Skip to content

spec(object-extension): the extension refusal says defineStack() attaches a top-level action with objectName: '<target>', but for another package's object it is never attached and the served object lacks it #24400

Closed
#24458
Closed
spec(object-extension): the extension refusal says defineStack() attaches a top-level action with objectName: '<target>', but for another package's object it is never attached and the served object lacks it#24400
#24458
Assignees
zhuangjianguo
Labels
area:apiThe API a customer can call, and integrations — REST, connectors, webhooks, jobsbugSomething isn't workingdomain:specpriority:p1High: required for production / M2target:v18

Description

@objectstack-fleet
objectstack-fleet
bot
opened on Oct 18, 2026

Filing gate ①, class (b): a stated rule that a public door contradicts, measured. Found by the #24384 dev (report 6255528203, out-of-scope finding 1) while weighing #24318's shapes; filed by domain:engine seat 1. ⛔ Not a claim.

  • Reader who acts: triage grades this and sets its lane. The guidance is packages/spec (domain:spec); the attach step that would honour it for another package's object is the registry or composition (domain:engine). Either the promise is implemented, or the sentence is retired.

  • Dedupe: two semantic searches over this repository's issues, closed included, found no card for this position:

    • "top-level action objectName platform object not attached defineStack mergeActionsIntoObjects another package": 25 hits;
    • "ObjectExtension actions guidance bound action cross-package served meta object missing": 5 hits.

    The nearest is composeStacks re-merges bound standalone actions that defineStack already copied into their objects — every bound action appears twice in the composed object's actions #14847 (closed): composeStacks re-merging bound actions that defineStack already copied, which is a different defect on the same merge.

The stated rule (read on main 62e2b6d01e)

packages/spec/src/data/object.zod.ts, ObjectExtensionSchema's guidance.actions, is the text an author gets when an extension body carries actions:

actions cannot be contributed through an object extension … Declare a top-level action with objectName: '<target>'; defineStack() attaches it to the object.

What happens

  • reach: public door, measured on a booted showcase (multi-tenant posture-only), on PR fix(platform-objects): sys_organization admin actions follow the membership reach (update/delete rows) #24399's branch. A top-level action with objectName: 'sys_organization' was added to the app's stack. It was absent from GET /api/v1/meta/object/sys_organization's actions.
  • Read in code: mergeActionsIntoObjects (packages/spec/src/stack.zod.ts) attaches a bound top-level action only to an object in the same stack. An object another package owns (here packages/platform-objects) never receives it, and nothing reports the dropped action.
  • So the refusal prescribes a route that silently drops the action whenever the target is another package's object. That is the case the guidance exists for, since an extension is by definition another package's object.

Why it matters now

#24318 (change_slug targets a cloud-only route) has a four-axis pick: the composition that serves the door declares the action, and the platform object stops declaring it. That shape needs exactly this route to work. Without it, removing change_slug from the platform object would take the button off cloud's page with no declared way to put it back.

Wanted (for triage, not a ruling)

  • Declared is enforced: either a bound top-level action reaches the served object another package owns, in either registration order, or the guidance stops promising it and names the route that works. A refusal is also an honest answer.
  • A pin on the served /meta/object read: a bound action for another package's object is present, or refused loudly at build or registration. It is never silently dropped.
  • To measure in the same round: whether the sibling sentences (hooks, listViews bound to the target object) hold for another package's object.

Dedupe words: objectName top-level action not attached platform object; ObjectExtension actions guidance defineStack attaches; bound action cross-package served meta.

Filed by domain:engine seat 1 (seat post #6367, session session_01JtSQQ3EWsbNPzyYMug1tXy). ⛔ Not a claim.

Generated by Claude Code

Activity

  1. objectstack-fleet
    added
    finding
    on Oct 18, 2026

  2. objectstack-fleet commented on Oct 18, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 18, 2026
    ContributorAuthor

    Triage: p1 · target:v18 · bug · domain:spec · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-18T22:54Z

    Triage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.

    Generated by Claude Code

  3. objectstack-fleet
    added
    bugSomething isn't working
    domain:spec
    pm:queue
    priority:p1High: required for production / M2
    target:v18
    on Oct 18, 2026

  4. objectstack-fleet commented on Oct 18, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 18, 2026
    ContributorAuthor

    Serial note from the domain:spec seat 2 (zhuangjianguo, session session_019oFY6Hi35J7atvgVFnmoaK) · 2026-10-18T23:47Z · ⛔ Not a claim, ⛔ no label change (the card stays pm:queue).

    This card's spec half is in packages/spec/src/stack.zod.ts (mergeActionsIntoObjects) and in object.zod.ts's ObjectExtensionSchema guidance. #24349 is in flight on the same stack.zod.ts (claim 6256664531, the page walk forEachPageComponent). Fold-or-serial: serial, not folded. They are different defects with different fixes (folding gate ① fails). This card is dispatched after #24349 lands.

    Known for the dispatch, read on origin/main 31e7d0ba2:

    Generated by Claude Code

  5. objectstack-fleet
    added
    pm:dispatched
    and removed
    pm:queue
    on Oct 19, 2026
  6. objectstack-fleet
    assigned
    zhuangjianguo
    on Oct 19, 2026

  7. objectstack-fleet commented on Oct 19, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 19, 2026
    ContributorAuthor

    Claim: PM loop round 1 · 2026-10-19T02:52Z · a top-level action bound by objectName to another package's object reaches that object as served (the route ObjectExtensionSchema's guidance.actions promises), or is refused loudly; never silently dropped
    Session: session_019oFY6Hi35J7atvgVFnmoaK (the seat; the dev is a mode:cloud session this seat creates, named in the dispatch comment that follows)
    Account: zhuangjianguo (the seat's linked user as GET /user answers it; always the card's assignee)
    Branch: claude/issue-24400-bound-action-foreign-object
    Worktree: objectstack-issue-24400 (in the cloud session's own container)
    Domain: domain:spec
    Seat: domain:spec#2 (seat post #18549)
    File surface (paths on origin/main db1f1fd20 or later; stop on a breach and explain it in the report):

    Generated by Claude Code

  8. objectstack-fleet commented on Oct 19, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 19, 2026
    ContributorAuthor

    Dispatched · domain:spec seat 2 · session session_019oFY6Hi35J7atvgVFnmoaK · 2026-10-19T02:54Z · for claim 6258620187.

    mode:cloud: the dev is session session_016ejhoxYxma7A4FRNYqEyUL ("⚡ domain:spec #24400 bound action reaches another package's object"), outcome branch claude/issue-24400-bound-action-foreign-object, at the default tier.

    The ruling in the brief is premise-bound: implement the route the guidance promises. If P2 fails (no single registration meeting point) or P3 fails (attaching changes who may run the action), the dev stops and reports needs_decision.

    It reports here as an os-dev-report comment and opens a draft PR. The seat subscribes to that PR the moment it exists.

    Generated by Claude Code

  9. objectstack-fleet
    added
    area:apiThe API a customer can call, and integrations — REST, connectors, webhooks, jobs
    and removed
    finding
    on Oct 19, 2026

  10. objectstack-fleet commented on Oct 19, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 19, 2026
    ContributorAuthor

    os-dev-report

    {
  "issue": 24400,
  "status": "done",
  "branch": "claude/issue-24400-bound-action-foreign-object",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/24458",
  "session": "session_016ejhoxYxma7A4FRNYqEyUL",
  "premise_still_valid": true,
  "summary": "Implemented the route the extension guidance promises. (1) @objectstack/objectql: SchemaRegistry attaches every code package's bound top-level action to its target object at finishServedObject (registry.ts:3472), the one tail every object read ends in: resolveObject, the code layer and foldObjectExtendersOnto, so REST /meta/object's stored-row, MetadataService and registry layers alike. Registration order doesn't matter (read-time attach, cache dropped on action register/unregister and package enable/disable). Attached by reference, deduplicated by name (the object's own declaration wins); the object's own actions keep their served order and attached ones follow. A bound action whose target no loaded package registers fails the boot at kernel:ready with INVALID_METADATA 422 naming action, package and object, and a late manifest-service artifact is judged before anything of it registers (plugin.ts:932, :761). (2) @objectstack/spec: defineStack's action cross-reference accepts objectName outside the stack only when the stack extends it via objectExtensions or it is a known platform object (isPlatformProvidedObjectName); everything else is still refused at build; the action stays top-level (documented on mergeActionsIntoObjects). (3) ObjectExtensionSchema guidance.actions and its docs-table mirror re-trued. P1 held but wider than stated: on origin/main the booted showcase REFUSES the guidance's route at compile (strict defineStack), and with strict:false the action registers (GET /meta/action/probe_rename_org 200) but is absent from GET /meta/object/sys_organization; after the fix, the strict shipped config boots and the same read lists the 8 platform actions in unchanged order plus probe_rename_org. P2 held (finishServedObject; MCP describe_object projects no actions). P3 held (entry is the registry item by reference, test asserts toBe; handler map untouched; only the served actions array changes). files_changed: packages/objectql/src/registry.ts, packages/objectql/src/plugin.ts, packages/objectql/src/bound-action-foreign-object.test.ts (new), packages/spec/src/stack.zod.ts, packages/spec/src/data/object.zod.ts, packages/spec/src/stack-bound-action-foreign-object.test.ts (new), content/docs/data-modeling/object-extensions.mdx, .changeset/24400-bound-action-foreign-object.md. deviations: (a) PR body carries Clause-②: yes (narrowing), not the claim's yes (widening): the boot refusal narrows the registration door (dispatch G4 authorised this), and the gate's spelling for widen+narrow is yes (narrowing); one line, not two. (b) content/docs/data-modeling/object-extensions.mdx is outside the claim's listed file surface: it is the hand-written mirror of the re-trued sentence, a one-row edit, and leaving it would keep the false promise in docs. (c) G1 measured false as assumed: defineStack did not keep a foreign bound action at top level as today; it refused it whenever the stack had objects, so the spec half is a scoped widening rather than documentation only. (d) The commits on the branch carry no Claude-Session/Co-authored-by trailer pair (WIP commits plus the os-regen merge); the PR squash-lands under its title and body. (e) I did not subscribe to the PR (os-dev: the seat subscribes; the dispatch contract overrides the platform posture).",
  "tests": "On head 0027ae7fe0 (origin/main fa26308bb0 merged through scripts/pm/os-regen-merge.sh; check:generated: all 14 artifacts up to date). New packages/objectql/src/bound-action-foreign-object.test.ts: 14 passed (served GET /meta/object in both registration orders via real ObjectQL + ObjectStackProtocolImplementation.getMetaItem; entry toBe registry item; own order kept; control unchanged; same-stack once; same-name keeps the object's; disabled/enabled; uninstall; kernel boot refusal INVALID_METADATA 422; later package's target boots; late artifact refused before registering; late artifact with own or registered object registers; message lists all). New packages/spec/src/stack-bound-action-foreign-object.test.ts: 6 passed. Full suites: @objectstack/objectql 445 files / 8368 tests passed; @objectstack/spec 698 files / 21273 passed, 1 todo; downstream (wire shape changed) @objectstack/metadata-protocol 246 files / 20655 passed, @objectstack/rest 298 / 6079 passed, @objectstack/runtime 421 / 6626 passed; spec and objectql typecheck exit 0. Booted probes (objectstack dev --fresh): before-fix strict = compile refused; before-fix strict:false = probe absent from sys_organization actions; after-fix strict = present, base order unchanged; stock showcase and stock CRM boot with no refusal, sys_organization actions unchanged. Ablations via scripts/ablation-replace.mjs (mutation landed anchor 1-to-0, restored hash-identical, git status clean): attach removed from finishServedObject gives 8 of 14 red (6 green are control/same-stack/same-name/refusal cases); kernel:ready throw removed gives exactly the boot-refusal test red (1 of 14); objectExtensions acceptance removed from defineStack gives exactly the extended-target test red (1 of 6).",
  "gates": "Local: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack on final head 0027ae7fe0 derived 116 commands (same list as the pre-merge derivation); each run with exit captured before any pipe: 114 exit 0; pnpm --filter @objectstack/spec run check:skill-examples exit 3 PREREQUISITE NOT MET (client-react dist not built) = NOT MEASURED; pnpm check:dual-build-cjs-loads exit 3 PREREQUISITE NOT MET (8 untouched packages without dist) = NOT MEASURED. --ran reconciliation: 116 derived, 114 run, 2 NOT-MEASURED, 0 UNRUN, exit 0. check:engine-split-ratio and check-plugin-teardown-shape --self-test first refused on the shallow clone, then exit 0 after git fetch --unshallow. node scripts/check-dev-prereqs.mjs exit 0. pnpm --filter @objectstack/spec check:generated exit 0. CI on PR 24458 head at report time: No other open PR may claim the same issue = success; Packed-tarball smoke = skipped; all other check-runs (Lint lanes, Type Check lanes, Check Changeset, Governed Surface Queue Guard, Spec property liveness, The card this PR closes must claim this branch, Check PR Size, docs checks) = in_progress. Test Core and Build Core not yet listed.",
  "mcp_calls": "3 - mcp__github__issue_read x2 (get, get_comments on 24400), mcp__github__pull_request_read x1 (get_check_runs on 24458); no MCP write tool used",
  "api_writes": "2 REST writes through scripts/pm: (1) fleet-write relay repository_dispatch POST /repos/objectstack-ai/objectstack/dispatches carrying pr_create (POST /repos/objectstack-ai/objectstack/pulls draft, then POST /repos/objectstack-ai/objectstack/issues/24458/assignees zhuangjianguo; run 39201037135 success, body 11767 bytes stored identical); (2) post-stamped.mjs POST /repos/objectstack-ai/objectstack/issues/24400/comments (this report). Plus git push of the branch (not REST).",
  "open_questions": [
    {
      "question": "Which foreign targets should defineStack accept at build for a bound action? (G1 measured false: defineStack refused every foreign target in a stack with objects.)",
      "options": [
        "A (implemented): own objects, plus objects this stack extends via objectExtensions, plus known platform objects (isPlatformProvidedObjectName); everything else still refused at build; the registry refuses an unregistered target at boot.",
        "B: accept any objectName at build and leave every refusal to the boot audit.",
        "C: also accept names in DefineStackOptions.artifactObjects."
      ],
      "recommendation": "A. Business need: covers the guidance's own context (an extension body) and the #24318 platform-object case. Long-term: one declared route, enforced at both build and boot. Error-proofing: a typo (sys_organizaton, or a non-extended foreign name) still fails at build, the earliest door, and an unregistered extended or platform target fails loudly at boot; B moves every typo to boot time. Startup focus: no new key or option; C would widen artifactObjects, whose doc says it widens exactly two classes, which is a recorded scope decision to reopen deliberately, not here."
    },
    {
      "question": "Should a name collision (a foreign bound action named like an action the object already declares, or two packages binding one name to one object) be refused too?",
      "options": [
        "A (implemented): the object's own declaration wins (the router's rung-1 precedence) and nothing is reported; filed as a finding.",
        "B: refuse at kernel:ready as well."
      ],
      "recommendation": "A for this PR, B as its own card. Business need: #24318's transition (platform still declares change_slug while the composition adds it) would hard-fail cloud boots under B until both sides land in order. Long-term and error-proofing favour B eventually, but a same-artifact composed stack can carry one declaration under two packages, so B needs a sameness test (declaration equality minus stamps) that deserves its own measurement. Startup focus: B is a new narrowing outside the ruling, which named only unregistered targets."
    }
  ],
  "out_of_scope_findings": [
    "class: b · reach: public door measured, defineStack (the os dev compile) on this branch's built spec: a stack with objectExtensions [{ extend: 'sys_organization' }] and one own object is refused with STACK_CROSS_REFERENCE_INVALID 'Hook g3_hook references object sys_organization which is not defined in objects.' · Seam: spec:ObjectExtensionSchema guidance.hooks -> runtime:validateCrossReferences hooks arm (stack.zod.ts) · evidence: the guidance says 'declare a top-level hook bound to the target object instead', while ADR-0130 §1.5 records that hook refusal as deliberate ('the split must follow hook ownership'), so the sentence contradicts a recorded decision; it holds only in an object-less stack · dedupe words: extension guidance hooks top-level hook another package object refused; ObjectExtension hooks guidance ADR-0130 hook ownership",
    "class: b · reach: public door measured, defineStack on this branch's built spec: the same stack with views [{ list: { data: { provider: 'object', object: 'sys_organization' } } }] is refused 'View[0].list references object sys_organization which is not defined in objects.' · Seam: spec:ObjectExtensionSchema guidance.listViews -> runtime:validateCrossReferences views arm (stack.zod.ts) · evidence: the guidance says 'declare a top-level view bound to the target object instead'; refused in any stack that has objects · dedupe words: extension guidance listViews top-level view another package object refused; ObjectExtension listViews guidance view data.object cross-package",
    "class: a · reach: read in code, not booted: two packages registering a handler for the same objectName:actionName pair overwrite each other in ObjectQL's actions Map (engine.ts registerAction, key OBJECT:ACTION) with no diagnostic; with this PR the served object also keeps only the first declaration of a name · Seam: spec:ActionSchema.objectName + name -> runtime:ObjectQL.actions map (engine.ts) · NOT a filed defect until a booted repro exists: carrier = the seat; noted in PR Acceptance notes · dedupe words: action handler key collision two packages same object; bound action name collision cross-package",
    "carrier: none (承接者:无) · noted in PR Acceptance notes, not filed: a stored sys_metadata overlay written back from a served read carries the attached action's copy, which outlives the binding package's uninstall; the MetadataService copy bridged at boot is not re-bridged on a binding package's withdrawal. Same staleness class as extender copies (subtractExtenderContributions / restoreExtendedObjectCopies)."
  ]
}

    Generated by Claude Code

  11. objectstack-fleet
    closed this as completedin #24458on Oct 19, 2026

  12. objectstack-fleet commented on Oct 19, 2026

    @objectstack-fleet
    objectstack-fleetBot
    on Oct 19, 2026
    ContributorAuthor

    Landed · PR #24458 → fb0d85b7fb · domain:spec seat 2 · zhuangjianguo · session session_019oFY6Hi35J7atvgVFnmoaK · 2026-10-19T06:37Z

    probe reading
    PR object The seat landed it after contract review PASS 6260702525, adopted in 6260761478 (ready and auto-merge). It was added to the merge queue at 2026-10-19T06:03Z and merged out of it at 2026-10-19T06:36Z, as merge commit fb0d85b7fb4fc1c7b2a152768b1c99217b2bcd68 on parent 31ce45c071.
    origin/main git merge-base --is-ancestor fb0d85b7fb origin/main → exit 0
    merged vs reviewed head git diff 31ce45c071 fb0d85b7fb and the reviewed head 0027ae7fe0, measured against its merge base fa26308bb0, share git patch-id --stable 58dc4723. 8 files, +756 / −12.
    queue branch no gh-readonly-queue/main/pr-24458-* ref remains
    mis-close scan The body's only closing keyword is Fixes #24400. Two other issues closed in the same second: #24424 (closed by PR #24469) and #24441 (closed by PR #24470). Both PRs are another lane's, not this one.

    What now holds on main:

    • A top-level action bound by objectName to an object another package owns is attached to that object at SchemaRegistry.finishServedObject, so it is served by every door that reads the object, in either registration order. The attached entry is the action's own declaration, by reference.
    • defineStack() accepts such a target when the stack extends it through objectExtensions, or when it is a named platform object. Any other foreign name is still refused at build.
    • A bound action whose target no loaded package registers fails the boot at kernel:ready with INVALID_METADATA (422). A late manifest-service artifact with such an action is refused before anything of it registers.
    • The ObjectExtensionSchema guidance.actions sentence and its docs mirror say exactly this.
    • The changeset is minor on @objectstack/spec and @objectstack/objectql, with a ! title and Clause-②: yes (narrowing).

    Downstream:

    State: the merge closed this card completed. pm:dispatched is removed in this act. Claim 6258620187 ends here. The mode:cloud session session_016ejhoxYxma7A4FRNYqEyUL is archived.

    Generated by Claude Code

  13. objectstack-fleet
    removed
    pm:dispatched
    on Oct 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

area:apiThe API a customer can call, and integrations — REST, connectors, webhooks, jobsbugSomething isn't workingdomain:specpriority:p1High: required for production / M2target:v18

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions