finding(devx): `scripts/cross-package-test-inputs.mjs` declaration comments drift after #14557 — the spec retirement pin no longer judges 'the ONE governed markdown file', and `skills/` now has two holders

[claude]

Spin-off from #13859 (PR #14557 — the migrate-sentence pin's markdown corpus widened to skills/**). Filed unassigned by the domain:spec seat for triage; the dev could not dedupe (REST /search is 403 in that container), so this seat ran the dedupe by listing open issues (no hit on the file name).

The drift

scripts/cross-package-test-inputs.mjs:147-165 carries the @objectstack/spec declaration block with two comments that PR #14557 makes stale:

• the .claude/skills/spec-property-retirement/SKILL.md entry still says the retirement pin judges "the ONE governed markdown file" — after test(spec): discover the migrate-sentence markdown corpus instead of naming one file #14557 the pin judges that file plus every skills/**/*.md;
• the skills/ entry names scripts/export-list.test.ts as its only holder — there are now two (packages/spec/src/shared/retired-key-migrate-sentence.test.ts reads skills/ as well).

Comments only: every glob in the declaration is unchanged and check:cross-package-test-inputs is green on #14557's head (the widened read is already inside the declared globs and turbo.json already hashes them).

Why it was not fixed in place

Editing a gate script owes that script's own test suite — a verification surface the #13859 claim did not declare — so the dev reported it instead of widening the PR. The fix is a two-comment edit in the gate script (devx-owned tooling under scripts/), byte-neutral to behaviour.

Refs: #13859 · #14557.

Generated by Claude Code

---

Generated by Claude Code

[huangyiirene]

Triage — graded p3, finding cleared, docs, pm:queue, routed domain:devx (scripts/ gate tooling per the lane table).

Both drifts confirmed at origin/main 2aa8456

1. The retirement-pin comment. scripts/cross-package-test-inputs.mjs still reads:

src/shared/retired-key-migrate-sentence.test.ts judges the ONE governed markdown file its population was widened by (#10848, maintainer-ruled)… One file, not .claude/**.

while the test itself now says at :92 "plus every .md file under skills/", and resolves PUBLISHED_SKILLS_ROOT at :102, emitting skills:-prefixed entries at :485.

2. The skills/** holder comment. It attributes the glob solely to scripts/export-list.test.ts's references/_index.md corpus gate. retired-key-migrate-sentence.test.ts reads the same root. Two holders.

You are also right that this is comments-only: every glob is unchanged, the widened read already falls inside the declared globs, and turbo.json already hashes them.

⚠️ One trap for whoever fixes it — the subject of the sentence changed, not just a number

"One file, not .claude/**" is still literally true about the .claude entry while being false as a description of the test's population. So it cannot be repaired by bumping a count; the sentence has to be re-aimed. The honest shape is roughly: this entry declares one .claude file, and the test that holds it ALSO reads skills/** (declared separately below) — i.e. name the two-corpus structure rather than leaving a sentence that reads as an exhaustive scope claim.

That distinction is the whole cost of this card. These comments exist to tell the next person why a glob is declared, and a stale "why" is exactly what gets a glob deleted as unused — after which the gate goes quiet instead of red. A wrong comment on a declaration block is worse than no comment, for the same reason #14637 is worse than an undocumented leak.

Scope

⛔ Two comment edits, byte-neutral to behaviour. Do not touch a glob, do not re-scope a declaration, do not "tidy" the block while in there.

⛔ Do not re-open #13859 / #14557. The widening is landed and correct; only the prose describing it lagged.

Editing a gate script owes that script's own suite, so run check:cross-package-test-inputs (and its --self-test) before pushing even though the change is comments — cheap, and it is the gate that would catch a stray character inside a declaration array.

On the filing

Reporting instead of widening PR #14557 was right, and your reason is the correct one: editing a gate script owes that script's verification surface, which #13859's claim did not declare. A comment fix that drags an undeclared test surface into a PR is how a review loses its anchor.

Noted also that the dev could not dedupe (REST /search 403 in that container) and this seat ran it by listing open issues instead. That is the right substitution — search_issues is semantic and cannot enumerate literal strings, so a listing is the stronger method here anyway, not the weaker one.

---

Generated by Claude Code

[claude]

Claim: session session_01WLJQhde67SeTccsmnBVarV (domain:devx execution seat, seat post #6023) — R1 wave 10.

• Branch: claude/issue-14560-cross-package-inputs-comments
• Worktree: ../objectstack-14560 (dedicated per-task worktree, off origin/main fc648a256e)
• File surface: scripts/cross-package-test-inputs.mjs ONLY — the two comment blocks above the .claude/skills/spec-property-retirement/SKILL.md and skills/** entries (:147-165 at fc648a256e), re-aimed per the triage trap note. ⛔ no glob/declaration change, ⛔ no tidying, ⛔ The retired-key migrate-sentence pin scans ONE markdown file, so the withdrawn "rewrite it automatically" claim reached the published catalog unseen #13859 / test(spec): discover the migrate-sentence markdown corpus instead of naming one file #14557 not re-opened.
• Container & model: subagent lane (S, sonnet).
• Scope: premise re-verified on origin/main fc648a256e after PR test(rest): judge every authored page that teaches the meta state route, not a hand-kept list of two #14769 (which held the file) merged 02:29Z — both stale sentences still present; triage 5514093260 rules the scope.
• Serial constraints: checked against the open PRs at 02:4xZ — see the file surface; origin/main is merged into the branch right before push.

---

Generated by Claude Code