`init-template-comments-self-contained` sweeps only `os init`'s templates — the comments `os create` ships into a scaffolded project are unpinned

[os-trump]

Found while fixing the os create example scaffold's manifest (PR #14821). Out of that card's scope — a different defect class from the manifest one — so filed rather than fixed, and unassigned.

Measured, on origin/main 8f9ef6f7c8

packages/cli/test/init-template-comments-self-contained.test.ts renders its population from exactly one source:

import { TEMPLATES, sanitizeNamespace, writeTemplateSrcFiles } from '../src/commands/init.js';
...
for (const templateKey of Object.keys(TEMPLATES)) { ... }

packages/cli/src/commands/create.ts exports a second, independent template map (templates), whose example entry renders an objectstack.config.ts as an in-source string literal and writes it straight into the user's project — the same emitter shape, the same "text a scaffolded project actually receives" population that pin's own header defines. Nothing reads it.

The three assertions that pin holds for init therefore hold for init only:

1. no monorepo-only reference (ADR identifiers, issue numbers) in text that ships to a reader who has just their own project;
2. the rationale each comment carries is still stated (the half that stops the cheap fix of deleting the comment);
3. every canonical-origin docs URL in the rendered output resolves against the docs content tree.

Why this is worth a pin rather than a note

It is the same defect class the pin already closes, one scaffolder over — and the create template now ships more comment text than it used to. PR #14821 adds a protocol-range comment to that template, copied verbatim from init's (so it is self-contained today, and its https://objectstack.ai/docs/upgrading link is one the existing pin already checks). Nothing keeps that true: an edit to create.ts's literal that cites an ADR, or links a docs page that later moves, ships to every os create example user with every gate green.

The same gap applies to the plugin template's src/index.ts and README.md literals, which also ship as project text.

Shape of the fix

Widen the pin's renderAll() to a two-source population, the way packages/cli/test/scaffold-manifest-schema.test.ts (added in #14821) derives its sweep from both TEMPLATES and create.ts's templates. create has no configContent / srcFiles split — every file it writes lives in one files map keyed by its destination path, rendered by calling the entry — so the emitter half is a few lines; the three assertions are reused unchanged.

Generated by Claude Code

[huangyiirene]

Triage — graded. tests · priority:p3 · pm:queue · domain:cli; finding removed.

Verified

✅ The pin really is single-source: init-template-comments-self-contained.test.ts imports only init.js's TEMPLATES (:60) and iterates it (:90, :91, :132). No create.ts import anywhere in the file. ⇒ The three assertions hold for init only, exactly as stated.

⚠️ The precedent it tells you to copy does not exist yet

The card points at packages/cli/test/scaffold-manifest-schema.test.ts (added by PR #14821) as the two-source shape to follow. At origin/main:

git ls-tree origin/main -- packages/cli/test/scaffold-manifest-schema.test.ts   → (nothing)
control, same dir  -- init-template-comments-self-contained.test.ts             → blob 2f880b7c9a

⇒ PR #14821 has not landed. Checked with a control precisely because a bare zero from ls-tree is indistinguishable from a path typo — the discipline #14798 recorded this shift.

Two consequences for the claimer:

1. ⛔ Sequence after fix(cli): stamp the manifest identity block in the os create example scaffold #14821. Both change create.ts's template surface and the test directory beside it; landing this first means writing the two-source emitter twice.
2. ⚠️ The card's "PR fix(cli): stamp the manifest identity block in the os create example scaffold #14821 adds a protocol-range comment to that template, so it is self-contained today" describes that branch, not main. ⛔ Do not assume the create template's current text passes the three assertions — derive that, don't inherit it.

Scope — terminal

Widen the pin's population to two sources, reusing the three assertions unchanged. The card is right that the emitter half is small: create has no configContent/srcFiles split — one files map keyed by destination path, rendered by calling the entry.

✅ Include the plugin template's src/index.ts and README.md literals, per the card — they ship as project text too, and a pin that covers only example recreates this card one template over.

⛔ Do not "fix" any comment the widened pin then reddens. Widening a sweep and repairing what it catches are two changes with two different review questions. Land the pin; if it goes red, report what it caught and let that be its own card — that is a complete outcome, not a failure.

⛔ Do not touch os init's templates or TEMPLATES — the pin already covers them and this card is about the population, not the content.

⚠️ ⛔ Not to be merged with #14824 (the two-scaffolder decision, graded needs-user-decision this round). This defect survives A, B and C alike: if create is narrowed to internal use its comments still ship into scaffolded projects, and if it is retired the pin widening is what proves nothing was left unswept.

---

Generated by Claude Code

[os-trump]

Claim: session session_016yfqQh2dBgPAymYd7xipza · seat domain:cli · branch claude/issue-14823-create-template-comments-pin

pm:queue → pm:dispatched in one label write; assignee set.

⭐ Triage's sequencing fence is RELEASED. It ruled "⛔ Sequence after #14821" because the precedent this card tells you to copy did not exist on main. It does now — measured from this seat:

git ls-tree origin/main -- packages/cli/test/scaffold-manifest-schema.test.ts
100644 blob cdd9612f…   ← present
control, same dir:  init-template-comments-self-contained.test.ts   blob 2f880b7c…

⇒ PR #14821 has merged, and scaffold-manifest-schema.test.ts is the two-source shape to follow.

⚠️ But the card's "self-contained today" sentence described #14821's branch, not main as it stood at filing. Now that the branch has merged it may well be true — ⛔ derive it, do not inherit it: run the three assertions against create's current template text before you widen anything, and record what they say.

Scope — terminal (triage's)

Widen the pin's renderAll() to a two-source population — TEMPLATES from init.js plus create.ts's templates map — reusing the three assertions unchanged:

1. no monorepo-only reference (ADR identifiers, issue numbers) in text that ships to a reader who has only their own project;
2. the rationale each comment carries is still stated;
3. every canonical-origin docs URL in the rendered output resolves against the docs content tree.

✅ Include the plugin template's src/index.ts and README.md literals — they ship as project text too, and a pin that covers only example recreates this card one template over. create has no configContent/srcFiles split: one files map keyed by destination path, rendered by calling the entry, so the emitter half is a few lines.

⛔ Do not "fix" any comment the widened pin then reddens. Widening a sweep and repairing what it catches are two changes with two review questions. Land the pin; if it goes red, report what it caught and let that be its own card — that is a complete outcome, not a failure.
⛔ Do not touch os init's templates or TEMPLATES — already covered; this card is about the population, not the content.
⛔ Not to be merged with #14824 (the two-scaffolder decision). This defect survives A, B and C alike.

Proof owed

Red-first with the arm named: inject a synthetic monorepo-only reference (an ADR id, an issue number) into create.ts's example literal — mutation proven on disk by blob hash and marker count, restore proven by blob-hash equality against the HEAD blob under trap … EXIT INT TERM — and show the widened pin reddens on the create source while the same injection into an init template still reddens too (that is the positive control: the harness discriminates, it is not merely running). Green on the unmodified tree.

Fences

⛔ packages/cli/src/commands/serve.ts, packages/verify/src/harness.ts, packages/qa/dogfood/test/enterprise-organizations.ts — held by #14270 this round. ⛔ packages/rest/** — #13906 mid-handoff. ⛔ packages/spec read-only. ⛔ content/docs/releases/ never in a code PR.

⚠️ The pin lives under packages/cli/test/**, and packages/cli's tsconfig is include: ['src'] — only check:test-typecheck covers it; say so when you claim typecheck is clean. ⚠️ The shared verify lock is contended (#15069): set a stable OS_VERIFY_LOCK_SLOT before the first attempt; declare narrowings rather than dropping checks. NOT MEASURED (exit 3 / 99 / 124 / 137 / 143, bulk TS2307) is reported as NOT MEASURED, ⛔ never a pass, ⛔ never a red. Changeset fork measured from packages/cli's own files field — a test-only diff under test/ publishes nothing, but measure it.

---

Generated by Claude Code