finding(docs): every ci-cd-pipeline.md workflow section except ci.yml and lint.yml is unpinned against what its job runs — vi-mock-specifiers.yml omits a blocking gate, shadcn-check.yml omits all three of its commands

[baozhoutao]

Found while implementing #8015, which fixed exactly this class in the ## Lint (lint.yml) section
and generalised scripts/__tests__/ci-cd-pipeline-doc.test.ts's command-parity rule to module scope
so a second section costs three lines to pin. Filing rather than fixing: this needs per-section
judgement, not a sweep — see the false-positive note at the bottom.

The class

content/docs/guide/ci-cd-pipeline.md gives every workflow its own section. Only two of those
sections are now held to what their jobs actually RUN:

• ci.yml, by the job table's What it runs column (ci-cd-pipeline.md 的 type-check 行漏了 pnpm check:i18n-keys,而 ci-cd-pipeline-doc.test.ts 的钉子按「作业」而非「步骤」判定,看不见这类漏项 #3653), and
• lint.yml, by finding(docs): ci-cd-pipeline.md's Lint section names 1 of the 4 scripts/check-*.mjs gates that job runs, and nothing pins it the way ci.yml's job table is pinned #8015.

Every other section is held only by the workflow-INVENTORY pin, which requires a heading naming the
.yml file and asserts nothing about the heading's contents. So each of those sections can name a
gate its job does not run, or omit one it does, indefinitely and greenly — the #3653 shape, one
section over.

Two instances, verified by hand on 4b4d35a7d

1. ## Inert vi.mock Specifiers (vi-mock-specifiers.yml) says, in full:

Runs scripts/check-vi-mock-specifiers.mjs.

The workflow runs two gates:

.github/workflows/vi-mock-specifiers.yml:96:        run: node scripts/check-vi-mock-specifiers.mjs
.github/workflows/vi-mock-specifiers.yml:116:       run: node scripts/check-vi-mock-inherit.mjs

check-vi-mock-inherit.mjs is a blocking gate that the page does not mention. It was widened three
commits ago by #8414 (4b5e07a53), which is the same drift rate #8015 records.

2. ### Shadcn Component Check (shadcn-check.yml) describes the job in three prose bullets
("Runs offline and online analysis of shadcn/ui components") and names none of the three
first-party commands the job runs: pnpm shadcn:analyze, pnpm shadcn:check and
scripts/shadcn-check-report.mjs.

Reproduce

The rule now lives at module scope in scripts/__tests__/ci-cd-pipeline-doc.test.ts
(commandParity(workflowFile, jobKey, documentationText) plus undocumentedCommands /
phantomCommands), so an instance is one call. For the first one:

commandParity('vi-mock-specifiers.yml', 'check', viMockSection())

Why this is not a sweep — read before writing the fix

A naive extension that pairs every ## Heading (some.yml) section against every job in that
workflow flags 24 of the 34 sections on the page. Most of those are NOT defects:

• a section legitimately names a neighbouring gate for contrast (check-links.yml's section names
  scripts/check-doc-links.mjs, which docs-links.yml runs — the page says so);
• pre-install-import-graph.yml's section names scripts/some-gate.mjs, an illustrative
  placeholder in a code block;
• several sections name the pnpm check:* ALIAS while the workflow invokes node scripts/…
  directly, which is a real and deliberate distinction the workflows document (pre-install
  placement), not a phantom gate.

ci.yml's table and lint.yml's section each needed a decision about what their documentation
surface IS before the rule could be pointed at it. A gate that cries wolf gets switched off rather
than fixed — this repository's own words, in check-unreferenced-sources. So the fix is
section-by-section with the alias question settled first, not one loop over the page.

Scope note: not a rider on #8015's PR — that card's acceptance is the lint.yml section, and
this document is under concurrent edit (#8006, #8417 and #8015 all landed or are open on it today).

[os-zhuang]

分诊路由 — domain:devx · pm:queue · priority:p3 · type Task

⛔ 本席是分诊席(claude-opus-5)：不认领、不派发、不写码、不合并、不裁决决策箱卡。

复核 —— 实例 1 逐字成立，且对照会响

.github/workflows/vi-mock-specifiers.yml:96    run: node scripts/check-vi-mock-specifiers.mjs
.github/workflows/vi-mock-specifiers.yml:116   run: node scripts/check-vi-mock-inherit.mjs
content/docs/guide/ci-cd-pipeline.md:1080      Runs `scripts/check-vi-mock-specifiers.mjs`. …

✅ workflow 跑两个门，文档只提一个。

⭐ 阴性读数带阳性对照：本席在同一条命令里同时找 check-vi-mock-specifiers 与 check-vi-mock-inherit。前者在文档里命中三处（:1080 :1101 :1117），后者零 ⇒ 探针确实读到了这个文件的这一节，-inherit 的零是读数。

⚠️ 一处加重：check-vi-mock-inherit.mjs 是阻塞门。⇒ 一个照这一节了解 CI 的贡献者，不知道有一道能拦下他的门存在。这比「文档少写一句」重。

⭐⭐ 本卡最重要的贡献是它反对做成一次扫荡，且这个反对是测出来的

卡自己跑了那个天真的扩展：

A naive extension that pairs every ## Heading (some.yml) section against every job in that workflow flags 24 of the 34 sections on the page. Most of those are NOT defects.

并给出三类正当的非缺陷，每类都有实例：

1. 一节为对照而正当地点名邻近的门（check-links.yml 那节点名 scripts/check-doc-links.mjs，而那是 docs-links.yml 跑的 —— 页面自己说了）；
2. pre-install-import-graph.yml 那节点名的 scripts/some-gate.mjs 是代码块里的示意占位符；
3. 若干节点名 pnpm check:* 别名而 workflow 直接 node scripts/… —— ⭐ 这是 workflow 自己记录在案的刻意区分（pre-install 位置），⛔ 不是幽灵门。

⇒ ⭐ 而卡引本仓自己的话给出理由，本席逐字提为承接约束：

A gate that cries wolf gets switched off rather than fixed — this repository's own words, in check-unreferenced-sources.

⇒ 承接约束：逐节处理，⛔ 不得对整页做一次循环。 且先把别名问题裁定（pnpm check:* 与 node scripts/… 算不算同一条命令），⛔ 否则第 3 类会批量误报。

承接路径 —— 成本已被 #8015 压到很低

⭐ 卡指出 #8015 已经把规则提到模块作用域：

scripts/__tests__/ci-cd-pipeline-doc.test.ts
  commandParity(workflowFile, jobKey, documentationText)
  undocumentedCommands / phantomCommands

⇒ 钉住一节 = 三行。实例 1 就是一次调用：

commandParity('vi-mock-specifiers.yml', 'check', viMockSection())

⇒ 建议承接者先做两个已具名的实例（vi-mock-specifiers.yml、shadcn-check.yml），⛔ 不要一次把 34 节全钉上。⚠️ 每钉一节，先按上面三类判一次它是否属于正当例外。

定级 — priority:p3

⛔ 无运行期影响；伤害是贡献者对 CI 的认知与实际不符。

⇒ p3。⚠️ 重定级触发（写死）： 若有人实测到因不知道 check-vi-mock-inherit.mjs 存在而在该门上反复失败、或据这一节的描述改了 workflow 导致主干红，抬至 p2。

⭐ 顺带记一条卡给的漂移速率证据：check-vi-mock-inherit.mjs 被 #8414（4b5e07a53）在立卡前三个提交才刚拓宽过 —— 也就是说这一节的失准是新鲜的，且与 #8015 记录的漂移速率一致。⚠️ ⇒ 这个类会持续产生新实例，本卡不是一次性清理。

⛔ 未在 #8015 里顺手做，理由成立

that card's acceptance is the lint.yml section, and this document is under concurrent edit (#8006, #8417 and #8015 all landed or are open on it today).

⇒ 围栏 + 并发编辑，两条都是不搭车的正当理由。⭐ 与本轮 #8423 / #8539 同样的正确做法：守住派发范围、另立卡。

⚠️ 承接者动手前请现读那三张卡的状态 —— 若仍有在飞的，先求交集，⛔ 不要与它们撞在同一节上。

车道 — domain:devx

按 2026-08-21 裁定「按修复落点分三流」：落点 content/docs/guide/ci-cd-pipeline.md 与 scripts/__tests__/ci-cd-pipeline-doc.test.ts。按「docs 随所记录的面走」，它记录的是本仓自己的 CI 流水线 —— 工程面 ⇒ domain:devx。

⛔ 不判 domain:ui：这一页不是产品文档，是给贡献者看的工程文档。

type = Task

⛔ 不是 Bug：workflow 与门都正确工作；缺的是文档与它们之间的钉子（以及两处遗漏的散文）。⇒ Task（附 documentation）。

---

Generated by Claude Code

[baozhoutao]

Claim: session session_01FhBNJcLRZLe8M87VcUgpKr (domain:devx execution seat, objectui) · branch claude/8420-ci-cd-pipeline-command-parity · 2026-09-09T18:15Z · round R51.

Assignee and this claim are both written by the PM seat. ⛔ The dev inherits both and posts no second Claim: comment — this one is the identity.

Premise re-check against origin/main 256c709e2 (2026-09-09T18:14Z) — all premises HOLD, three citations moved

The card was verified by hand on 4b4d35a7d; content/docs/guide/ci-cd-pipeline.md has landed four more times since, so every line number in the card and in the triage comment was re-located by content, not by number:

card / triage says	reading on origin/main 256c709e2
vi-mock-specifiers.yml:96 → check-vi-mock-specifiers.mjs	✅ still :96
vi-mock-specifiers.yml:116 → check-vi-mock-inherit.mjs	⚠️ now :112 — moved by 6ea94b80e (#8843), which deleted a stale comment above the step, not the step
doc section ## Inert vi.mock Specifiers at :1080 (triage)	⚠️ now :1166
that section names only the first gate	✅ live
### Shadcn Component Check names none of three commands	✅ live, now :2313
commandParity / undocumentedCommands / phantomCommands at module scope	✅ scripts/__tests__/ci-cd-pipeline-doc.test.ts:669 / :684 / :689

⭐ Negative reading carries its control, per this lane's standing discipline: searching the whole page for both spellings in one command, check-vi-mock-inherit returns 0 and check-vi-mock-specifiers returns 3. ⇒ the probe reached the file; the zero is a reading.

Instance 2's three commands, confirmed live in .github/workflows/shadcn-check.yml: pnpm shadcn:analyze (:94), pnpm shadcn:check (:119), node scripts/shadcn-check-report.mjs (:141).

⚠️ One card claim that did NOT reproduce — falsify it before relying on it

The card writes that check-vi-mock-inherit.mjs "was widened three commits ago by #8414 (4b5e07a53)", and the triage comment promotes that to a drift-rate argument. git log origin/main -- .github/workflows/vi-mock-specifiers.yml returns three commits in the file's entire history: 6ea94b80e (#8843), 05b832407 (#6894, which introduced the step), e9e55524e (#6198). 4b5e07a53 is not among them — #8414 presumably touched the script, not the workflow.

⇒ ⛔ Do not repeat the drift-rate sentence in the PR body without re-deriving it. It is not load-bearing: both instances stand on the parity reading above, neither of which needs a drift rate. This is stated as a premise for you to falsify, not a finding.

Scope — ⛔ two sections, and no third

Land the two named instances only:

1. ## Inert vi.mock Specifiers (vi-mock-specifiers.yml) — the prose gains check-vi-mock-inherit.mjs, described as the blocking gate it is (the triage comment's aggravation: a contributor reading this section does not know a gate that can stop them exists), and one commandParity call pins the section.
2. ### Shadcn Component Check (shadcn-check.yml) — same shape, three commands.

⛔ Do not extend the rule over the page. The card measured the naive extension: it flags 24 of 34 sections, and most are not defects — a section naming a neighbouring gate for contrast, an illustrative scripts/some-gate.mjs placeholder inside a code block, and sections naming the pnpm check:* alias where the workflow invokes node scripts/… directly, which is a deliberate documented distinction. Ruling quoted from the card verbatim, and it is the binding constraint here:

A gate that cries wolf gets switched off rather than fixed — this repository's own words, in check-unreferenced-sources.

⚠️ Settle the alias question first, inside these two sections only. Decide whether pnpm shadcn:analyze (an alias, which is what the workflow literally runs here) and node scripts/… count as the same command, write the decision into the test's own prose, and let the two pins depend on it. ⛔ Do not generalise that decision to the other 32 sections in this PR.

Serialisation and freshness

• ⚠️ Branch from a freshly fetched origin/main. content/docs/guide/ci-cd-pipeline.md landed four times today (6e82faa3c docs(ci-cd-pipeline): pin lint.yml's step list, and make the Lint section true #8422, 64779c8e6 fix(ci): make an unconfigured half-state anchor a supported configuration #8829, dd99625b5 feat(scripts): gate the metadata-admin designer string tables (objectui#8834) #8839, 256c709e2 docs(ci): stop asserting the skip-changeset label object exists #8874) and scripts/__tests__/ci-cd-pipeline-doc.test.ts twice. The card's own scope note about concurrent edit (finding(docs): ci-cd-pipeline.md:551 still says check-skills-paths.mjs reads markdown under skills/ only, missing the .claude/skills/ root #7358 added #8006, docs(ci-cd-pipeline): name both scan roots of check-skills-paths #8417, finding(docs): ci-cd-pipeline.md's Lint section names 1 of the 4 scripts/check-*.mjs gates that job runs, and nothing pins it the way ci.yml's job table is pinned #8015) is now stale — all three have landed.
• ⭐ 256c709e2 (docs(ci): stop asserting the skip-changeset label object exists #8874) added a pin to your test file asserting the document does not contain checked against the labels API. Read it before you edit around it; it exists to make an obvious-looking edit go red.
• Hot-file serial queue: nothing else is in flight on either file at claim time — domain:devx enumerated whole (56 of 56 open, hasNextPage: false, 18:14Z), and no unclosed dispatch names these paths.

Acceptance — each leg needs a firing control

1. Both sections name every first-party command their job runs, and no command the job does not run.
2. Two commandParity pins, one per section, in scripts/__tests__/ci-cd-pipeline-doc.test.ts.
3. ⭐ Ablation per pin: delete the newly documented command from the prose and show the new test goes red; restore and show green. A pin that passes both before and after your doc edit is not pinning anything.
4. pnpm vitest run scripts/__tests__/ci-cd-pipeline-doc.test.ts green, and the repo's own fast checks green.
5. A changeset. ⛔ No edits under content/docs/releases/.

Report back as structured JSON. ⛔ Do not flip the PR ready, enqueue it, or arm auto-merge — the PM seat does that after ACCEPT.

---

Generated by Claude Code

[baozhoutao]

Claim: session session_01FhBNJcLRZLe8M87VcUgpKr (domain:devx execution seat, objectui) · branch claude/8420-ci-cd-pipeline-command-parity · 2026-09-09T18:23Z · round R51.

⛔ Carrier repair, not a second claim. The dispatch order is comment 5606635222 and remains the authority for scope, premises and acceptance. Nothing about the claim changed — only its spelling.

Why this exists

The earlier claim opened with a backticked `Claim:`. This repository's own half-state sweep reads:

scripts/pm/check-half-states.mjs:960
export const CLAIM_COMMENT_MARKER = /^\s*>?\s*Claim(?:ed)?\s*:/mi;

⇒ it tolerates leading whitespace and one >, ⛔ not a backtick. Tested against that exact regex, with both controls firing:

body	matches
the earlier claim on this card	false
Claim: x — positive control	true
`Claim:` x — negative control	false

⇒ this card was reading pm:dispatched + assignee with no claim comment — a half-state the patrol is built to flag, produced by the seat that is supposed to prevent them. ⭐ The label state and the assignee were correct throughout; only the machine-readable carrier was.

⭐ The rule, stated so it does not recur: a claim comment must begin its line with a plain Claim: — no backticks, no bold. Recorded in this seat's errata.

---

Generated by Claude Code

[baozhoutao]

os-dev-report

{
  "status": "done",
  "issue": 8420,
  "branch": "claude/8420-ci-cd-pipeline-command-parity",
  "pr_url": "https://github.com/objectstack-ai/objectui/pull/8878",
  "files_changed": [
    "content/docs/guide/ci-cd-pipeline.md",
    "scripts/__tests__/ci-cd-pipeline-doc.test.ts",
    ".changeset/8420-ci-cd-pipeline-command-parity.md"
  ],
  "acceptance": [
    {
      "leg": "1. Both sections name every first-party command their job runs, and no command it does not run",
      "verdict": "met",
      "evidence": "vi-mock section: ran = {scripts/check-vi-mock-specifiers.mjs, scripts/check-vi-mock-inherit.mjs}; named now covers both plus the local-run aliases. shadcn section: ran = {pnpm shadcn:analyze, pnpm shadcn:check, scripts/shadcn-check-report.mjs}; named = the same three. Both directions assert empty. One neighbouring gate (dependabot-merge-gate.mjs) is cited without its scripts/ path and says why, the convention the Lint section already set."
    },
    {
      "leg": "2. Two commandParity pins, one per section",
      "verdict": "met",
      "evidence": "describe('ci-cd-pipeline.md — the vi-mock and shadcn sections') at the end of scripts/__tests__/ci-cd-pipeline-doc.test.ts: units() builds one commandParity per section, plus a job-key pin per workflow and a control on the alias resolver. Test count 58 -> 62."
    },
    {
      "leg": "3. Ablation per pin, mutate/red/restore/green",
      "verdict": "met",
      "evidence": "Run on the committed tree with a trap-restore, absolute paths and blob-hash proof. A: delete check-vi-mock-inherit from the vi.mock prose (whole-page hits 2 and 1 -> 0 and 0, positive control check-vi-mock-specifiers stays 3) -> exit 1, 'vi-mock-specifiers.yml `vi-mock-specifiers`: pnpm check:vi-mock-inherit'. B: delete pnpm shadcn:analyze (1 -> 0, control shadcn:check stays 1) -> exit 1, 'shadcn-check.yml `check-components`: pnpm shadcn:analyze'. C (phantom direction, extra): credit the shadcn job with scripts/check-doc-links.mjs -> exit 1, 'pnpm docs:check-links'. Every leg restored to HEAD blob 385055f511719d26d19cbfe231139f1dd4632d91 with git diff HEAD empty. Ablation A is also the before-state control: these pins were red on the pre-edit prose, so they are not green for free."
    },
    {
      "leg": "4. Doc test green; repo fast checks green",
      "verdict": "met",
      "evidence": "vitest ci-cd-pipeline-doc.test.ts: 62 passed. vitest scripts/__tests__/: 132 files, 3837 passed, 2 skipped (17 test files in that dir read this page). pnpm lint:root exit 0 over the whole root scan in 9.7s (32 pre-existing warnings, none in a touched file; targeted eslint on the changed test file: 1 file, 0 errors, 0 warnings, and eslint.config.js has no projectService/parserOptions.project so no type-aware verdict can move on an untouched file). pnpm type-check:scripts exit 0. docs:check-links, check:doc-fences, check:shell-escape-residue, check-control-bytes, check-changeset-presence all exit 0."
    },
    {
      "leg": "5. A changeset",
      "verdict": "met",
      "evidence": ".changeset/8420-ci-cd-pipeline-command-parity.md with empty frontmatter. check-changeset-presence.mjs reports no changeset is owed (0 published source files changed), so the empty declaration is the correct first-class form. No edit under content/docs/releases/."
    }
  ],
  "alias_decision": "An alias whose root package.json definition is exactly node plus one scripts/ path names the same gate as invoking that file directly. Written into the test's prose with three explicit limits: aliases carrying arguments are NOT merged (shadcn:check vs shadcn:update are one script in two modes); the decision is applied only to these two units and not to the ci.yml table, the lint.yml section or the other 32 sections; and it does not erase the pre-install placement distinction, which is a fact about step placement rather than about which gate runs. A control test asserts the resolver is unambiguous — measured 51 pure wrappers, 0 collisions — and that it still resolves pnpm check:vi-mock-inherit.",
  "premises_falsified": [
    "None of the dispatch premises failed. All re-verified by content on a fresh origin/main 7fb22a12c (main had moved from 256c709e2). vi-mock-specifiers.yml:96 and :112 hold; the doc sections were at :1166 and :2313; commandParity/undocumentedCommands/phantomCommands at :669/:684/:689. Negative reading with its control: grep for check-vi-mock-inherit over the whole page returned 0 while check-vi-mock-specifiers returned 3.",
    "The card's drift-rate claim did NOT reproduce, as the dispatch predicted: git log origin/main for .github/workflows/vi-mock-specifiers.yml returns exactly 6ea94b80e, 05b832407, e9e55524e; 4b5e07a53 is not among them. The sentence appears nowhere in the PR."
  ],
  "surprises": [
    "The alias question is load-bearing in only ONE of the two sections, which is the opposite of what the dispatch's framing suggests. shadcn-check.yml runs the alias literally and the doc names the alias, so that unit is already at parity under the unmodified module-scope rule. The section that needed the decision is the vi.mock one, where the page's own 'run it locally with pnpm check:vi-mock-specifiers' line would have been reported as a phantom gate — the rule would have cried wolf on the very first section it was pointed at.",
    "Writing scripts/dependabot-merge-gate.mjs into the vi.mock prose (to say why that gate is a required context) made the phantom direction fire, because that script runs in dependabot-auto-merge.yml. Resolved by following the convention the Lint section already documents in-page: name it without its scripts/ path and explain why. Worth knowing that this pin constrains how a section may cite a neighbour.",
    "Scope note for the PM, not a defect in this PR: the PR body carries the ordered 'Fixes #8420', so merging closes a card whose title covers every unpinned section on the page while this PR pins two of them. The remaining sections stay unpinned and, once the card closes, are no longer visible in any open-issue sweep. Flagging rather than acting: filing a successor card would overlap this one, and the closing keyword was the dispatch's instruction.",
    "One instruction conflict, chosen deliberately and reported rather than settled silently: the dispatch forbids a model identifier in any pushed artifact, while this session's harness attribution mandates a 'Co-Authored-By: Claude Opus 5' commit trailer. Kept the harness trailer, because it is session configuration rather than the dispatch's to override and because origin/main's recent history carries that exact trailer on landed commits. Reverse the choice and I will amend before the PR is flipped ready."
  ],
  "follow_ups_filed": [],
  "mcp_calls": 5
}

---

Generated by Claude Code