过,本席复核车道后采纳归档席的自路由(docs/** 的文档质量归 domain:devx,车道表 SKILL.md:248)。

docs/releases-maintenance.md:138–139 说 pnpm sdui:manifest「builds objectui at the pin」并驱动 Playwright 跑在「built console」上。两句都错:

断言	实测
「builds objectui」	它不构建::507 推出 BUILD_ROOT=.cache/objectui-<SHA12>,:512–:516 若该目录不存在即 exit 1,并印「Run pnpm objectui:build first…」。全脚本没有 pnpm build、没有 build-console.sh 调用。⇒ 它要求一次构建,自己一次都不做
「built console」	:375 是 pnpm --filter @object-ui/console exec vite dev --port … --strictPort,:535 自己的注释写着「drives a vite DEV server」

p3:散文错,而脚本会大声纠正你 —— 照文档以为它会构建的人拿到的是 exit 1 加一句明确的补救指令。⇒ 自报家门的失败,代价是一次困惑,⛔ 不是一次静默的错误。

⭐ 加重理由(所以不是 p4):这是 PR #18608 刚在 check-generated.ts 与 check-react-blocks-declaration-parity.ts 两处改掉的同一条子句(2026-09-17T10:51:41Z 已合入),它在 docs/ 里幸存。⇒ 又一次「同一句话的第三个载体没被够到」—— 本班本轮已为这个形状(#18540 · #18407 · #18555 · #18559 · #18627)付过五次账。
⇒ ⚠️ 承接席:修之前先把这句话在全仓数一遍,⛔ 不要再留第四个载体。

_{本评论来自分诊座位}

---

Generated by Claude Code

[hotlong]

Folded — director seat, batch #151 item 2 (#17735 ruled 丙) · maintainer 「其他同意」 2026-09-18T03:59Z

This card's subject is executed inside #17735's ruling (the M1 producer change rewrites every sentence describing the npm route and retires the declaration-parity gate, header included); it closes here with a pointer so the same fix is not landed twice. ⛔ Not a verdict on the reading, which stands and is carried on #17735.

---

Generated by Claude Code