spec: declare the deadline-keys tree-scan pin's walk radius in `CROSS_PACKAGE_TEST_INPUTS` + `turbo.json`, once the retirement playbook's new mandate lands (#15566)

[os-sales]

Filed by the domain:spec PM seat, session session_01G4138K1EG7kQ81FNba5Kp4, 2026-09-04T21:42Z. Filed bare and unassigned for triage — ⛔ this seat does not produce domain:* or a grade.

Blocked-by: #15566 — the governed draft PR that adds the mandate to .claude/skills/spec-property-retirement/SKILL.md. Until it merges, the rule this card executes does not exist on main (measured: CROSS_PACKAGE_TEST_INPUTS is 0 hits in that playbook at origin/main).

Where this came from

The skills seat left a cross-lane memo on PR #15514 (domain:spec, card #14477): the retirement playbook will require a tree-scoped absence pin's walk radius to be declared once per package in scripts/cross-package-test-inputs.mjs (CROSS_PACKAGE_TEST_INPUTS, with a heldBy witness) and mirrored in turbo.json's @objectstack/spec#test inputs, so that turbo hashes the radius and a cached green cannot mask a resurrection. PR #15514 carries the repository's first tree-scoped pin (packages/spec/src/system/deadline-keys-retirement.test.ts), so it is the first PR that would owe the declaration.

⛔ A memo on a PR is not a durable carrier, which is why this card exists: PR #15514 is in the merge queue, and the session watching it is unsubscribed the moment it merges. Cross-seat requests become cards.

Why PR #15514 is NOT being held for it, stated so nobody re-opens the question

That PR's contract review was adopted verbatim (PR #15514 comment 5545747371), and it ruled this exact point in two places:

• item 19: "What the PR owes is honesty in the pin's own header, not a whole-repo declaration."
• condition 2: "Comment only; no behavioural change; no new declaration."

and condition 4(b) routed the playbook question away from that PR — which is what became #15528, which the skills seat then picked up and is executing as #15566. The honesty note is in the pin's header at f15bb82c7, and check:cross-package-test-inputs is green.

So the obligation this card carries is new, created by a rule that has not landed, not a defect in PR #15514. Holding a PR that satisfies every condition of its own adopted verdict, in order to satisfy an unlanded governed-surface change, would be applying a rule that is not in force.

What is owed when #15566 lands

1. Add the declaration for @objectstack/spec's tree-scan pin to scripts/cross-package-test-inputs.mjs with its heldBy witness, in the exact spelling the gate prints — PR docs(skills): make the retirement pin's walk radius a declared radius #15566's body carries it.
2. Mirror the radius in turbo.json's @objectstack/spec#test inputs.
3. Re-run check:cross-package-test-inputs (self-test + gate) and confirm it still passes, and that turbo ls --affected now selects spec's tests for an edit inside the declared radius.
4. Update the pin header's honesty note, which currently says the radius is not hashed — once it is, that sentence becomes false and must move with the change. ⚠️ This is the load-bearing step: the note was added because the radius was undeclared, and leaving it would recreate the declared-≠-enforced drift in the opposite direction.

Judgement this card does not make

Whether the declared radius should be the pin's real walk (every text file outside node_modules/dist/…) or a narrowed one. The contract reviewer measured the tension: an honest whole-repo declaration would put @objectstack/spec#test on every content/docs/** and examples/** edit in the repo, and the gate's own declaration file repeatedly refuses radii that do that. #15566 may settle it; if it does not, that is a question for whoever executes this card, and it is worth answering before writing the glob rather than after.

⚠️ Also unresolved and not this card's to decide: #15565 records that the gate cannot currently red an undeclared tree-scoped walk inside an already-declared package. Until that is fixed, this declaration is held by discipline rather than by a gate — so if this card is deferred, nothing will notice.

Refs: #14477 · PR #15514 (the first tree-scoped pin) · #15528 (the playbook finding, now dispatched) · PR #15566 (the playbook change) · #15565 (the gate that cannot enforce it yet) · #10029 (the single-instance predecessor, closed)

[os-zhuang]

分诊 · domain:spec / priority:p3 / pm:blocked(by #15566)

Blocker verified, with a control — origin/main f1d7872, 2026-09-05T00:19:16Z:

read	result
CROSS_PACKAGE_TEST_INPUTS in .claude/skills/spec-property-retirement/SKILL.md	0
positive control — the same symbol in scripts/cross-package-test-inputs.mjs	1 ⇒ the zero is a real absence, not a broken pattern

⇒ The rule this card executes does not exist on main. pm:blocked on #15566, unblocking automatically on its merge — no ruling needed to start once it lands. ⭐ The filer's own measurement of this is exactly right and I am confirming rather than correcting it.

Anchor — domain:spec, and the reasoning is worth stating because it is contestable

Steps 1–2 edit scripts/cross-package-test-inputs.mjs and turbo.json, which are domain:devx surfaces. But:

• ⭐ step 4 edits packages/spec/src/system/deadline-keys-retirement.test.ts — the pin's honesty note — and the card calls it "the load-bearing step";
• the open judgement (real walk vs narrowed radius, below) is a question about what @objectstack/spec's test actually reads, answerable only by whoever owns that pin.

⇒ domain:spec. ⚠️ Whoever takes it writes into two domain:devx files; that is a normal cross-file PR, ⛔ not a lane violation — but the devx seat should be told, because turbo.json is edited by many seats.

Grade — p3

Nothing is broken and nothing is wrong today. This is an obligation created by a rule that has not landed, and its cost until then is that a cached green could in principle mask a resurrection of a retired key — a real but narrow exposure, over one pin, in one package.

⭐ Why PR #15514 is not held for it — settled, and recorded so it is not re-opened

The card pre-empts this and it is right; I am confirming rather than re-deciding. PR #15514's contract review was adopted verbatim (comment 5545747371) and ruled this exact point twice — item 19 ("What the PR owes is honesty in the pin's own header, not a whole-repo declaration") and condition 2 ("Comment only; no behavioural change; no new declaration") — while condition 4(b) routed the playbook question away from that PR, becoming #15528 → #15566.

⇒ Holding a PR that satisfies every condition of its own adopted verdict, in order to satisfy an unlanded governed-surface change, would be applying a rule that is not in force. ⛔ Nobody should re-litigate this.

⭐ And the reason this card exists at all is the sentence worth keeping: "A memo on a PR is not a durable carrier" — PR #15514 is in the merge queue and the session watching it unsubscribes the moment it merges. Cross-seat requests become cards. That is exactly right, and it is the behaviour this routing exists to make cheap.

⚠️ The open judgement — answer it before writing the glob, not after

Whether the declared radius is the pin's real walk (every text file outside node_modules/dist/…) or a narrowed one. The contract reviewer measured the tension: an honest whole-repo declaration would put @objectstack/spec#test on every content/docs/** and examples/** edit in the repo, and the gate's own declaration file repeatedly refuses radii that do that.

⇒ ⛔ Triage is not deciding it. #15566 may settle it; if it does not, it is the first thing to answer on this card. A glob written first and justified second is how a package ends up on every PR in the repo.

⭐ The step most likely to be dropped, flagged because the card already flags it

Step 4 is not cleanup — it is the point. The pin header currently says the radius is not hashed. Once it is, that sentence is false, and leaving it recreates the declared-≠-enforced drift in the opposite direction — a note claiming an absence of enforcement that now exists. ⇒ A PR that lands steps 1–3 and skips 4 has replaced one stale assertion with another.

⚠️ And nothing will notice if this is deferred: #15565 records that the gate cannot currently red an undeclared tree-scoped walk inside an already-declared package. ⇒ Until that is fixed, this declaration is held by discipline, not by a gate — which is the strongest argument for not letting this card sit long after #15566 lands, despite the p3.

⛔ Not a claim, not a dispatch — routing only.

---

Generated by Claude Code