Remove the `(p as any)` cast at the REST `publishMetaItem` call site once `MetadataProtocol` declares the member (#11006 follow-up)

[os-zhuang]

Blocked-by: #11006

Filed by the domain:engine PM seat as the downstream half of a maintainer ruling, so the follow-up is a card rather than something a seat has to remember. ⛔ Not dispatchable until #11006 closes.

Why this card exists

The maintainer ruled B on #11006 (2026-08-22 live session, 「同意所有」, item 5): MetadataProtocol gains an optional publishMetaItem member. The ruling record assigns the consumer-side cleanup to this lane by name:

follow-up in the engine lane removes the (p as any) cast at the REST call site once landed (the spec PR body names that consumer so the engine seat picks it up)

That sentence was the only record of this work. It lives in a comment on a card in another lane which will close when the spec PR merges — at which point the follow-up becomes invisible to every queue view in this repo. Hence a card.

What the work is

Once #11006's spec change lands, the REST call site's (p as any) cast can come off, because the cast will no longer be carrying anything. That is the measured point from #11006, and it is worth restating precisely because it is counter-intuitive:

删掉 cast 得到的是 TS2339 — Property 'publishMetaItem' does not exist on type 'RestProtocol',而不是 TS2353(对象字面量属性不匹配)。也就是说 cast 承的是「成员存在性」的重,不是「请求形状」的重

So this is not a request-shape fix and adding fields to metadata-protocol's own request type could never have achieved it — the cast exists solely because the interface had no such member. #11006 adds the member; this card removes the prop that was holding the gap closed.

Expected shape after the ruling lands: the cast degrades to the existing optional-call idiom (matching getMetaItemLayered / deleteMetaItem), and with #9741's TransportScopedMetaRequest wrapper applied, an undeclared key at the call site becomes a compile error — which is the entire point of the exercise and the acceptance signal for this card.

Before dispatching, re-verify rather than assume

• Confirm on merged main that [裁决] MetadataProtocol 是否应声明 publishMetaItem 成员 —— 半声明的门（响应已声明、请求未声明） #11006's member actually landed as an optional member with the request type, not some narrowed variant. The scope may have shifted during spec-seat implementation, and this card's premise is exactly that member's existence.
• Establish that removing the cast produces a clean compile, not a different error. If a second, independent reason for the cast surfaces, that is a finding — report it rather than re-casting.
• ⛔ packages/spec is off limits from this lane. If the cleanup needs anything further in spec, it goes back to the domain:spec seat, not into this card's PR.

Labelling note

domain:engine here transcribes the ruling's own routing ("follow-up in the engine lane") rather than being a fresh triage judgement — this seat does not produce domain:*. Triage should correct it freely if the landing point argues otherwise. Size looks S (one cast, one compile check), but the sizing is this seat's guess and not measured.

Related: #11006 (the ruling, upstream) · #10350 (the parent card, landed as PR #11005 — packageId declared, three stale comments corrected) · #9741 (the meta-read sibling and the transport wrapper) · #7294 (declared the response side) · ADR-0076 D9

[claude]

Unlock scan (triage daily round, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24): released to pm:queue — blocker #11006 closed 2026-08-23T18:01Z (completed). Same-repo consumer, not release-gated. Dispatch obligation: re-verify the card's file surface on the merged ref — the #11006 landing (publishMetaItem optional member, ruled B) may have absorbed or reshaped part of this follow-up; if fully absorbed, close with evidence instead of dispatching.

---

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 1
Session: session_01VK8rFDtg8eREaxBGX99Csn
Branch: claude/issue-11145-drop-publishmetaitem-cast
Worktree: objectstack-issue-11145
Domain: domain:engine
File surface: packages/rest/src/rest-server.ts (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus (tier: dispatch-gates.mjs --tier packages/rest/src/rest-server.ts returns no path-derived mandate, quoted not recalled. Floor is sonnet and the compile check is mechanical, but the card carries a judgement branch — "if a second, independent reason for the cast surfaces, that is a finding" — and misjudging that branch is how a cast gets silently re-added. Default judgement tier.)
Clause-②: no — this removes a cast that carried member existence, not request shape. The accept set does not move in either direction; the declared member already shipped in packages/spec. This is declared = enforced catching up on the consumer side.
Serial constraints cleared: ⚠️ PR #11641 is open on this exact file (packages/rest/src/rest-server.ts, card #11546, domain:cli seat). Regions measured disjoint — its hunks are @@ -3622,2 +3622,40 @@ and @@ -3627 +3665 @@, i.e. the openapi.json info.version block at ~3619-3665. This card's door is at 5940 / 6051, ~2300 lines away in a different method. Measured from the diff, ⛔ not from the PR's prose — the region-from-prose mistake is on this seat's record and cost a wrongly-withheld dispatch. Other seats do not stop because this seat holds a file, so the region criterion is the only one that applies to #11641. No other open PR touches this file.

Premise re-verified on the merged ref — NOT absorbed

The unlock scan's dispatch obligation was to re-verify on the merged ref and close with evidence if the #11006 landing absorbed this. It did not. Measured on origin/main @ cad8b42:

• packages/spec/src/api/protocol.zod.ts:2535 declares publishMetaItem?(request: PublishMetaItemRequest): Promise[PublishMetaItemResponse] — an optional member carrying the request type, which is exactly the shape this card's premise requires (ruled B on [裁决] MetadataProtocol 是否应声明 publishMetaItem 成员 —— 半声明的门（响应已声明、请求未声明） #11006, landed via PR feat(spec): declare publishMetaItem as an optional MetadataProtocol member with PublishMetaItemRequest #11426). Not a narrowed variant.
• Both cast sites are still live: (p as any).publishMetaItem at 5940 (the existence check) and 6051 (the call).
• ⚠️ The handover recorded these at 5916 / 6027. They moved. Line numbers are taken fresh at dispatch, never copied from the card.
• Positive control: 166 occurrences of as any in this file, so the zero-elsewhere readings are readings and not a broken matcher.

What the acceptance signal actually is

Per the card, removing the cast should degrade to the existing optional-call idiom used by getMetaItemLayered / deleteMetaItem, and with #9741's TransportScopedMetaRequest wrapper an undeclared key at the call site becomes a compile error. That compile error is the point of the exercise, not a side effect.

⚠️ There is a docblock at ~6029-6031 that exists solely to explain why the cast is there (it names the TS2339 it would otherwise produce). It is falsified by this change. Leaving it behind would be a declaration that outlived its subject — the same defect class as the sibling card #11222 in this batch.

⛔ packages/spec is off limits from this lane. If the cleanup turns out to need anything further in spec, that goes back to the domain:spec seat as its own card — ⛔ not into this PR.

H17 on-hold trigger index — zero intersection

Intersected against all 17 indexed holds on the patrol anchor (#9857). No hit on packages/rest/src/rest-server.ts.

---

Generated by Claude Code