Repository navigation
meta: with two package holders of one dashboard name, GET /meta/dashboard/X/layers silently returns one holder's overlay while package-less save and delete answer 409 #28337
Description
domain:devx#2 · os-justin · session_01QSeKYmTYfaPbLQD4BJfomW
Filing gate: ① an observed, reproduced inconsistency from the 18.3.0 release-candidate run. W3-C2b, record #27438, at c4c515ad, parked on the wave anchor #27377 as a ruling finding. Item: platform-core.metadata-registry-serving. ⛔ Not a claim; triage grades and routes. Suggested labels are suggestions only.
Observed
The record (W3-C2b comment 6494110310 on #27377) reads, verbatim:
with two package holders,
GET /meta/dashboard/X/layerssilently returns one holder's overlay, while the package-less save and delete on that name answer 409.
When two packages both hold a dashboard of the same name:
- the read door (
/layers) picks one holder silently; - the write doors (a save or delete without
?package=) refuse with 409, because the name is ambiguous.
One name gives two answers to the same ambiguity. A reader of /layers, whether the Studio or an agent, cannot tell that a second holder exists, or which holder it got.
Why it may matter
- Who: anyone who installs two packages that ship the same dashboard name, and the Studio or agent reading layers for it.
- What happens: the layer stack shown or edited belongs to whichever holder
/layerspicked, while a save on the same name is refused. This invites a wrong edit target or a confusing refusal. - Not measured: how often two holders of one dashboard name occur in real installs, and which holder
/layerspicks (registration order?).
Possible directions (for the owner and triage; not a recommendation)
/layersrefuses the same way the write doors do (409 plus the holders' package ids), or requires?package=;- or
/layersreturns every holder's stack, keyed by package; - or the current behaviour is ruled correct and documented, in which case the checklist item records it.
Dedupe
- Read metadata: a published org overlay on the built-in system_overview dashboard is never served — the plain read and the rendered board keep the shipped body while ?layers=true reports the overlay as effective #20680 and metadata-protocol: a package-less direct save of an item a package owns answers 200 and writes a package-less row the item read never serves; the draft save already resolves the owning package (#11087 / #22128) #23165: related, closed, different. The census's semantic search found no open card.
- Search terms:
layers,two holders,package holder,metadata-registry-serving.
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 31, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: 错的必须被响亮拒绝并给处方 | none | P③
Triage: p3 ·
domain:server·area:studio· 2026-10-31T05:56Z⛔ Not a claim.
/layerssilently picks one holder, while the package-less writes refuse with 409. Two package holders of one name are rare, so the grade is p3./layerswithout?package=refuses the same way the writes do (409, naming the holders, with the?package=prescription). One answer to one ambiguity. It is a narrowing,Clause-②: no.Generated by Claude Code
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_0186vfjHHMsRXC9EUT9hTFMo
Claim: PM loop round 4 · card thread · 2026-11-01T04:28Z
Session:
session_0186vfjHHMsRXC9EUT9hTFMo(the seat); the card thread issession_01FEkQ5o95tvF7A7y85W37iRAccount:
os-project-manager(the seat's linked user; the card's assignee)Branch:
claude/project-thread-0r3khj(the card thread's harness-pinned branch; it names no issue, so this claim is the identity record)Worktree: the card thread's own container checkout
Domain:
domain:serverSeat:
domain:server#2Ruling-ref: the maintainer, 2026-11-01T03:46Z: 「server 车道现在就是你负责，任务有一大堆啊，都可以派，包括 p3」, with 2026-10-26T10:29Z 「p1 p2 处理完了，就可以处理 p3」. Triage
6539811535:/layerswithout?package=refuses the same way the package-less writes do (409, naming the holders, with the?package=prescription); a narrowing. The order reuses the writes' judge, with no new refusal shape.File surface: the layered metadata read (
GET {metaPath}/:type/:name/layersand?layers=true) inpackages/rest/src/meta-*andpackages/metadata-protocol/src/**, the existing ambiguous-name judge it reuses, tests, and changesets. ⛔ Noui/**edit.Container & model:
M,mode:thread,model: opus(--tiernames no path mandate: PM's call, default),effort: highClause-②: no (narrowing — a read answered 200 today is refused; a contract review at
CONTRACT_REVIEW_TIERis owed before the queue)Metadata-protocol: none (expected; the dev measures and declares)
Responsibility: the code is the metadata protocol's layered read · the platform path is the Studio's and agents' layer view of an item · reached by anyone reading layers for a name two installed packages both hold
Thread-read: none
Premise: holds (per the RC record 6494110310 on #27377; the dev reproduces it on
origin/mainfirst)Serial constraints cleared: no open PR touches the layered read at this read (the dev re-checks)
Resume-from: none — fresh from
origin/mainState: →
pm:dispatched, assigned toos-project-manager, recorded by this claim.Generated by Claude Code
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_01FEkQ5o95tvF7A7y85W37iR
os-dev-report
{
"issue": 28337,
"status": "done",
"branch": "claude/project-thread-0r3khj",
"pr": "#28583",
"session": "session_01FEkQ5o95tvF7A7y85W37iR",
"premise_still_valid": true,
"summary": "Red first on origin/main: GET /meta/dashboard/twin_dash/layers with twin_dash held by two packages answered 200 with one holder's overlay, while PUT/DELETE naming no package answered 409 RESOURCE_CONFLICT. The shared layered chain createMetaLayeredAnswer (packages/rest/src/meta-item-read-gate.ts) gains step 1c: with no ?package= it asks the new protocol judge packagelessLayeredReadRefusal (metadata-protocol), which reuses packagelessAddress + packagelessAddressAmbiguityRefusal (verb 'layered read'), and throws that refusal to each transport's existing catch, so /layers and ?layers=true answer the same code, status, holders and message family as the package-less save on RestServer and on the runtime dispatcher. ?package= still reads that holder's stack; a name one package holds, or a package-unbound row, answers as before; in-process getMetaItemLayered is unchanged (its package-less in-process callers are not doors). Card assignee untouched; PR opened draft, no assignees, no labels (changesets present).",
"tests": "New: metadata-protocol protocol.packageless-layered-read-refusal.test.ts 3/3 (full refusal sentence pinned); rest meta-layers-ambiguous-package.test.ts 3/3 (real protocol + registry, in-memory sys_metadata); runtime domains/meta-layers-ambiguous-package.test.ts 4/4 (better-sqlite3, real protocol, RestServer handlers + catch-all into HttpDispatcher.dispatch, both spellings, read compared with the same door's PUT/DELETE answer). Red first: base rest src answered {status:200} for the two-holder read. Ablation:
if (ambiguity) throw ambiguity;->void ambiguity;, rest rebuilt, ablation-dist-preflight confirmed marker absent in dist; only two-holder cases red (rest 1/3, runtime 2/4); restored blob == HEAD, rebuilt, runtime 4/4. Package suites: metadata-protocol 290 files passed / 4 skipped, 21897 tests passed; rest 363 files, 6907 tests passed (both at b8f5998); runtime 500/500 files, 8095 tests passed (913e990 tree; b8f5998 only adds the changeset/ledger lines). typecheck exit 0 for all three; eslint exit 0 on the ten touched files.","gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack: 71/71 exit 0 at b8f5998 (exit codes captured before any pipe), --ran reconciliation 0 NOT-MEASURED. First-loop non-zero exits fixed then re-run: check:engine-double-contract exit 1 -> ledger --write committed -> 0; check-plugin-teardown-shape --self-test exit 3 (shallow clone missing its pinned commit) -> fetched 621a487 -> 0; check:dual-build-cjs-loads exit 3 (no dist) -> pnpm build --concurrency=2 -> 0; check:type-check-debt exit 3 (lock queue behind dogfood) -> 0. platform-checklist, commit trailers, nul-bytes 0. PR check-runs at b8f5998 when this report was written: Governed Surface Queue Guard success; Build Core, Check Changeset, Lint families, Type Check families, PR-body guards, Test UI, Live E2E UI, Dogfood Regression Gate queued/in_progress; opt-in families skipped. Not re-run locally; the seat reads them before landing.",
"dogfood_battery": "b8f5998e · VERDICT command-exit 0 · Test Files 363 passed / 3 skipped (366) · Tests 2932 passed / 28 skipped · no failures",
"mcp_calls": "0",
"api_writes": "2 — fleet-write relay: pr_create (#28583, draft, read back 12799/12799 bytes identical, assignees none); post-stamped comment on #28337 (this report). Plus git push (not a REST write).",
"open_questions": [],
"out_of_scope_findings": [
"Seam (ui lane, not filed, no ui/** edit): spec:GET /meta/:type/:name/layers (package-less) -> renderer: Studio MetadataClient.layered and other package-less /layers callers in ui/. After this lands, a name two packages hold answers those callers 409 RESOURCE_CONFLICT instead of one holder's stack; they need to pass ?package= (or surface the refusal). reach: public door, measured by the runtime test above. carrier: ui lane via seat 2. dedupe words: layers package-less 409, MetadataClient.layered package, Studio layered view ambiguous package",
"carrier: seat 2 · noted, not filed — open PRs #28554 (protocol.ts, engine-double-contract.pinned.json) and #28512 (protocol.ts) touch the same files; textual overlap only, whichever lands second rebases"
]
}
Generated by Claude Code
objectstack-fleet commentedon Nov 1, 2026
domain:server#2 · os-project-manager · session_0186vfjHHMsRXC9EUT9hTFMo
Landing record · #28337 · PR #28583
domain:serverseat 2 ·session_0186vfjHHMsRXC9EUT9hTFMo· 2026-11-01T07:09Zgit ls-remote origin 'refs/heads/gh-readonly-queue/main/pr-28583-*'git merge-base --is-ancestor cd065d36 origin/maingit grep -c packagelessLayeredReadRefusal origin/main -- packages/metadata-protocol/src/protocol.tscd065d36. The layered view of a name two packages hold, read with no?package=, now refuses with the same 409 the package-less save and delete answer, naming the holders and prescribing?package=. It no longer silently picks one holder. Both/:type/:name/layersand?layers=truebehave this way, on both doors.?package=<id>reads that holder's stack. A name with one stored holder reads as before, including a base and its.customoverride.domain:ui): two Studio callers should pass the owning package.Release: claim
6553412526(session_0186vfjHHMsRXC9EUT9hTFMo, branchclaude/project-thread-0r3khj) · landed.pm:dispatchedremoved; assignee cleared; the card closed on merge.Generated by Claude Code