docs(qa): start QA rounds on the maintainer's word and add the scenario half (#31489)

<!-- claude-projects-attribution: {"github_login":"zhuangjianguo"} -->
_Requested by **Steedos** · [project
thread](https://claude.ai/code/project/chan_018ETKaNwMZCNRYD7D8Q9Ubb?thread=cmsg_018ETKaNwMZCNRYD7D8Q9Ubb9R81zJBEnwtS9WxYnDKFP7)_

Fixes #31434

Clause-②: no
Metadata-protocol: none

## 维护者速读

- 改了什么：QA 席不再按每周定时跑，改为你说「一波开发完了」才开一轮；每轮在清单回归之外加跑场景库的 `starter`
卡（用本仓未发布的包）。
- 为什么改：照你在 QA 席会话里的两句裁决「也不需要每周跑，一波开发完了，我会和你说」与「纳入，本轮一起跑，立卡交 skills 车道」。
- 风险与回滚：只改三份内部规则文本，不动任何脚本、选择器（`owed:weekly` 原样）或产品代码；回滚即 revert 本 PR。
-
先看：`.claude/skills/pm-dispatch/references/lanes/qa.md`（〈每轮〉新增四行场景半边）、`.claude/skills/checklist-test/SKILL.md`
§6、`lanes/skills.md` 第 25 行。
- 两分钟验证：`pnpm check:pm-skill-ratchet`，预期三份文件行数都在上限内、全绿（qa.md
29/29，SKILL.md 279/279，skills.md 32/33）。
- 需要你做的：无，席位落地（Tier S，契约复审归总监席）。

## What changes

- `lanes/qa.md` 〈座位与唤醒〉: the weekly Routine fire is gone; a round opens
when the maintainer says a development wave is done, and a post-release
round still runs the npm smoke. The "first round, then every fire" line
is removed (its content is the next section's first line).
- `lanes/qa.md` 〈每轮〉: the scenario half, four lines: the `starter` cards
of `docs/qa/scenarios` with `--modes scratch --packages workspace` (a
wave is unreleased); one card per command (`--only ID`, the 2-hour
background cap) and at most two in parallel per container, citing
`packages/qa/scenarios/README.md` 〈Running the cards in a Claude Code
cloud session〉 for the subscription `oauth_token` login, the two API
variables left unset and a permission mode other than Auto (named, not
restated); the scorecard goes in the round report, not a `qa-run`
record; a failing half judged a product defect is extracted like a
checklist `fail`, its `QA-source:` line pointing at the round.
- `lanes/qa.md` 〈边界〉: a QA wave (release-candidate full run,
maintainer-named target) stays maintainer-triggered and distinct from
the wave-end rounds; the seat runs the section 6 routines plus the
scenario half.
- `checklist-test` section 6 points at `lanes/qa.md` for when a round
opens (one written copy of the cadence); "weekly / last week" become
"routine / last round" in section 6 and in retest trigger 4 of section
5.
- `lanes/skills.md` line 25: the section 6 routines and the scenario
half run per development wave and are not a QA wave.

Line budget, paid inside each file (no re-wrap): `qa.md` 29 → 29 (four
scenario lines paid by deleting the first-round line, folding the
`retest-owed` line into the four-record line, and replacing the three
restated section 5 lines on closing and Supersede with one pointer);
`checklist-test/SKILL.md` 279 → 279; `lanes/skills.md` 32 → 32. Every
changed line is at most 120 bytes.

Facts checked against `main` before writing: the six `starter` cards are
the top-level `tier` of `asset_lifecycle`, `leave_balances`,
`order_to_cash`, `requisition_to_receipt`, `service_desk` and
`travel_expense`; `--only`, `--modes scratch` and `--packages workspace`
are parsed in `packages/qa/scenarios/src/runner/cli.ts`; the API-billing
refusal is `API_BILLING_ENV` in `driver.ts`. No script reads
`owed:weekly` (`git grep` hits only the two rule files), so no selector
moves.

## Why

The maintainer's two rulings quoted above, recorded on the card.

## Verification

17 derived gates on head `7d0c77e4`, every one exit 0, reconciled with
`dispatch-gates --ran` (17 derived, 17 run, 0 NOT-MEASURED); after
merging `main` (head `d983bdef`, no overlap with these files)
`check:pm-skill-ratchet` and `check:pm-skill-id-lint` re-ran, exit 0.

<details><summary>Gate commands</summary>

```text
node scripts/check-closing-keyword-parity.mjs :: exit 0
node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0
node scripts/check-comment-mask-corpus.mjs :: exit 0
node scripts/pm/check-harness-current.mjs --self-test :: exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0
pnpm check:agent-test-spelling :: exit 0
pnpm check:cross-package-test-inputs :: exit 0
pnpm check:doc-authoring :: exit 0
pnpm check:driver-memory-census :: exit 0
pnpm check:gitlink-declared :: exit 0
pnpm check:nul-bytes :: exit 0
pnpm check:pm-governed-merges :: exit 0
pnpm check:pm-skill-id-lint :: exit 0
pnpm check:pm-skill-ratchet :: exit 0
pnpm check:refd-timer-probe :: exit 0
pnpm check:skill-frame-sync :: exit 0
pnpm check:watch-hint-literal :: exit 0
```

</details>

## Acceptance notes

- `skip-changeset`: `.claude/**` only, nothing publishes.
- Governed surface `.claude/**` (Tier S): lands on a contract-review
record for the current head.

## Human Minimum

- [x] 1. Schemas start as Zod; `packages/spec` holds no runtime logic.
(n/a)
- [x] 2. Every name I added uses the casing for its kind. (n/a)
- [x] 3. No formatter ran; `pnpm lint` passes. (no code)
- [x] 4. One issue, one branch off an up-to-date `main`, one pull
request.
- [x] 5. A change that publishes carries a changeset, and both
declaration lines are filled in here and in the changeset. (publishes
nothing)
- [x] 6. `packages/spec` touched: its generated artifacts are
regenerated and committed. (n/a)
- [x] 7. No hand edits to `content/docs/references/`,
`packages/*/CHANGELOG.md` or `content/docs/releases/`.
- [x] 8. Title, body and comments are in English.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---
_Generated by [Claude
Code](https://claude.ai/code/session_01QsR2jUgpaxRFL1ara3J4y8)_

Co-authored-by: Claude <noreply@anthropic.com>

Author: objectstack-fleet[bot]

.claude/skills/checklist-test/SKILL.md

@@ -234,7 +234,7 @@ QA-source: #运行记录号 · area.item · 条款
    单 —— 与跨仓联动同一纪律:第一产者是接受座位。安全加固没有卡:落地它的 PM 会话自
    己重跑该项,⛔ 不在任何卡上点名它。
 3. **维护者点名目标**。
-4. **常驻回归**(维护者所定,§6)⇒ 每周回归与发版后 npm 冒烟。
+4. **常驻回归**(维护者所定,§6)⇒ 例行回归与发版后 npm 冒烟。
 
 ⛔ 其它一切都不是重测触发:凭手感「再跑一遍」只增加收口债,不增加覆盖。
 
@@ -244,16 +244,16 @@ QA-source: #运行记录号 · area.item · 条款
 
 ## 6. 常驻回归 —— 两次发版之间也重跑
 
-维护者所定的两条例行(裁定评论 6501246005)归 QA 席 `qa#1`(pm-dispatch `lanes/qa.md`);其定时任务
-由维护者为该席建,⛔ 不是仓库代码,也不是合并门禁。
+维护者所定的两条例行(裁定评论 6501246005)归 QA 席 `qa#1`;何时开轮只见 pm-dispatch `lanes/qa.md`,
+⛔ 本节不另写节奏;例行不是仓库代码,也不是合并门禁。
 
-- **每周回归**,对 `main` HEAD,四张 `qa-run` 记录,一个选择器一张:
+- **例行回归**,对 `main` HEAD,四张 `qa-run` 记录,一个选择器一张:
   - 欠测清单:标题选择器写 `owed:weekly`(⛔ 不带冒号的词会被读成区),清单里每个 id
     按 §0 裸 id 逐项解析;
-  - `touched:<上周 touched 记录的 subject sha>`;首周取最近一张发版全量重跑记录的 subject;
+  - `touched:<上轮 touched 记录的 subject sha>`;首轮取最近一张发版全量重跑记录的 subject;
   - `priority:P0` 与 `surface:api`。
-  §5 的 Supersede 让每条例行至多开着一张:选择器周周相同的三张直接适用;`touched:` 的 sha
-  每周都变,本周那张照样 supersede 上周那张,判定结转照 §5。
+  §5 的 Supersede 让每条例行至多开着一张:选择器轮轮相同的三张直接适用;`touched:` 的 sha
+  每轮都变,本轮那张照样 supersede 上轮那张,判定结转照 §5。
 - **发版后 npm 冒烟**:每次发布后跑 `smoke:npm-<版本>`,形状照上一张同类记录;上一轮冒烟抽出
   的修复卡在这里重判(它们的溯源行见 §5 的冒烟例外)。
 - **欠测清单**由 `node scripts/qa/retest-owed.mjs` 只读算出(⛔ 不写 GitHub):

.claude/skills/pm-dispatch/references/lanes/qa.md

@@ -4,26 +4,26 @@
 ## 座位与唤醒
 
 - 席位 `qa#N`,首席 `qa#1`;签名、座位贴、信箱同他席;无 `domain:*` 标签、无 `pm:queue`。
-- Routine 席:维护者在 Routines UI 建每周 fire,发版后敲一轮;⛔ 本席不自建、不改频率。
+- 维护者说一波开发完了才开一轮,发版后另敲一轮冒烟;⛔ 本席不自启、不建定时任务。
 - 发版只认维护者一句或总监席摘要的新 tag 族读数;⛔ 不自判发版。
-- 开席首轮:先 `retest-owed.mjs`,再立第一张 `owed:weekly` 记录;此后每 fire 一轮。
 
 ## 每轮
 
-- 首动作 `node scripts/qa/retest-owed.mjs`(只读);欠测 id 即本轮 `owed:weekly` 记录的范围。
-- 每周回归照 §6 四张:`owed:weekly`、`touched:` 上周 subject、`priority:P0`、`surface:api`。
+- 先 `retest-owed.mjs`,再照 §6 四张:`owed:weekly`、`touched:` 上轮 subject、`priority:P0`、`surface:api`。
 - 发版后 `smoke:npm-<版本>` 照 §6;上轮冒烟抽出的修复卡在此重判。
+- 场景半边跑 `docs/qa/scenarios` 的 `starter` 卡:`--modes scratch --packages workspace`(开发波未发布)。
+- 一卡一命令 `--only ID`(后台限 2 小时),每容器至多两条并行;照 `packages/qa/scenarios/README.md`。
+- 其〈Running the cards in a Claude Code cloud session〉:订阅 `oauth_token` 登录、无 API 变量、非 Auto。
+- 计分卡进轮次记录(非 `qa-run`);失败半边判为产品缺陷 ⇒ 照 `fail` 抽卡,`QA-source:` 指本轮。
 - 执行照 checklist-test 与 RUNNER;本席即派 runner 的会话,runner 的安全复现私报本席。
-- 每张记录当轮照 §5 收口:四分法处置、收口评论机器行、抽取卡带 `QA-source:` 行。
-- 收口完即关记录 `completed`,一行指向收口评论;遗留项留在各自卡上,⛔ 不为等修复开着。
-- 新记录覆盖旧地面 ⇒ 先结转、再关旧记录,照 §5 Supersede 协议。
+- 每张 `qa-run` 记录当轮照 §5 收口、关闭与 Supersede;⛔ 不为等修复开着。
 - 轮末 `node scripts/qa/qa-rollup.mjs` 读开着的 `qa-run` 数;非零逐张点名欠什么,进轮次记录。
-- 每轮记花费行(环境准备与测试墙钟、runner 花费);三周读数进座位贴,频率由维护者定。
+- 每轮记花费行(环境准备与测试墙钟、runner 与场景半边花费);三轮读数进座位贴。
 
 ## 边界
 
 - ⛔ 不派 `os-dev`、不改产品代码、不开 PR;§3 清单修订同断言缺陷立卡交该项车道。
 - 抽出的缺陷卡照常进分诊首触,⛔ 不定级、不入队;安全发现只照 `../security-findings.md`。
-- 波次(发版候选全量、维护者点名目标)仍由维护者手动触发;本席只跑 §6 两条例行。
+- QA 波次(发版候选全量、维护者点名目标)另由维护者触发;本席只跑 §6 例行与场景半边。
 - 技能文本归 `domain:skills`、RUNNER 归 `domain:devx`;本席见文本问题以 `finding` 报,⛔ 不自改。
 - GitHub 写只走 checklist-test §4 写入门;推维护者只限需其出手,⛔ 无周报聊天表。

.claude/skills/pm-dispatch/references/lanes/skills.md

@@ -22,7 +22,7 @@
 - finding 归分诊首触,自分诊见 `triage-duties.md`;可附证据与「Instruments serve the fleet」答案。
 - 「Instruments serve the fleet」答不出的 finding ⇒ 建议 p3 或不做,⛔ 不自入队;定级归分诊。
 - 内部指令面中文,机器判据恒英文;认领前重读全线程;标签写 read-modify-write 加回读。
-- QA 波次由维护者手动触发,⛔ 不自启;checklist-test §6 两条例行按期由 `qa#1` 跑,不是波次。
+- QA 波次由维护者手动触发,⛔ 不自启;`qa#1` 的 §6 例行与场景半边逐开发波跑,不是波次。
 
 ## 席内判断