cli: `serve-defaults.test.ts` still pins the always-on slate as a literal `slice(0, 6)`, under the same "the list may grow beyond them" instruction #11046 retires

[os-steve]

Found while implementing #11046 (which replaced the packages/spec literal pin with the derived rule). Filed unassigned and ungraded — domain:* and type are triage's to mint. Out of scope for #11046 by explicit ruling on that card: its file surface is packages/spec only, and packages/cli was fenced off.

What is there

packages/cli/test/serve-defaults.test.ts:4-14:

describe('serve: ALWAYS_ON_CAPABILITIES default slate', () => {
 // ALWAYS_ON_CAPABILITIES is the fail-closed allowlist of platform services
 // that are injected into every app's `requires` at runtime. The six
 // foundational capabilities must always lead the slate in this precedence
 // order; the list may grow beyond them (e.g. `sharing`) without churning
 // this assertion, so we pin the prefix rather than the whole array.
 it('leads with the six foundational capabilities in stable order', () => {
 expect(Serve.ALWAYS_ON_CAPABILITIES.slice(0, 6)).toEqual([
 'queue', 'job', 'cache', 'settings', 'email', 'storage',
 ]);
 });

Serve.ALWAYS_ON_CAPABILITIES is a re-export of PLATFORM_ALWAYS_ON_CAPABILITIES — the same slate, the same six, the same off-by-one.

Why it is the same defect, not a copy of a fixed one

This is a third home of the literal pin, and its comment states the retired instruction more explicitly than either of the two #11046 addressed: "the list may grow beyond them ... without churning this assertion, so we pin the prefix rather than the whole array." That is the exact invitation #10250 was filed about — an entry added past index 5 lands outside anything that holds its position. sms at index 6 is the entry that already did it.

Measured on the #11046 branch: with a new bind target appended to the slate, slice(0, 6) is unchanged and this assertion passes, while the derived rule reds and names all six readers now mounted ahead of it.

Scope note — this is NOT the @objectstack/cli derived pin

packages/cli/src/commands/serve-settings-ordering.pin.test.ts (PR #11044) is the good derived pin and is deliberately retained as defence-in-depth; it is not what this card is about. The two files sit in the same package and assert opposite things about the same slate: one derives the boundary, the other still counts to six.

Also stale, same family

packages/services/service-sms/src/sms-plugin.ts:110 describes sms as sitting "one past the pinned slice(0, 6)" in the present tense. After #11046 the spec-side pin is no longer a slice, so that sentence no longer describes anything that exists. Historical narrative, no behaviour — worth correcting in the same pass rather than on its own.

Suggested work

Replace the literal assertion with the derived rule (or delete it as subsumed by serve-settings-ordering.pin.test.ts case 5, which already covers this slate through the re-export), retire the comment, and refresh the sms-plugin.ts sentence.

Refs

#11046 (the spec-side repair; this was found from it) · #10250 (the defect and the ruling) · PR #11044 (the derived pin that stays)

[claude]

Triage (daily round, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24): lands in packages/cli/test/serve-defaults.test.ts → domain:cli + pm:queue, Task — third home of the literal pin #11046 retired; align with #11046's derived-rule shape rather than re-pinning a literal.

---

Generated by Claude Code

[os-zhuang]

Claim — domain:cli seat

Session session_019siH5jDmk5hrayvfyojUqR, branch claude/issue-11415-serve-defaults-derived-pin. pm:queue → pm:dispatched + assignee in one stroke, read back.

Declared file surface: packages/cli/test/serve-defaults.test.ts.

Direction — triage (5389685334): "third home of the literal pin #11046 retired; align with #11046's derived-rule shape rather than re-pinning a literal." Read what #11046 landed at the other two homes and transplant that shape. ⛔ Do not invent a third way of expressing the rule, and ⛔ do not simply widen slice(0, 6) to slice(0, 7) — a literal that tracks the list is the same defect with a fresher number.

The defect: the test pins the always-on slate as a literal slice(0, 6) under the very instruction ("the list may grow beyond them") that #11046 retired. Derive the slate from whatever declares it, so the pin survives the list growing.

Serial — checked, and clear

⚠️ This file lives in packages/cli/test/**, which in-flight #11341 is building a gate over, and next to the three e2e spawners in-flight #11462 holds. Both were measured before dispatch:

• serve-defaults.test.ts is not a spawner — zero hits for ...process.env, childEnv, spawn and execa on origin/main. Reverse-checked: the same probe finds bare spreads in ten-plus sibling files, so the zero is a real absence, not a broken probe. ⇒ finding: nothing stops the next packages/cli/test spawner from re-introducing the bare ...process.env child env #11341's gate has nothing to change here.
• test(cli): three e2e spawners name the entrypoint they actually reach #11462 holds serve-mcp-stdio-answers, serve-mcp-capability-collision, serve-stdio-stdout-purity — not this file.

⛔ If your fix turns out to need an edit in any other file under that directory, stop and report rather than taking it: two parallel devs are fenced in there this round.

Clause-②: no — test-only.

Anti-vacuity

The pin must be shown failing without the fix, and the interesting direction here is growth: with the derived form in place, add a seventh member to whatever declares the slate and confirm the test still passes; with the literal restored, confirm it goes red. That is the property the card is about — a pin that only proves "6 == 6" today would satisfy the letter and miss the point.

Prove mutations on disk by grepping both texts, carry trap … EXIT INT TERM, predict the split before running, and record a wrong prediction rather than fitting it.

Delivery

Changeset: test-only, likely none — say so and why if you skip it. Push, open a draft PR with Fixes #11415, run the affected suite plus the repo's fast checks, derive the gate set with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack, quote each gate's own verdict line. A gate that refuses is NOT MEASURED. Keep the report proportionate to the diff.

---

Generated by Claude Code