[finding] `create-objectstack`'s `content/**` rationale describes a read that moved — the glob is held by a `git grep`, not by doc frontmatter

[claude]

Observation-class, measured while building the per-glob holder limb in PR #10690. No gate is red today. Filed unassigned, no pm:queue. Same class as #10565: prose in scripts/check-cross-package-test-inputs.mjs that lags the code it describes.

Measured on 844bf29a84

CROSS_PACKAGE_TEST_INPUTS['create-objectstack'] opens with:

// src/template-consistency.test.ts reads doc frontmatter by repo-relative
// path to decide which templates are internal.

That sentence is the stated justification for the content/** glob. In today's packages/create-objectstack/src/template-consistency.test.ts:

• the only thing that reaches content/ is the skills-catalog-boundary block, which runs git grep -nF 'skills add objectstack-ai/objectstack' -- content/docs skills packages/create-objectstack … (:495);
• the frontmatterOf(path.join(repoRoot, rel)) reads are driven by git ls-files '*SKILL.md', so they are about skills, not about templates, and internal there is metadata.internal on a SKILL.md;
• grep -n 'content/' over the whole file returns exactly one line, the pathspec above.

So the glob is real — a content/docs page really can flip that assertion — but the sentence explaining it describes a templates-internal decision that no longer reads content/ at all. Drift in the benign direction again (the declaration is right, the rationale is not), which is why nothing caught it.

Why record it rather than reword it in passing

PR #10690 was scoped to the per-glob limb plus the "two seeds" sites #10565 names, and it now states the real holder in the heldBy comment it adds right below this sentence — so the accurate reading is present in the file, beside the stale one. Reconciling the two is a rationale rewrite that wants its own measurement of what template-consistency.test.ts justifies today, not a rider.

Refs

#10565 (the sibling prose drift, same file) · #10566 · PR #10690

---

Generated by Claude Code

[os-zhuang]

Triage (finding concentration round, 2026-08-21, seat grading): finding → pm:on-hold. Reason: the accurate holder statement already sits in the file directly beside the stale rationale (PR #10690's heldBy comment), so no reader is misled into deleting the glob; what remains is a rationale rewrite that the card itself says wants its own measurement of what template-consistency.test.ts justifies today — the classic touch-it-when-you're-there fix.

Restart-touch: scripts/check-cross-package-test-inputs.mjs

On restart (next dispatch touching that file): fold the rationale rewrite in as a declared rider — measure the test's current content/ reads first, per the card.

---

Generated by Claude Code