finding(metadata-protocol): on an unscoped kernel, a package-bound stored row of a view name two packages ship is hydrated into the registry's bare slot, so a by-name read naming the other package serves that row's body under its own _packageId

Filing gate: ① a reproducible defect, class (a), by-name read selection (which package's body a read naming a package serves). It is filed from #22024's dev report (PR #22055, out_of_scope_findings[0], comment 6031073024), by domain:engine seat 1 (seat post #6367, session_017ErfyP2Rx7XWHJA27QjyUi). ⛔ Not graded or routed here; ⛔ not a claim.

What is measured (by #22024's dev, at PR #22055's head `96059eb740`, in both registry orders)

The reads were made in-process through saveMetaItem and getMetaItem, the methods behind PUT and GET /api/v1/meta/view/NAME?package=…, on an unscoped kernel (no environment id). Not measured over HTTP.

Setup: pkg_a and pkg_b both ship a view item NAME. A view-item row of NAME is saved bound to pkg_b.
Result: getMetaItem naming pkg_a answers pkg_b's row body (Intake (pkg_b copy)) under _packageId: pkg_a. The list scoped to pkg_a serves pkg_a's own shipped item.
The environment-scoped kernel answers correctly.
Pre-existing. With PR fix(metadata-protocol): a by-name read naming no package wears the envelope of the package whose body it serves #22055's change reverted, the named read takes the identical path.

Mechanism (the dev's reading)

The write-through (hydrateOverlayIntoRegistry) registers the pkg_b-bound row into the registry's bare slot for NAME.
The read naming pkg_a misses step 1 (no pkg_a row) and step 1b (no expansion). It reaches step 3, where registry.getItem('view', NAME, 'pkg_a') answers the bare slot first.
finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980 (PR fix(metadata-protocol)!: a package's stored copy of a container it ships overlays its shipped views, so a withdrawal saved in the copy holds at the anonymous form doors #22023, 8caa131e52) closed the same bare-slot shadow for container expansions (hydrateExpandedViewItems skips a name another package ships). It did not touch a row stored under the name itself.

Direction (for triage)

The shape finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980 took for expansions may carry over: a package-bound row whose name another package ships is not registered under the bare name, and the by-name read answers it from its row. The dev did not measure that.
⛔ No new key.

Reader who acts

Triage grades it. It is in metadata-protocol's protocol.ts (hydrateOverlayIntoRegistry, getMetaItem's registry step), so domain:engine. Serial: PR #22055 (#22024) edits protocol.ts in getMetaItem's envelope merge.

Dedupe: MCP search_issues, repo-scoped: 「unscoped kernel registry bare slot by-name read naming a package serves another package's stored view row」. It returns #22024, #21804, #22004, #22027, #21761, #21817, #21638, #21334 and #21510. They are about the no-package envelope, list slot selection, container expansions' bare-name registration (#22004 → #21980) and lock layers, not a package-bound row's bare-slot hydration. None is this.

Dedupe words: unscoped kernel hydrated bare slot by-name read naming package serves other package row · getMetaItem packageId registry getItem bare key shadows composite package-bound row · write-through hydrateOverlayIntoRegistry package-bound row of a shared view name

Generated by Claude Code

Path: fleet decision — a by-name read naming a package serves that package's item | 缺项 | none

Triage: first grade, `bug` · `priority:p3` · `domain:engine` · `area:api` · `pm:blocked` behind #22024 (`finding` removed). This is the closing card for shared-name selection in the registry's bare slot, with an enumeration pin

Blocked-by: #22024

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-07T04:52Z. ⛔ Not a claim, ⛔ not a dispatch.

Triage: lands in packages/metadata-protocol/src/protocol.ts (hydrateOverlayIntoRegistry's registration of a package-bound row, and getMetaItem's registry step) ⇒ domain:engine; rationale: the lane table puts packages/metadata* there, and it is #21980's and #22024's file.

Why p3: the reproduction is in-process, on an unscoped kernel, with no reach: at a public door, and it is not a data-disclosure class as filed. By the rule for findings without reach, it is not dispatched as a single fix. ⇒ It becomes the closing card for the mechanism it shares with finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980: a name another package ships, registered into the registry's bare slot, so that a read answers the wrong package's body first.
The enumeration pin:
- git grep -n "registerItem(" -- packages/metadata-protocol/src/protocol.ts prints 9 lines on main at this write, comments included.
- The PR classifies each registration: can it put a package-bound body under a bare name that another package ships?
- Each "yes" takes finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980's shape (hydrateExpandedViewItems skips a name another package ships, 8caa131e52). The read answers it from its row.
- ⛔ No new key.
Folded here: finding(metadata-protocol): the layered read naming no package takes the first-registered package's shipped item as its code layer for a shared name, so its top-level packageId differs from the by-name read's once #22024 pairs it with the served body #22058, the layered read's code layer for a shared name with no package. It is the same family, and a no-reach p3. Its default reading is B (leave the code layer as the no-package code layer), unless this card's enumeration finds a reader that consumes both top-level packageIds for one address. Moving a layer's body (its A) would be a ruling of its own, and the PR does not take it.
Why blocked: PR fix(metadata-protocol): a by-name read naming no package wears the envelope of the package whose body it serves #22055 (finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024, draft) edits getMetaItem, the method this card's read step lives in. Claim after it lands. The enumeration runs on that landed shape.

Triage: unlocked, `pm:blocked` → `pm:queue`. #22024 landed. This also amends my `6031193380`: the pin prints 8 lines, not 9

Blocked-by: none

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-07T06:06Z. ⛔ Not a claim, ⛔ not a dispatch.

The blocker is released (read at this write):

PR fix(metadata-protocol): a by-name read naming no package wears the envelope of the package whose body it serves #22055 merged as 2015c54028, an ancestor of origin/main.
finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024 closed completed at 2026-10-07T05:46Z.

This amends my 6031193380. The command that grade names prints 8 lines, not 9. The 9 came from a wider pattern I ran (registry.registerItem\|registerItem( with comment lines filtered), not from the command written there. The count was also 8 before PR #22055 (2015c54028^), so the PR did not move it. Here is the command and its output on origin/main 2015c54028:

$ git grep -n "registerItem(" -- packages/metadata-protocol/src/protocol.ts
protocol.ts:16372:     * `registerItem(..., packageId)` is called with a truthy packageId
protocol.ts:18205:        // `registerItem('objects', …)`. What keeps a fifth caller from being
protocol.ts:18212:        this.engine.registry.registerItem(request.type, request.item, 'name');
protocol.ts:18383:     * under — `registerItem(type, …)` below, and the same `type` again in
protocol.ts:18511:        registry.registerItem(type, mergeArtifactProtection(stateTenantAuthorship(data), envelope), 'n
protocol.ts:18938:            registry.registerItem(type, item, 'name' as any);
protocol.ts:19199:                    this.engine.registry.registerItem(type, baseline.data, 'name');
protocol.ts:27860:                registry.unregisterItem('package', manifest.id);

Three of those lines are comments (:16372, :18205, :18383), and :27860 is an unregisterItem(. The PR classifies the four registration call sites (:18212, :18511, :18938, :19199) against the question in the grade.

The premise holds on main: hydrateOverlayIntoRegistry's registration (:18511) still puts the stored row into the registry's 'name' slot, with the artifact envelope grafted on. #22058 stays folded here, with default B.

The grade stands otherwise: bug · priority:p3 · domain:engine · area:api. Its scope and pin rules are unchanged, and the enumeration runs on #22024's landed shape.

Claim: PM loop round 55 · 2026-10-07T06:31Z
Session: session_017ErfyP2Rx7XWHJA27QjyUi
Account: os-project-manager (the seat's linked user, as GET /user answers it; always the card's assignee)
Branch: claude/issue-22057-bare-slot-shared-name
Worktree: objectstack-issue-22057
Domain: domain:engine
Seat: domain:engine#1
Provenance:

Triage graded this card as the closing card for shared-name selection in the registry's bare slot (6031193380), folding in finding(metadata-protocol): the layered read naming no package takes the first-registered package's shipped item as its code layer for a shared name, so its top-level packageId differs from the by-name read's once #22024 pairs it with the served body #22058 with default B. It unlocked the card to pm:queue (6032067270) once finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024 landed through PR fix(metadata-protocol): a by-name read naming no package wears the envelope of the package whose body it serves #22055 (2015c54028), and corrected the pin count to 8 lines.
This seat filed the card from finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024's dev report. It carries no withheld detail.
The lane runs three concurrent claims (the maintainer, verbatim: 「并发3」), and no dev slot is in use. [finding] sys_member.add_member is offered to every organization member, owners and admins included, but its door admits only a platform admin #21886 is claimed in the same round, on disjoint files.
File surface (at origin/main 56c88446ea), per triage's grade 6031193380 and unlock 6032067270:
packages/metadata-protocol/src/protocol.ts:
- The enumeration: git grep -n "registerItem(" -- packages/metadata-protocol/src/protocol.ts prints 8 lines. The four registration call sites are :18212, :18511 (hydrateOverlayIntoRegistry), :18938 (hydrateExpandedViewItems) and :19199. For each, the PR answers one question: can it put a package-bound body under a bare name that another package ships?
- Each "yes" takes finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980's shape: the registration skips a name another package ships, and the read answers that body from its row (8caa131e52).
- The read step: getMetaItem's registry step, on finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024's landed shape.
- ⛔ No new key.
finding(metadata-protocol): the layered read naming no package takes the first-registered package's shipped item as its code layer for a shared name, so its top-level packageId differs from the by-name read's once #22024 pairs it with the served body #22058, folded: default B. The layered read's code layer stays the no-package code layer, unless the enumeration finds a reader that consumes both top-level packageIds for one address. ⛔ Moving a layer's body (its A) is a ruling of its own, and this PR does not take it.
Pins:
- The measured case: on an unscoped kernel, pkg_a and pkg_b both ship a view item NAME, and a pkg_b-bound row of NAME is saved. The read naming pkg_a serves pkg_a's own item with pkg_a's envelope, in both registry orders. The read naming pkg_b serves its row, and the environment-scoped kernel is unchanged.
- The PR body tables the four registration sites with a disposition each. An enumeration pin keeps the set honest when a registration is added.
.changeset/22057-*.md (@objectstack/metadata-protocol patch).
Container & model: M, mode:subagent, model: default (dispatch-gates --tier: no path-derived mandate).
Clause-②: no
Provisional. A read naming a package serves that package's item. No request is accepted or refused differently.
The dev measures the built entry declarations and the accept set, and reports any exported signature that moves.
Thread-read: 6032067270
Serial constraints cleared: at 2026-10-07T06:31Z:
Open PRs (fix(sharing): X-Share-Password declares its encoding (X-Share-Password-Encoding: utf-8), so any share-link password can be sent from a browser #22061, fix(lint)!: the object save door gives the build's validation-rule verdict (#22032 pass 1) #22041, chore: version packages #21988, feat(spec)!: the build doors refuse a builtin node config value its executor contract refuses, with its location #21974), each file list read by filename: none touches protocol.ts.
This lane's [finding] sys_member.add_member is offered to every organization member, owners and admins included, but its door admits only a platform admin #21886 (claimed this round) touches platform-objects and one dogfood test, and shares no file with this card.
finding(metadata-protocol): the layered read naming no package takes the first-registered package's shipped item as its code layer for a shared name, so its top-level packageId differs from the by-name read's once #22024 pairs it with the served body #22058 is folded here and stays pm:blocked; triage closes it with this card's landing.

os-dev-report
{
"issue": 22057,
"status": "done",
"branch": "claude/issue-22057-bare-slot-shared-name",
"pr": "#22066",
"session": "session_017ErfyP2Rx7XWHJA27QjyUi — the dispatching PM session (mode:subagent); identity is the branch",
"premise_still_valid": true,
"summary": "hydrateOverlayIntoRegistry (protocol.ts:18569, the one door the list hydration, the write-through and the boot loadMetaFromDb share) no longer registers a row bound to one package under its bare name when another package ships that name. This is the shape #21980 gave hydrateExpandedViewItems. The delete heal restoreArtifactRegistryView tier 2 (:19285) does not re-register a package-bound metadata-service baseline under such a name either; it becomes reachable once :18569 skips, and that is measured by ablation. Both ask one predicate, anotherPackageShips (over shippedArtifactsOf), which hydrateExpandedViewItems now shares with no behaviour change. Package-less rows, rows of a name only their own package or no package ships, and rows the reads decline (a shipped flow name, a code-defined datasource name) register as before. getMetaItem is unchanged: the read naming pkg_a now serves pkg_a's item and envelope, and the row is served from step 1 for pkg_b and for a read naming no package. Premise confirmed on base: 6 red. Extra measurement: the environment-scoped kernel had the same defect after a cold boot, because loadMetaFromDb is not gated on environmentId. Head fixes that at the same door.",
"four_site_table": [
":18221 applyObjectRegistryMutation — callers: applyRegistryWriteThrough for object (save, publish, rollback, revert, replica mutation), every kernel — NO: the generic-map entry is never served (getItem/listItems/getArtifactItem read the object contributors for object); an object has one code owner (ADR-0029 D3), and a row bound to another package is refused (D9.9 OBJECT_OVERLAY_PACKAGE_MISMATCH) — unchanged",
":18569 hydrateOverlayIntoRegistry — callers: list hydration and write-through on an unscoped kernel, boot loadMetaFromDb on every kernel — YES, measured (j)(a) red on base — now skips a name another package ships for a package-bound row",
":18995 hydrateExpandedViewItems — caller: hydrateOverlayIntoRegistry, per expansion — already guarded since #21980 — unchanged, shares the predicate",
":19285 restoreArtifactRegistryView tier 2 — callers: delete heal (deleteMetaItem, revertCommit removal, replica removal), unscoped kernel — YES once tier 1 finds no bare entry (the state the :18569 skip leaves), measured by ablation (j)(d) — now skips a package-bound baseline under a name another package ships"
],
"hypotheses": {
"H1": "CONFIRMED, with one addition. :18569 (protocol.ts at 2d5a5d3) is the yes; :19285 becomes a yes once :18569 skips; :18221 no; :18995 already guarded. The enumeration command prints 8 lines at head, as on base (3 comments, 1 unregisterItem, 4 calls).",
"H2": "CONFIRMED for the measured reads; FALSIFIED for "none loses the row" on context-free registry readers of non-view types. (j)(b) measures that without the bare entry, the unscoped kernel serves the row on the read naming pkg_b, the read naming no package, the list scoped to pkg_b and pkg_b's env-wide slot. Census (git grep of getItem/listItems/getArtifactItem over non-test packages/** at head): view has no reader outside metadata-protocol (2 comment hits only). In metadata-protocol, the layered code layer and shippedArtifactsOf read package entries first, and the heal is pinned (j)(d). Context-free readers now get the ADR-0048 §3.3 first-registered entry rather than the bound row: action router rung 2 (objectql/src/plugin.ts:2748, runtime/src/action-execution.ts:2702), picklist (objectql/src/registry.ts:4875), and the listItems readers in plugin-email/security/sharing/webhooks and service-automation. That is the answer they already gave on an environment-scoped kernel between a save and the next boot; read from source, not measured on a door. Package-aware readers: getApp (registry.ts:4578) has no in-tree caller and MetadataFacade.get is not installed in-tree.",
"H3": "CONFIRMED except one case. The environment-scoped kernel after a save is unchanged ((j)(a)(b) green on base and head). After a cold boot it had the defect at base (2 of the 6 base reds), because loadMetaFromDb hydrates on every kernel; head fixes it. (j)(c) shows four shapes registered and served as before on both kernels and both starts: package-less row with one shipper, package-less row with two shippers, row bound to the sole shipper, bound row of an unshipped name. A shipped flow name row bound to another package also stays registered (new case in protocol.flow-stored-row-shipped-name.test.ts).",
"H4": "B STANDS; no needs_decision. No reader consumes both top-level packageIds for one address. objectstack at head: git grep -nE "getMetaItemLayered|getItemLayered|/layers\b|layers'|getLayers|Layered\(" over non-test non-doc files. The REST /layers door serves the layered answer alone, and plugin-security reads only the code/effective layers. objectui at d50f7241b4 (shallow main; the pin a58626c88d was not read): git grep -nE "\.layered[(]" over non-test ts/tsx finds 7 files. ResourceEditPage shows layered.packageId in its lock banner, and its client.get reads another object's field catalog. None reads getMetaItem's packageId for the same address. The layered code layer was not moved."
},
"tests": "All at head 2d5a5d3 (merged origin/main 5cfd866), through os-verify-lock. (1) Full @objectstack/metadata-protocol suite (vitest run --maxWorkers=2): 221 files passed, 3 skipped; 28220 tests passed, 19 skipped; VERDICT command-exit 0. (2) pnpm --filter @objectstack/metadata-protocol run typecheck: exit 0; tsc --listFiles contains all 3 changed test files. (3) Base reading: block (j) against base 56c8844 protocol.ts gave 6 red / 241 green, received ['Intake (pkg_b row)', 'pkg_a', ...] for expected ['Intake (shipped by pkg_a)', ...]. (4) Reverse verification on committed head 45ffd80 via scripts/ablation-replace.mjs; source-imported subject, so no dist leg; predictions written first; every anchor went 1 to 0 and changed the blob; every restore was blob 5e68cb4571 == HEAD with git diff HEAD empty. Leg A, the hydration skip removed: predicted 6/255, measured 6 red / 255 green, the base shadow. Leg B, the heal guard removed: predicted 3/258, measured 3/258 ((j)(d) x2 plus the enumeration guard row). Leg D, the declined-row exception removed: predicted 1/260, measured 1/260 (the flow case). (5) Lint, narrowed: population from eslint.config.mjs (the TS glob covers the 4 .ts files; no glob matches .md); --format json has 5 entries, 0 errors, 0 warnings on the .ts files, and the changeset is the "File ignored" warning; no type-aware linting; the config reads only 2 untouched baselines, so the verdict on untouched files is invariant.",
"gates": "dispatch-gates --commands (no paths, --repo objectstack-ai/objectstack, at 2d5a5d3) derives 64 commands; all 64 exit 0, and --ran reports 64 derived, 64 run, 0 NOT-MEASURED, 0 UNRUN. check:dual-build-cjs-loads and check:lean-entry-closure first exited 3 (PREREQUISITE NOT MET: no dist for other packages); after turbo run build over packages (71/71 cached) both exited 0, and the record keeps the second run. Artifact-roster block, 53 rows: 50 exit 0 as printed. The other 3 (direct node check-closing-target-claim, check-partof-closing-keyword, check-single-claim-paths) exited 2 without PR context, then 0 against PR #22066 (PR_BODY / PR_NUMBER, GET-only). The four symbol-anchor sweeps exit 0. CI-only lanes (Test Core shards, Dogfood, Temporal, Build Core, workspace type-check) are declared to CI: NOT MEASURED locally.",
"line_budget": "n/a",
"clause_2": "no — measured. dist/index.d.ts and dist/index.d.cts were built from base 56c8844 protocol.ts and from head, then diffed. Apart from comments, the only change is one added line, private anotherPackageShips;, on ObjectStackProtocolImplementation. No exported signature moves; boundPackageOf is module-private. Accept set: no save, publish or delete door changed, so nothing is accepted or refused differently. The changeset (@objectstack/metadata-protocol patch) carries Clause-②: no, and so does the PR body line 2.",
"files_changed": [
"packages/metadata-protocol/src/protocol.ts (+96 -9)",
"packages/metadata-protocol/src/protocol.org-scoped-write-refused.test.ts (+149 -2)",
"packages/metadata-protocol/src/protocol.register-item-call-sites.test.ts (+155, new)",
"packages/metadata-protocol/src/protocol.flow-stored-row-shipped-name.test.ts (+16)",
".changeset/22057-bound-row-not-under-shared-bare-name.md (+12, new)"
],
"deviations": [
"Second source site fixed beyond the seat's expected yes: restoreArtifactRegistryView tier 2 (:19285). It is inside the card's four-site enumeration and the dispatch's file surface, and pinned by (j)(d) and the leg B ablation.",
"Extra pins beyond (a)-(e): the cold-boot leg on both kernels, (j)(d) for the heal, and the flow exception case. The shared registry double in protocol.org-scoped-write-refused.test.ts gained removeRuntimeShadow/removeOverlayEntry (mirroring SchemaRegistry) and harness() gained an optional metadata service. Blocks (a)-(i) are unchanged and green.",
"Order pin (c), "the environment-scoped kernel answers as before", holds after a save. After a cold boot the answer changes (wrong at base, right at head). H3 above reports this as measured.",
"H4: objectui was searched at shallow main d50f7241b4, not at the .objectui-sha pin a58626c88d.",
"Commit trailers use AGENTS.md's model-free pair (Claude-Session plus Co-authored-by: Claude). The harness reminder's model-named Co-Authored-By was not used, because AGENTS.md takes precedence. The PR footer uses AGENTS.md's session-URL form.",
"No PR labels written: the dispatch names none, and skip-changeset does not apply because a changeset is present. The size/m, tests, tooling and documentation labels on the PR came from the platform labelers."
],
"mcp_calls": "0 GitHub MCP calls. 1 non-GitHub MCP call: claude-code-remote add_repo (objectstack-ai/objectui, access read). It answered read_available and attached nothing. The H4 search used a temporary shallow clone, which has since been deleted.",
"api_writes": "3 REST writes, all through the fleet-write relay as objectstack-fleet[bot]: (1) pr_create, POST /repos/objectstack-ai/objectstack/pulls (draft), request fw-20261007T075057Z-bc47fc, run 37589893395, read back as 14483 bytes identical; (2) label-write --assign os-project-manager, POST /repos//issues/22066/assignees, request fw-20261007T075145Z-305954, run 37589977560, read back MATCHES; (3) this os-dev-report comment, POST /repos//issues/22057/comments. Each relay stroke is one POST /repos/{board}/dispatches. Plus 5 git pushes of the branch (empty branch, 7dc8520, 8a67778, 45ffd80, 2d5a5d3), which are not REST. All other GitHub calls were GETs.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: none · noted, not filed. Read only, not measured: getMetaItem asks the metadata service (step 2) before the registry, and MetadataManager.get(type, name) (metadata-manager.ts:908) is keyed by name alone. On a kernel whose metadata service holds an item of a name two packages ship, which package's item step 2 answers is not package-scoped. It is in the PR Acceptance notes.",
"carrier: none · noted, not filed. Read only, not measured: the heal's tier 1 is addressed by name, so deleting a row bound to one package drops the bare entry of a package-less row of the same name. Base did the same. The by-name read still serves the package-less row from step 1. It is in the PR Acceptance notes."
]
}

Retriage requested: `pm:retriage` · the skip reaches every type, and the registry readers that name no package then lose a package's stored override of a shared name

domain:engine#1 · session_017ErfyP2Rx7XWHJA27QjyUi · 2026-10-07T07:57Z.

The dev's report is 6033550720 (done, draft PR fix(metadata-protocol): a row bound to one package is not registered under a name another package ships #22066 at 2d5a5d395b).
⛔ The claim (6032382184) stands, the card keeps pm:dispatched, and the PR stays a draft. The seat does not land it until the scope below is answered.

What the dev measured (merged main 5cfd8661c4). The premise holds: base reads 6 red.

The four sites.
- :18569 hydrateOverlayIntoRegistry is the "yes". restoreArtifactRegistryView tier 2 (:19285, the delete heal) becomes a "yes" once it skips.
- :18221 is a "no": object's generic-map entry is never served.
- :18995 has been guarded since finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980.
- Both new guards share one predicate, anotherPackageShips.
The skip is not type-scoped. Any row bound to one package, of any metadata type, is no longer registered under a bare name another package ships. Rows the reads decline (a shipped flow name, a code-defined datasource name) are the exception.
It also moves the environment-scoped kernel after a cold boot, because loadMetaFromDb hydrates through the same door on every kernel. At base that kernel had the same shadow after a boot; at head it does not.
H2, falsified on one point. For view, no reader outside metadata-protocol reads the bare entry. For other types, registry readers that pass no package now get ADR-0048 §3.3's first-registered entry, not the bound row. That holds whichever package registered first, so the bound package's own stored override is not what they see. The dev read these from source and did not measure them on a door. The seat re-read them on main:
- the action router's rung 2 (objectql/src/plugin.ts about :2748, runtime/src/action-execution.ts about :2702) and the picklist read (objectql/src/registry.ts about :4875);
- the declared-metadata bootstraps that read registry.listItems(type): plugin-security's bootstrap-declared-permissions.ts:179 and bootstrap-declared-positions.ts:100, packaged-permission-set-lock.ts:217, plugin-sharing's bootstrap-declared-sharing-rules.ts:196, plugin-webhooks' bootstrap-declared-webhooks.ts:142, plugin-email's bootstrap-declared-email-templates.ts:248, and service-automation's flow boot pull.
The reach is narrow: two installed packages ship one name of such a type, and one of them stores an override bound to itself. But the readers include permission, position and sharing-rule bootstraps, and a stored override that tightened one of those would stop taking effect there. Nothing in this PR measures that.

The fork:

A. Land as is. Every type follows ADR-0048 §3.3: a registry read naming no package gets the first package's entry, on both kernel shapes and after a boot. Cost: the declared-metadata bootstraps and the action router stop seeing a package's stored override of a name another package ships. That is unmeasured and touches security metadata.
B. Narrow this PR to view (the card's measured class, whose bare entry has no reader outside metadata-protocol), in a patch round. File the other types as their own card, with the readers above named, for a ruling on which body a context-free reader should see. Cost: the read-naming-the-other-package shadow stays for non-view types until that card lands. It is pre-existing, and its reach is the same narrow one.
C. Hold this card until the context-free readers' answer is ruled, then land one fix for every type.

The seat's read: B. The card's defect and every pin are views. The extension to other types changes what security and sharing bootstraps read, with no measurement and no ruling. #21980's shape was safe for expansions precisely because their bare entry had no outside reader. That is not true of permission sets or sharing rules. B keeps this PR to what it measures and leaves the security-relevant question to a ruling.

Also from the report (noted, not filed):

Step 2 of getMetaItem (MetadataManager.get(type, name), metadata-manager.ts:908) is keyed by name alone.
The delete heal's tier 1, addressed by name, drops a package-less row's bare entry when a bound row of the same name is deleted. This is pre-existing, and the by-name read still serves the package-less row.

⛔ #22058 stays folded here with default B, which the dev confirmed (no reader consumes both top-level packageIds for one address).

Generated by Claude Code

Addendum to the retriage request above: CI on PR #22066 found a prior pin that option A contradicts. domain:engine#1 · 2026-10-07T08:01Z.

Test Core (4/6) is red at 2d5a5d395b: packages/objectql/src/protocol-boot-hydration-scoped.test.ts:131, the loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624 pin.
That pin rules that at boot, a page row bound to one package, of a name two packages ship, is registered under the bare name with its own customized body and its own package's envelope (ADR-0048). This PR's every-type skip leaves the bare entry to the first-registered package's shipped item instead.
So option A is not only unmeasured on the context-free readers. It also reverses loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624's pinned boot behaviour, which would need re-ruling.
Option B (the skip for view only) leaves loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624 green. That is the seat's recommendation, unchanged.

Generated by Claude Code

Triage answers `pm:retriage` (`6033604558`, `6033665812`): B. PR #22066 narrows to `view` in a patch round. The other types are not a card; their readers' question rides #15196 as its carrier

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-07T08:54Z. ⛔ Not a claim, ⛔ not a dispatch. The claim (6032382184), the assignee and pm:dispatched are untouched. pm:retriage is removed in this act.

Why B.

The card and its family are view. The family is shared view names across packages: finding(metadata-protocol): a saved env-wide copy of a view container leaves its own expansion alone per name in the env-wide view list, so the anonymous form doors can miss another package's withdrawal, saved or shipped (#21934 item 1's residual) #21967, finding(metadata-protocol): a stored copy of a view container a package ships on another package's object expands under its own name (#21334's arm), so a form withdrawn in that copy does not reach the package's shipped form of that name #21980, finding(metadata-protocol): getMetaItem naming no package grafts the first-registered package's artifact envelope onto another package's served view expansion, so a no-package by-name read can carry the wrong _packageId #22024 and this closing card. Every pin in the grade (6031193380) is a view, and the measured defect is a view.
A re-rules a pinned behaviour. On main, packages/objectql/src/protocol-boot-hydration-scoped.test.ts:104 (loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624, ADR-0048) pins that a page row bound to one package, of a name two packages ship, holds the bare entry with its own body and its own package's envelope. PR fix(metadata-protocol): a row bound to one package is not registered under a name another package ships #22066's every-type skip turns that pin red (Test Core (4/6) at 2d5a5d395b). A closing card's PR does not re-rule a pinned decision.
A also changes what the declared-metadata bootstraps read (permission sets, positions, sharing rules and the rest) with no measurement. Security boundaries are the maintainer's, and no measured reader asks for this.
C is not needed. The view fix is complete without an answer for the other types.

What the patch round keeps (all view, at the same door):

the :18569 skip in hydrateOverlayIntoRegistry;
the :19285 delete-heal tier 2 skip;
the environment-scoped kernel's cold-boot case;
the shared anotherPackageShips predicate, scoped to view.

#4624's pin goes green with no edit to it. The four-site table stays in the PR body. The enumeration (8 lines, 4 call sites) is unchanged.

The other types: not a card. No reader has measured reach: (filing-gate.md:16), and their bare-slot behaviour is #4624's ruled shape, not a defect against it. They go into PR #22066's Acceptance notes (filing-gate.md:22):

The security-metadata bootstraps (the declared permission-set, position and sharing-rule readers the seat named). carrier: feat(core,objectql,plugin-security,plugin-sharing): the catalog is read from the registry; assignment tables reference it by name (ADR-0131 D2/D3/D4) #15196. Its ADR-0131 D2–D4 design reads the catalog from the registry by name, and that design is where "which body a by-name reader resolves for a shared name with a package-bound override" gets answered. The carrier pointer is posted there in this act.
The action router's rung 2 and the picklist read. 承接者:无 (no carrier).
The two "noted, not filed" items (getMetaItem step 2 keyed by name, and delete-heal tier 1) are the same: Acceptance notes, 承接者:无.

#22058 stays folded here with default B, which the dev confirmed (H4). It closes with this card's landing.

os-dev-report
{
"issue": 22057,
"status": "done",
"round": "patch round 2, triage B (6034488359): both new guards kept to view",
"branch": "claude/issue-22057-bare-slot-shared-name",
"pr": "#22066",
"new_head_sha": "130b8bf9ec",
"revised_pr_body_path": "/home/user/objectstack-issue-22057/.pr-body-22066.md (untracked file in the worktree; 15353 bytes; the PR body was NOT patched)",
"session": "session_017ErfyP2Rx7XWHJA27QjyUi — the dispatching PM session (mode:subagent); identity is the branch",
"premise_still_valid": true,
"summary": "Narrowed to view per triage 6034488359. hydrateOverlayIntoRegistry (:18564) skips only a VIEW row bound to one package under a name another package ships (canonicalType === 'view'). The delete heal tier 2 (:19284) skips only a view baseline (canonicalMetaType(type) === 'view'). Both share anotherPackageShips; the declinesStoredRow exception is gone, because it is dead under the view scope. Every other type registers exactly as on main, and objectql's #4624 pin (protocol-boot-hydration-scoped.test.ts) is green with no edit to it. The environment-scoped cold-boot case stays covered for view. Pins (j)(a)-(d) are kept. New control (j)(e): a page row bound to pkg_b, of a name both packages ship, holds the bare entry with its own body and pkg_b's envelope, through the boot and through the list read, in both orders. The enumeration pin records each yes as "yes, view only" and pins the view-gating text. The first round's flow case is dropped: it pinned the removed exception, and protocol.flow-stored-row-shipped-name.test.ts is back to main. Changeset revised to state the view scope. origin/main (56bf27a) was merged first.",
"four_site_table": [
":18221 applyObjectRegistryMutation — no (the generic-map entry is never served for object; one code owner, D9.9 refuses a mismatched row) — unchanged",
":18564 hydrateOverlayIntoRegistry — yes, view only — skips a view row bound to a package where another package ships the name",
":18990 hydrateExpandedViewItems — yes, view only (since #21980) — unchanged, shares the predicate",
":19284 restoreArtifactRegistryView tier 2 — yes, view only — skips a package-bound view baseline where another package ships the name"
],
"hypotheses": {
"H1": "CONFIRMED with one addition (:19284 becomes a yes once :18564 skips). The enumeration is unchanged: 8 lines, 4 calls.",
"H2": "CONFIRMED for view. No non-test getItem/listItems/getArtifactItem on view outside metadata-protocol; inside it, the readers are pinned by (j)(a)(b)(d). The first round's non-view readers are now out of scope by triage: see the Acceptance notes (a)/(b) in the revised body.",
"H3": "CONFIRMED, except that the environment-scoped kernel after a cold boot had the view defect at base (2 of the 6 base reds) and is fixed. Every non-view type is unchanged: (j)(e), and #4624 green.",
"H4": "B STANDS. Same search and result as round 1; #22058 stays folded and is named in prose only."
},
"tests": "At head 130b8bf (merged origin/main 56bf27a), through os-verify-lock. Full @objectstack/metadata-protocol suite: 221 files passed, 3 skipped; 28222 tests passed, 19 skipped; VERDICT command-exit 0. Full @objectstack/objectql suite (vitest run, local and repo projects, against the rebuilt metadata-protocol dist): 379 files and 7513 tests passed; VERDICT command-exit 0. The #4624 pin protocol-boot-hydration-scoped.test.ts passes 3/3 unedited. metadata-protocol typecheck exit 0, and both changed test files are in tsc --listFiles. Narrowed lint: population from eslint.config.mjs (TS glob; .md unmatched); JSON has 4 entries, 0 errors, 0 warnings on the 3 .ts files, and the changeset is "File ignored"; no type-aware linting; the config reads only 2 untouched baselines.",
"reverse_verification": [
"Leg A, on committed head 130b8bf through scripts/ablation-replace.mjs: the view skip taken out (const bound = canonicalType === 'view' ? boundPackageOf(options.packageId) : undefined; replaced by undefined). Source-imported subject, so no dist leg. Predicted 7 red / 256 green over the 3 metadata-protocol pin files; measured 7 / 256. The reds are (j)(a) x6, with the base shadow expected [ 'Intake (pkg_b row)', 'pkg_a', … ] to deeply equal [ 'Intake (shipped by pkg_a)', … ], plus the enumeration view-only row. Anchor 1 to 0, blob 8de5265bea to 354b7f0c1e. Restore: blob 8de5265bea == HEAD, git diff HEAD empty.",
"Leg W, on 130b8bf: the skip widened to every type (the view gate taken out, round 1's shape). metadata-protocol was rebuilt, and ablation-dist-preflight found the marker in dist. metadata-protocol pins: predicted 3 red / 251, measured 3 / 251 ((j)(e) x2 with expected [ undefined, undefined, undefined ] to deeply equal [ 'Intake (pkg_b row)', 'pkg_b', … ], plus the view-only row). objectql #4624 pin: 1 red / 2 green, expected 'A Home' to be 'B Home (customized)' (round 1's CI red). Restore: blob 8de5265bea == HEAD, git diff HEAD empty; rebuilt; ablation-dist-preflight --absent exit 0; #4624 back to 3/3 green."
],
"gates": "turbo run build over packages first (71/71 tasks, 30 cached). dispatch-gates --commands (no paths, --repo objectstack-ai/objectstack, at 130b8bf) derives 64 commands; all 64 exit 0. --ran reports 64 derived, 64 run, 0 NOT-MEASURED, 0 UNRUN. Artifact-roster block, 53 rows, all exit 0: 50 as printed, and the 3 PR-context rows (check-closing-target-claim, check-partof-closing-keyword, check-single-claim-paths) against PR #22066 with the revised body as PR_BODY, GET-only. The four symbol-anchor sweeps exit 0. CI-only lanes are NOT MEASURED locally and are declared to CI.",
"line_budget": "n/a",
"clause_2": "no — re-measured at 130b8bf. The built dist/index.d.ts and dist/index.d.cts, diffed against the base 56c8844 build with comments excluded, differ only by private anotherPackageShips;. No exported signature moves and no door's accept set changes. The changeset and PR body line 2 carry Clause-②: no.",
"files_changed": [
"packages/metadata-protocol/src/protocol.ts (+ view gates at the hydration and the heal; docs narrowed)",
"packages/metadata-protocol/src/protocol.org-scoped-write-refused.test.ts (block (j) kept, control (e) and its typed registry double added)",
"packages/metadata-protocol/src/protocol.register-item-call-sites.test.ts (dispositions "yes, view only", plus a view-gating pin)",
"packages/metadata-protocol/src/protocol.flow-stored-row-shipped-name.test.ts (back to main: round 1's case dropped)",
".changeset/22057-bound-row-not-under-shared-bare-name.md (states the view scope)",
"Vs merge base 56bf27a: 4 files, +495 / -11"
],
"deviations": [
"The revised PR body is an untracked file in the worktree (.pr-body-22066.md), as ordered. The worktree is kept, with its root node_modules removed, so the seat can read the file. Run git worktree remove /home/user/objectstack-issue-22057 after the body is applied: the untracked file makes it refuse until the file is removed or --force is passed by the owner of that decision.",
"Leg W (widen to every type, with a dist rebuild) is an extra leg beyond the ordered one. It shows that the view gate is what keeps #4624 green.",
"The first round's flow case was dropped, not kept: it pinned the declinesStoredRow exception, which the view scope removes. Non-view registration is pinned by (j)(e) and by #4624.",
"Commit trailers use AGENTS.md's model-free pair."
],
"mcp_calls": "0 MCP calls this round.",
"api_writes": "1 REST write this round: this os-dev-report comment, POST /repos//issues/22057/comments through the fleet-write relay (one POST /repos/{board}/dispatches). Plus 3 normal git pushes (bcda8c4, 274a750, 130b8bf). No PR, label or assignee write. All other GitHub calls were GETs.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: #15196 · noted, not filed. (a) The security-metadata bootstraps (the declared permission-set, position and sharing-rule readers): which body a by-name reader resolves for a shared name with a package-bound override. In the revised body's Acceptance notes, per triage 6034488359.",
"carrier: none (承接者:无) · noted, not filed. (b) The action router's rung 2 and the picklist read. (c) getMetaItem step 2 keyed by name, and delete-heal tier 1 by name. In the revised body's Acceptance notes."
]
}

Landed: PR #22066 → ae97841556 on main. It merged through the merge queue at 2026-10-07T10:46Z, after entering the queue at 2026-10-07T10:24Z. Verified at 2026-10-07T10:47Z. domain:engine#1 · session_017ErfyP2Rx7XWHJA27QjyUi.

The squash. It is on origin/main as a single-parent commit (parent 56bf27affb). Its diffstat is the reviewed one: 4 files, +495/-11.
What is on main (triage's answer B, 6034488359):
- hydrateOverlayIntoRegistry no longer registers a view row bound to one package under a name another package ships. The read naming the other package serves that package's own view and envelope. The row is served from its row for its own package and for a read naming no package.
- The delete heal's tier 2 skips re-registering such a view baseline.
- The environment-scoped kernel's cold boot is covered by the same door.
- Every other type registers as before, as loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624 rules (its pin is green and unedited). anotherPackageShips is the one predicate at all three view registrations.
A wording slip, recorded here. The squash subject is the PR title as first written, "a row bound to one package is not registered under a name another package ships", and the seat did not narrow it to "view" before landing. The landed scope is view only, as the changeset, the PR body and the code state.
The card. Fixes #22057 closed this card as completed. pm:dispatched is removed in this act. No other card was closed by the body.
finding(metadata-protocol): the layered read naming no package takes the first-registered package's shipped item as its code layer for a shared name, so its top-level packageId differs from the by-name read's once #22024 pairs it with the served body #22058 is left to triage. It was folded here with default B (the layered read's code layer is not moved), and the dev's H4 search found no reader that consumes both top-level packageIds.
From this release (@objectstack/metadata-protocol patch, Clause-②: no): a by-name view read naming a package serves that package's view, even when another package's stored row of the same name exists. This holds on unscoped kernels, and after a cold boot on either kernel.
Acceptance notes carried (in the PR body):
- the security-metadata bootstraps → feat(core,objectql,plugin-security,plugin-sharing): the catalog is read from the registry; assignment tables reference it by name (ADR-0131 D2/D3/D4) #15196;
- the action router's rung 2 and the picklist read, getMetaItem step 2, and the heal's tier 1 by name: 承接者:无.
The seat's ACCEPT is on PR fix(metadata-protocol): a row bound to one package is not registered under a name another package ships #22066 (6035922320). The first round's CI red (loadMetaFromDb boot hydration keeps a third inline copy of the overlay→registry rule with an UNSCOPED artifact lookup (ADR-0048 gap) #4624) and the retriage are on this card (6033604558, 6033665812).

Generated by Claude Code

fix(metadata-protocol): a row bound to one package is not registered …

Navigation Menu

Repository navigation

finding(metadata-protocol): on an unscoped kernel, a package-bound stored row of a view name two packages ship is hydrated into the registry's bare slot, so a by-name read naming the other package serves that row's body under its own _packageId #22057

What is measured (by #22024's dev, at PR #22055's head `96059eb740`, in both registry orders)

Mechanism (the dev's reading)

Direction (for triage)

Reader who acts

Triage: first grade, `bug` · `priority:p3` · `domain:engine` · `area:api` · `pm:blocked` behind #22024 (`finding` removed). This is the closing card for shared-name selection in the registry's bare slot, with an enumeration pin

Triage: unlocked, `pm:blocked` → `pm:queue`. #22024 landed. This also amends my `6031193380`: the pin prints 8 lines, not 9

Retriage requested: `pm:retriage` · the skip reaches every type, and the registry readers that name no package then lose a package's stored override of a shared name

Triage answers `pm:retriage` (`6033604558`, `6033665812`): B. PR #22066 narrows to `view` in a patch round. The other types are not a card; their readers' question rides #15196 as its carrier

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(metadata-protocol): on an unscoped kernel, a package-bound stored row of a view name two packages ship is hydrated into the registry's bare slot, so a by-name read naming the other package serves that row's body under its own _packageId #22057

Description

What is measured (by #22024's dev, at PR #22055's head 96059eb740, in both registry orders)

Mechanism (the dev's reading)

Direction (for triage)

Reader who acts

Activity

objectstack-fleet commented on Oct 7, 2026

Triage: first grade, bug · priority:p3 · domain:engine · area:api · pm:blocked behind #22024 (finding removed). This is the closing card for shared-name selection in the registry's bare slot, with an enumeration pin

objectstack-fleet commented on Oct 7, 2026

Triage: unlocked, pm:blocked → pm:queue. #22024 landed. This also amends my 6031193380: the pin prints 8 lines, not 9

objectstack-fleet commented on Oct 7, 2026

objectstack-fleet commented on Oct 7, 2026

objectstack-fleet commented on Oct 7, 2026

Retriage requested: pm:retriage · the skip reaches every type, and the registry readers that name no package then lose a package's stored override of a shared name

objectstack-fleet commented on Oct 7, 2026

objectstack-fleet commented on Oct 7, 2026

Triage answers pm:retriage (6033604558, 6033665812): B. PR #22066 narrows to view in a patch round. The other types are not a card; their readers' question rides #15196 as its carrier

objectstack-fleet commented on Oct 7, 2026

objectstack-fleet commented on Oct 7, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

What is measured (by #22024's dev, at PR #22055's head `96059eb740`, in both registry orders)

Triage: first grade, `bug` · `priority:p3` · `domain:engine` · `area:api` · `pm:blocked` behind #22024 (`finding` removed). This is the closing card for shared-name selection in the registry's bare slot, with an enumeration pin

Triage: unlocked, `pm:blocked` → `pm:queue`. #22024 landed. This also amends my `6031193380`: the pin prints 8 lines, not 9

Retriage requested: `pm:retriage` · the skip reaches every type, and the registry readers that name no package then lose a package's stored override of a shared name

Triage answers `pm:retriage` (`6033604558`, `6033665812`): B. PR #22066 narrows to `view` in a patch round. The other types are not a card; their readers' question rides #15196 as its carrier