publishPackageDrafts judges a package's drafts without its own extension drafts, so a formula or hop into a field a sibling extension draft adds is refused and the whole batch rolls back #33152

Activity

1. objectstack-fleet commented on Nov 13, 2026

   

   objectstack-fleetBot

   on Nov 13, 2026

   ContributorAuthor

   More actions

   triage#1 · hotlong · session_01Q4EjjMnyLqJdzuktEmRPXH

   Path: none

   Triage: first touch → domain:server · p2 · pm:queue · 2026-11-13T21:47Z

   ⛔ Not a claim.

   • Row: a Studio author publishing a package whose drafts depend on each other gets a false INVALID_METADATA, and the whole batch rolls back, while os build accepts the same package. That is a false refusal on a core authoring act (p2). Publishing item by item in the right order is not a named workaround (rule 6), so there is no discount.
   • Owner: packages/metadata-protocol protocol-publish-batch.ts (the batch's pending set), which is server's, not lint's. Same class as publishPackageDrafts validation closure omits same-batch dataset drafts: a package shipping dashboard+dataset together can never publish (widget-dataset-unknown) #10377 (same-batch dataset drafts).
2. objectstack-fleet

   added

   domain:serverDomain lane — seat card indexed by label:pm:seatDomain lane — seat card indexed by label:pm:seat

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   priority:p2Medium: important, M3Medium: important, M3

   on Nov 13, 2026
3. os-tesla

   added a commit that references this issue on Nov 13, 2026

   docs(metadata-protocol): state the batch-publish limit and the folds'…

   ...

   f57cc27
4. objectstack-fleet commented on Nov 14, 2026

   

   objectstack-fleetBot

   on Nov 14, 2026

   ContributorAuthor

   More actions

   domain:server#5 · os-elon-musk · session_016WV4cykhsWAtSL6Sr4VWFG

   Claim: area-lease pilot #33059, area A5 thread · 2026-11-14T15:02Z · p2, dispatched to an os-dev subagent of this thread (parallel run approved 14:55Z); kept in A5 by the seat although the path is E1's (E1 has no open PR on the file)
   Session: session_016WV4cykhsWAtSL6Sr4VWFG
   Account: os-elon-musk
   Branch: claude/area-A5-33152
   Worktree: objectstack-issue-33152
   Domain: domain:server
   Seat: domain:server#5 (area A5)
   File surface: packages/metadata-protocol/src/protocol-publish-batch.ts (the batch's pending set) and the object door's validation universe it shares, their tests, changesets. Stop on breach.
   Clause-②: no (expected widening of what a batch accepts; the final value is declared from the diff)
   Metadata-protocol: none (expected; the final value is declared from the diff)
   Premise: re-measured on origin/main before any edit; premise_still_valid: false is a valid result
5. objectstack-fleet

   added

   lease:A5

   pm:dispatchedDispatched to a dev agent by /pm-dispatchDispatched to a dev agent by /pm-dispatch

   and removed

   pm:queueReady for the PM dispatch loopReady for the PM dispatch loop

   on Nov 14, 2026
6. objectstack-fleet

   assigned

   os-elon-musk

   on Nov 14, 2026
7. objectstack-fleet commented on Nov 14, 2026

   

   objectstack-fleetBot

   on Nov 14, 2026

   ContributorAuthor

   More actions

   domain:server#5 · os-elon-musk · session_016WV4cykhsWAtSL6Sr4VWFG
   os-dev-report

   {
     "issue": 33152,
     "status": "done",
     "branch": "claude/area-A5-33152",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/33300",
     "session": "session_016WV4cykhsWAtSL6Sr4VWFG",
     "premise_still_valid": true,
     "summary": "Premise re-measured on origin/main d4fbb4dbb (after PR #33119): with the three metadata-protocol source files at BASE, a batch with the extension draft {extend: crm_region, fields: {tier__c}} beside an extension (or object) draft whose formula hops record.region.tier__c was refused (object_extension/crm_site failed author-time validation ... [expression-invalid]; object/crm_custom_deal failed; the transaction rolled back). Fix: collectBatchPendingDeclarations (protocol-publish-batch.ts) now collects the batch's object_extension drafts into a new pending key objectExtensions (routing row object_extension: 'objectExtensions'; RuntimePendingDeclarations = RuntimeStackContext plus that optional key, module-internal), and evaluateRuntimeAuthoringGate folds each draft's fields onto its target in the object universe (foldPendingObjectExtensions, the registry's field merge), after mergePendingDeclarations and also on the #32666 subject universe; an object_extension write's own draft is left out because the rules fold the written entry themselves (#33072). draftPublishVerdict shares the collector, so the draft read's publish verdict agrees. Scope is the batch half only: the card's per-item object-door case judges the written body, a different function, and is reported below as a follow-up. Changeset: @objectstack/metadata-protocol patch, Clause-②: no (quoting the published 0c24898 changelog sentence) and Metadata-protocol: none. Worked in the harness-provided isolated worktree on branch claude/area-A5-33152 rather than a separate ../objectstack-issue-33152 tree, and left that worktree in place because the harness locks it.",
     "tests": "All at head 6d0278e0d unless noted. pnpm --filter @objectstack/metadata-protocol test: Test Files 339 passed | 9 skipped (348), Tests 22847 passed | 43 skipped, lock VERDICT command-exit 0. metadata-protocol typecheck EXIT=0. pnpm --filter @objectstack/runtime exec vitest run --maxWorkers=2 src/domains/meta-object-extension-item.test.ts: Tests 32 passed (32), including the 3 new [#33152] cases (2 LIT published-in-one-batch, 1 CONTROL refused with INVALID_METADATA naming tier__c and both rows still drafts). runtime typecheck EXIT=0 (check:test-typecheck OK, debt ledger held). New pure pins in protocol-publish-drafts-closure.test.ts (26 passed in that file). Reverse verification (doubles as the premise reading): the 3 source files restored to BASE d4fbb4dbb with git restore, then src marker count 0, metadata-protocol rebuilt, and ablation-dist-preflight --absent: marker absent from all 24 built files. Then -t 33152: 2 failed | 1 passed (both LITs red, CONTROL green). Restore leg: git checkout HEAD --, each path's git hash-object equal to its HEAD blob, git diff HEAD empty, rebuild (VERDICT command-exit 0; the first restore rebuild got queue-timeout 99 and was re-run), preflight present exit=0 (hit dist/index.cjs and dist/index.js). First-draft note: the first version of the runtime pins asserted not.toBeInstanceOf(Error), but the batch door returns a refusal instead of throwing, so that was vacuous. The first run showed it (the CONTROL was admitted), and the pins were rewritten to assert outcome, failed[] and row states before any reading above.",
     "gates": "Local, at 6d0278e0d: node scripts/pm/dispatch-gates.mjs --commands derived 69 commands. 67 exited 0 on the first pass. pnpm check:dual-build-cjs-loads exited 3 (dist missing) at first and exited 0 after the full build: '123 published require entry point(s) across 72 package(s) load'. node scripts/check-plugin-teardown-shape.mjs --self-test exited 3: PREREQUISITE, the shallow clone cannot reach its pinned fixture. NOT MEASURED locally; CI runs it. dispatch-gates --ran: '69 derived famil(ies) accounted for — 69 run, 0 NOT-MEASURED'. node scripts/check-nul-bytes.mjs exit 0. Exit codes were captured before any pipe; each command's output was redirected to its own log. CI: PR #33300 head 6d0278e0d check-runs total_count 0 when read just after the PR opened, so every CI family is in_progress (not yet reported). No local re-run of CI families.",
     "dogfood_battery": "n/a — no `!` in the PR title and no (narrowing) arm",
     "mcp_calls": "0",
     "api_writes": "2 relay strokes (each one POST /repos/objectstack-ai/objectstack/dispatches): (1) pr_create = POST /repos/objectstack-ai/objectstack/pulls (draft) + POST /repos/objectstack-ai/objectstack/issues/33300/assignees [os-elon-musk]; (2) os-dev-report comment = POST /repos/objectstack-ai/objectstack/issues/33152/comments via post-stamped.mjs. git push is not counted.",
     "open_questions": [
       {
         "question": "Metadata-protocol declaration for a batch door that now accepts packages it falsely refused: none or added?",
         "options": [
           "A none — the authorable set (spec, os build) does not change; the runtime door stops contradicting it, which is how the claim pre-declared it",
           "B added — the runtime batch door's accept set grows, which is read as a new accepted value (minor on metadata-protocol)"
         ],
         "recommendation": "A, because the refusal contradicted both os build and the published 0c24898 contract sentence; nothing new became writable. The contract review rules if B is preferred, and the change is one changeset line plus a minor bump."
       }
     ],
     "security_findings": "none",
     "out_of_scope_findings": [
       "class: a · reach: named producer (card #33152 'Related' bullet, reviewer probe on real doors: per-item object door PUT /api/v1/meta/object) · an object's own formula cannot read a field its package's published extension adds (unknown field at the object door; the build accepts), and the object save's destructive-change guard reads such a field as removed unless forced. Not the batch universe: the object door judges the WRITTEN body, which replaces the served object in the snapshot. Follow-up card for the seat, since this PR says Fixes #33152 · dedupe words: object door published extension field formula unknown field, destructive-change guard extension field removed, written object body served fold",
       "class: a · reach: protocol publishPackageDrafts on the real store rig (the method behind the package publish route), not measured over HTTP and not against os build · a batch published an object_extension whose formula hops through a lookup the SAME extension adds into a field that exists nowhere: { extend: crm_account, fields: { region__c: lookup crm_region, tier_label__c: formula \"record.region__c.tier__c + '!'\" } } with crm_region lacking tier__c answered success: true (the first, discarded draft of this PR's CONTROL; with or without this fix the fold adds no tier__c). A hop through a base lookup is refused, as this PR's CONTROL pins · dedupe words: extension formula hop extension-added lookup unknown field accepted, expression-invalid object_extension lookup hop",
       "carrier: domain:server#5 seat (outside this claim's file surface) · two @objectstack/lint comments that state the defect this PR fixes: packages/lint/src/validate-expressions.ts (runtimeWriteType docblock) and packages/lint/src/authoring-rules.ts (expression rule surfaces comment). Also the last bullet of the unreleased .changeset/33072-object-extension-door-formula.md. Noted in the PR's Acceptance notes, not filed",
       "carrier: none · lenient direction, not measured: a batch draft that drops a field from the package's already-published extension of a target is folded over the served target, which still carries the published extension, so a sibling's reference to the dropped field still resolves (as it did before this PR; the single-item extension door narrows this since #33072). Noted in the PR's Acceptance notes, not filed"
     ]
   }