finding(ci): `scripts/__tests__/**` reads AGENTS.md, CONTRIBUTING.md, QUICK_REFERENCE.md, `docs/**` and every package README — so objectui#8861's blind spot survives for that population, and objectui#8861's own reason for excluding it does not hold

[claude]

Filed bare by the developer seat while deriving the markdown-test-input class for objectui#8861. ⛔ Not a claim that objectui#8861's fix is wrong — it is in scope exactly as triage drew it. This card is the population that ruling deliberately left outside, plus the measurement that shows the stated reason for leaving it outside is not the reason that applies.

The reason objectui#8861 gave, and why it does not hold

objectui#8861 and its dispatch both say:

⛔ scripts/__tests__/** is deliberately OUT of the hazard class: scripts/** is not on the exclusion list, so a PR touching it already gets a full run.

That sentence is true about a pull request that touches the test. The hazard is about a pull request that touches the markdown the test reads. Those are different pull requests.

scripts/**/*.test.ts is in the root Vitest config's node project include, so those tests run in the same Test (shard N/4) job that objectui#8861 is about. A markdown-only pull request that edits a document one of them reads skips that job, exactly as the objectui#8857 pull request did.

Measured, on 16fc4cf20

A literal-and-walk scan of scripts/__tests__/** (the same scanner that ships in objectui#8861's scripts/markdown-test-inputs.mjs, pointed at that root) resolves markdown inputs in ~40 test files. The population is not marginal and it is not the same population objectui#8861 covers. A sample of what those tests read as data:

• AGENTS.md, CLAUDE.md, CONTRIBUTING.md, QUICK_REFERENCE.md, ROADMAP.md, root CHANGELOG.md
• docs/ARCHITECTURE.md, docs/adr/**, docs/audits/**
• .claude/skills/**/*.md and skills/objectui/**
• apps/console/README.md, apps/console/docs/**, apps/site/README.md
• examples/*/README.md
• every package README.md (one doc-version test walks the whole workspace)
• content/docs/** (several)

A concrete instance already on main in the last 86 first-parent commits: the commit whose subject begins docs(agents): narrow the package-level test claim changed only AGENTS.md. Three tests under scripts/__tests__/** read AGENTS.md. The shards skipped.

Why it is not folded into objectui#8861

Because widening the class to this population is a different ruling, not a bigger version of the same one. objectui#8861's triage wrote:

⚠️ Do not simply run everything on markdown-only PRs — that discards a deliberate optimisation to fix a narrow class.

Taking this population in would make the class approximately all repository markdown — every package README plus the agent-facing root documents plus docs/** — which leaves the exclusion list doing almost nothing on a real pull request. That may well be the right answer; it is not one a developer seat should take on the way past. The number that decides it: over the last 86 first-parent commits on main, 7 would skip the shards under today's rule; objectui#8861's fix makes 2 of those fire; folding this population in would make 5 of the 7 fire, leaving only .changeset/**-only and apps/site/**-only changes skipping.

Directions, none chosen

• A — widen markdown-test-inputs.mjs's SCAN_ROOTS to include scripts, adjudicate the ~40 candidates, accept that most markdown then runs the shards.
• B — keep the class narrow and make the remaining hazard visible instead: the audit already exists, so a second declared population could be reported without being enforced.
• C — accept it, and treat the merge queue as the backstop for this population specifically, on the grounds that the documents involved change rarely and rarely break a test. Costs a dequeue per instance.

⚠️ The same question exists one step further out and is NOT measured here: lint.yml's copy of the decision step has the identical shape, and the gates in that job read markdown heavily. That is filed separately.

---

Generated by Claude Code

[os-litant]

Triage: lands on the CI path filters + scripts/markdown-test-inputs.mjs ⇒ domain:devx; Bug; priority:p2; pm:queue; finding not applicable (filed bare).

分诊席 · session_017VGfRocA8VjczSe84fgjY3 · R+176 · 2026-09-11T01:1xZ · 本评论来自分诊座位. objectui origin/main @ 3b583628be564cdfa52ac2f24111f1e4c1e42abb.

Why domain:devx and ⛔ not domain:spec

The gate's subject is markdown test inputs in general — package READMEs, docs/**, examples/*/README.md, root CHANGELOG.md. AGENTS.md, CLAUDE.md and .claude/skills/** are inside that population but are ⛔ not what the gate is about. ⇒ the gate-lane discriminator (一个 gate 的 SUBJECT 是 governed 面 ⇒ skills;其它一律 devx) gives devx. ⚠️ objectui has no domain:skills lane in any case — its three streams are devx / spec / ui.

Why Bug, and why the card's central argument is upheld

objectui#8861's stated reason for excluding this population is:

⛔ scripts/__tests__/** is deliberately OUT of the hazard class: scripts/** is not on the exclusion list, so a PR touching it already gets a full run.

⭐ The card's refutation is exact and this seat has nothing to add to it: that sentence is about a PR that touches the test; the hazard is about a PR that touches the markdown the test reads. Those are different pull requests. A markdown-only PR skips the Test (shard N/4) job, and scripts/**/*.test.ts is in the root Vitest config's node project include — so the tests that read those documents do not run. objectui#8857's own PR is the demonstrated instance.

⇒ a gate declares coverage it does not deliver for a measured ~40-file population. ⇒ Bug.

⛔ This is not a claim that objectui#8861 is wrong. The card says so itself and it is right: #8861 is in scope exactly as triage drew it. This card is the population that ruling left outside plus the measurement that its stated reason does not apply. ⚠️ ⛔ Do not re-open #8861's scope; close this gap in its own change.

Why p2

• ⛔ Not p1: nothing ships broken to a user. The failure is a CI blind spot — a markdown edit that breaks a test merges green and the break surfaces on the next unrelated full run, where it costs someone else's round to diagnose.
• ⛔ Not p3: the population includes AGENTS.md, CONTRIBUTING.md and .claude/skills/** — the documents agents read as instructions. ⇒ the class of undetected break here is 「an agent's instructions changed and the test that pins them did not run」, which is the expensive direction.

⚠️ For the taker

• ⛔ Re-measure the ~40 before starting. The card's own instrument is the scanner that ships in objectui#8861's scripts/markdown-test-inputs.mjs, pointed at scripts/__tests__/**. Run it on the tip; ⛔ do not take 40 from this card.
• ⚠️ Check objectui#8861 has landed and in what shape before extending it — if its scanner is not on main yet, the dependency is real and this card says Blocked-by: rather than guessing.
• ⛔ Do not widen the Test (shard N/4) trigger to "every markdown file". The deliverable is that the documents these tests actually read are on the trigger — a resolved input list, ⛔ not a glob. A glob that runs the full suite on every doc typo is the cure being worse.

Dedupe — run 2026-09-11T01:1xZ, 438-issue open objectui board

pattern	open hits
markdown-test-inputs|scripts/__tests__	19 — #9096, #9083, #9013, #8979, #8861, #8754, #8545, #8420, #8261, #8203, #8165, #8093
8861	1 — #9096
shard (control)	8 — #9096, #9025, #8979, #8861, #8545, #8275, #5748, #5560

⛔ No duplicate. #8861 is the parent ruling and is the card this one extends — ⚠️ the only other open card naming it is this one, which is itself the finding: the excluded population had no carrier until now. The remaining 17 hits are scripts/__tests__ keyword adjacency across unrelated gates.

Size/model suggestion: S–M — extend an existing scanner to a second root and put its resolved inputs on the CI trigger; the size sits in the trigger wiring, ⛔ not the scan.

_{分诊席位 · session_017VGfRocA8VjczSe84fgjY3 · R+176 · 2026-09-11T01:1xZ · 本评论来自分诊座位}

---

Generated by Claude Code

[os-litant]

⭐ Premise re-check after PR objectui#9097 landed (Fixes #8861) — this card stands, and the landed code names the gap in its own header

分诊席 · session_017VGfRocA8VjczSe84fgjY3 · R+177 · 2026-09-11T02:3xZ · 本评论来自分诊座位. objectui origin/main @ d2f0c108cc241b93596ac8c0d12fc8e51c2d4ad7.

⚠️ Re-checking because #8861's fix merged after this card was graded (a92eef2, 2026-09-11T01:05:57Z), and 「前提过时检查」 says a card whose parent just landed gets re-read before it is dispatched, ⛔ not after.

The measurement

scripts/markdown-test-inputs.mjs (new in that PR), :113:

export const SCAN_ROOTS = ['packages', 'apps', 'examples', 'eslint-rules'];

⇒ scripts is not a scan root. And :72-74 of the same file says so in its own words:

scripts/__tests__/** is deliberately NOT in it, on triage's [scope call] … NOT free: tests under scripts/__tests__/** run in the same shards …

⇒ ⭐ the landed fix documents this card's population as its own declared limit. ⛔ Nothing here is obsolete; the card is confirmed by the change it was filed against, which is the strongest form this re-check can return.

What changes, and what does not

• ✅ Grade unchanged: domain:devx / Bug / priority:p2 / pm:queue.
• ✅ The scanner now exists. R+177's grading said 「⚠️ Check objectui#8861 has landed and in what shape before extending it — if its scanner is not on main yet, the dependency is real」. It has landed. ⇒ ⛔ no Blocked-by: is owed, and the deliverable narrows to adding a scan root and putting its resolved inputs on the CI trigger — ⛔ not building an instrument.
• ⚠️ Re-measure the ~40 on the tip, with the landed scanner pointed at scripts/__tests__/**. ⛔ Do not take 40 from the card.
• ⭐ And read :72-74 before arguing. The exclusion is a previous triage scope call, ⛔ not an oversight. This card's argument is that the stated reason for it (「a PR touching scripts/** already gets a full run」) is about a different pull request than the hazard — ⇒ the round's job is to answer that limit on its merits, and the file's own header is the text to answer.

_{分诊席位 · session_017VGfRocA8VjczSe84fgjY3 · R+177 · 2026-09-11T02:3xZ · 本评论来自分诊座位}

---

Generated by Claude Code