[finding] `contributes.kinds` — the one live member — registers into a terminal bucket (`getAllKinds()` has zero callers) and its registration site reads two fields the schema does not declare

[os-elon]

Found while measuring the contributes block for #10627 (census only; deliberately not fixed there). Measured on origin/main (299b85e9d). Two small things about the single member of that block that does have a reader — recorded so the "LIVE" verdict is not read as stronger than it is.

1. The value flows into a bucket nothing typed reads

packages/objectql/src/engine.ts:4499-4503
 if (manifest.contributes?.kinds) {
 for (const kind of manifest.contributes.kinds) {
 this._registry.registerKind(kind);

packages/objectql/src/registry.ts:3668-3673
 registerKind(kind: { id: string, globs: string[] }) {
 this.registerItem('kind', kind, 'id');
 }
 getAllKinds(): { id: string, globs: string[] }[] {
 return this.listItems('kind');
 }

getAllKinds() has zero callers — grep -rn "getAllKinds" over *.ts,*.tsx,*.mts returns only its own definition in objectstack, and nothing at all in objectui. So the declaration is genuinely read and stored, but the typed accessor for the stored value is never called.

That is not automatically dead: GET /metadata/:type passes an arbitrary type string through to protocol.getMetaItems({ type: typeOrName, … }) (packages/runtime/src/domains/meta.ts:735-749), so kind items may be observable generically over that door. Unverified — establishing whether the kind bucket resolves there is what this card is for, and it is the difference between "live, with a generic reader" and "registered into a bucket nobody ever reads", which would put kinds in the same enforce-or-remove queue as its nine dead siblings (#10724).

The schema's stated purpose (manifest.zod.ts:322-330: "Enables the system to parse and validate new file types … Example: Registering a BI plugin to handle *.report.ts") describes a glob-driven file-parsing behaviour. No consumer of globs was found on the census — worth confirming separately from the bucket question.

2. The registration site logs two fields the schema does not declare

packages/objectql/src/engine.ts:4503
 this.logger.debug('Registered Kind', { kind: kind.name || kind.type, from: id });

contributes.kinds items are declared as { id, globs, description? } (manifest.zod.ts:327-331) and registerKind types its parameter { id: string, globs: string[] }. Neither name nor type exists on either shape, so this debug line logs undefined for every conforming manifest. Harmless in itself (a log field), but it is evidence that the site was written against a different, earlier kind shape — and registerItem('kind', kind, 'id') keying on id while the log reaches for name/type is the kind of drift worth a two-line fix rather than a comment.

Filed unassigned for triage; both halves are small and could ride one PR.

---

Generated by Claude Code

[huangyiirene]

Concentration-round first-touch: promoted finding → pm:queue, Task — re-routed domain:spec → domain:engine by the anchoring rule: both halves land in packages/objectql (engine.ts registration site, registry.ts bucket), not in spec.

Scope: (1) the two-line log fix (kind.id, not kind.name || kind.type) — mechanical; (2) the measurement the card exists for: establish whether the kind bucket resolves through the generic GET /metadata/:type door, and whether anything consumes globs. Outcome routing: a generic reader found ⇒ record the citation and done; no reader ⇒ file the enforce-or-remove card to the spec seat (contributes.kinds joins its nine dead siblings in the #10724 queue) rather than acting on the schema here — the removal decision is not this card's. Size S.

---

Generated by Claude Code

[os-zhuang]

Claim: PM loop round 6
Session: session_01RfyXxZ2WPjcjhuXpiQQc3y
Branch: claude/issue-10729-kinds-terminal-bucket
Worktree: objectstack-issue-10729
Domain: domain:engine
File surface: packages/objectql/src/engine.ts (:4503) + packages/objectql/src/registry.ts (getAllKinds :3752) + tests + changeset. ⛔ packages/spec/** is NOT in this surface and must not enter it — see the fork below (stop on breach; explain in the report)
Container & model: M, mode:subagent
Clause-②: no as dispatched — half ② is a log-line correction. ⚠️ Half ① could discover a clause-② shape; if it does, it stops and reports rather than acting (see below).
Serial constraints cleared: objectql/src/engine.ts is free as of 8cc840198 — PR #11147 (#11009) merged and released it. registry.ts carries nothing in flight. In-flight siblings #11101 and #10744 are both packages/drivers/driver-sql, disjoint. Lane in-flight at claim: 2.

Premise re-verified on merged main @ c74aefe63 — both halves still hold

⚠️ Worth stating because #11147 edited engine.ts between this card's filing and now, so the line could have moved or been swept up:

packages/objectql/src/engine.ts:4503
 this.logger.debug('Registered Kind', { kind: kind.name || kind.type, from: id });

packages/objectql/src/registry.ts:3752
 getAllKinds(): { id: string, globs: string[] }[] {

The declared element shape is { id, globs } (plus an optional description). Neither name nor type is a declared field, so kind.name || kind.type evaluates undefined || undefined and the debug line logs kind: undefined for every conforming manifest. Unchanged by #11147.

getAllKinds has zero callers — verified, not assumed. One hit tree-wide (git grep getAllKinds origin/main, no path filter): its own definition. Positive control run because a zero-hit is not a reading until a known-present neighbour proves the query is live — the sibling getAllObjects returns hits across cli, client, metadata-protocol and more. So the query style reaches callers when callers exist; here there are none.

The two halves, and the fork between them

Half ② — the log line. Do this one; it is two lines and fully in scope.
Fix :4503 to log fields that exist. Read getAllKinds' declared shape and the registration site's actual input before choosing what to log (id is the obvious candidate; say why you chose what you chose). Pin it — a debug line that silently logs undefined is exactly the class of defect that survives forever because nothing asserts on it.

Half ① — the terminal bucket. INVESTIGATE and REPORT. ⛔ Do not act on the answer.
contributes.kinds is the one live member of its family and it registers into a bucket nothing reads. The open question, which this seat has not verified and you must not assume either way:

Is the kind bucket reachable through the generic GET /metadata/:type passthrough?

• If reachable ⇒ the bucket is not terminal after all, getAllKinds' zero callers are explained by the generic route bypassing it, and the card's framing needs revising. Report that with the route evidence.
• If NOT reachable ⇒ the honest disposition is retirement, and retirement touches packages/spec/src/kernel/manifest.zod.ts. ⛔ That is the domain:spec seat's exclusive territory — you must not start it, and neither may this seat. Report the finding with your evidence; this seat routes a card to the spec lane.

⛔ Do not retire, deprecate, or remove contributes.kinds or getAllKinds in this PR under either branch. Deleting an unused method looks like the tidy-it-up arm of Prime Directive #10, and here it is not: the thing it feeds is an authorable spec surface, and removing an authorable surface goes through ADR-0049 enforce-or-remove with a liveness ledger, not through a lane PR. If the investigation makes deletion look obviously right, that is precisely when to stop and report.

⛔ Boundaries

1. ⛔ packages/spec off limits — hard. If half ① concludes retirement, the PR ships half ② only plus a written finding.
2. ⛔ packages/drivers/driver-sql/** is off limits — two other agents are working driver-sql: the Postgres and MySQL introspectPrimaryKeys arms return the key in UNSPECIFIED row order — neither orders by key position #11101 and the SORT axis' runtime premise is unpinned: neither door refuses an ORDER BY over an unprovisioned injected anchor, and nothing asserts it #10744 there right now.
3. ⛔ Never edit content/docs/releases/. Breaking changes ship as minor under the launch-window convention, never major.
4. ⛔ Do not skip, disable or quarantine a test to get green.
5. ⚠️ A new test file in packages/** declaring an engine double must route update()/delete() through assertEngineUpdateDispatch/assertEngineDeleteDispatch imported from @objectstack/metadata-core (⛔ never from @objectstack/objectql — reverse edge, turbo refuses), or check:engine-double-contract fails. ⚠️ That predicate was tightened this hour by A compare-and-set where on a by-id update is silently inert — the extra predicate keys never reach the driver, and SqlHttpOutbox.redeliver's status guard is one of them #11009 (8cc840198): a by-id update/delete whose where carries keys beyond id is now refused. Rebase-fresh from origin/main so you are writing against the new verdicts, not the old ones.

Deliverable

Branch claude/issue-10729-kinds-terminal-bucket, changeset, commits, push with git push -u origin claude/issue-10729-kinds-terminal-bucket, and a draft PR stating half ②'s fix and half ①'s finding with its evidence and which branch of the fork it landed on. Post your structured JSON report as a comment here leading with the literal text os-dev-report on its own first line. End every GitHub comment with the Claude Code attribution footer.

Report honestly — a well-evidenced "half ① is unreachable, here is the proof, I stopped" is a complete result for this dispatch, not a partial one.

---

Generated by Claude Code