spec parity: the ten blocks @objectstack/spec 17.5.0 newly carries, with their unpublished inputs, refused arms and member pins — declare each by measurement (objectui#11111 decision 3 = B)

[objectstack-fleet]

Filing gate: ③ maintainer-directed. objectui#11111 decision 3 = B, ruled by the maintainer 「11111 同意」 (record 5902351047). The ruled option names this card: "objectui#8649, objectui#8652 and one new triage card (the ten blocks and the keys with no card) declare each key by its own measurement". Filed by domain:devx seat 2 (objectui#10917), session_01TdiauJaVCHuj45EzZGUxHh.
Unblocked: #11073 closed (PR #11086 merged); see 5903802870.

• Who acts on it: the domain:spec seat, once PR chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086 (the 17.5.0 bump, objectui#11073) lands. The work needs 17.5.0 installed on main.
  • Until then, chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086 carries ledger entries in the two guards, each naming this card and an expiry.
  • Each slice this card lands strikes its entries. The last owner card to land (this one, objectui#8649 or objectui#8652) restores the empty ledger and cap 0.
• Duplicate check:
  • Titles of all domain:spec issues, open and closed (290), against action:button|action:group|action:menu|action:icon|definition-list|element:repeater|UNJUDGED|registry-inputs|spec-carried|unjudged|lazy block|object-gantt|object-map|object-timeline|object-tree: 3 hits. objectui#10289, finding(types): ObjectTreeSchema declares no filter on either face, yet filter is delivered to the object-tree node through ListView's baseProps spread #9549 and plugin-gantt/types: the registered gantt node spelling has no declared schema type — ObjectGanttSchema.type is the literal 'object-gantt', so a typed author cannot write the key the registry (and the README's table) teaches #8008 are all closed, older and a different scope.
  • All open objectui issues (102) against 17.5.0|action:button|definition-list|element:repeater|registry-inputs|spec-carried|unjudged|never judged: objectui#10872 is a different defect (objectui validate refusing namespaced types), and objectui#11073 is the bump itself.
  • No card owns this population.

Population, measured on 17.5.0

Source: the objectui#11073 dev's report 5891483523, Q10. Guard: apps/console/src/__tests__/registry-inputs-spec-parity.test.ts.

Ten blocks newly spec-carried at 17.5.0 (specCarried 45 → 55):

• Now judged (6): action:button, action:group, action:icon, action:menu, element:definition-list, element:repeater.
• Registered lazily and not loaded by the guard file, so never judged (4): object-gantt, object-map, object-timeline, object-tree.

Spec keys not published as inputs:

• action:button, 22 keys: visible, disabled, params, description, openIn, endpoint, method, bodyExtra, bodyShape, operation, patch, confirmText, successMessage, errorMessage, refreshAfter, undoable, recordIdField, locations, toast, resultDialog, onSuccess, objectName.
• action:icon, 20 keys: the same list without undoable and recordIdField.
• action:group: location, visible.
• action:menu: size, visible.

A declared input the spec refuses: action:group.name.

Arms the spec refuses:

• action:group.actions as an object (the kind itself).
• action:group.size 'md'.
• action:menu.actions as an object (the kind itself).
• element:definition-list.columns '1' and '2'.
• object-form.layout 'inline' and 'grid'.

Member-shape pins owed: action:group.actions, action:menu.actions, element:definition-list.items, and element:repeater.fields, .filter and .sort.

Not this card's:

• navigation on object-kanban and object-calendar belongs to objectui#8652.
• enforceFieldSecurity, redactFields and requiredPermissions on record:details, record:highlights and record:related_list belong to objectui#8649.

How it is decided (the ruling's terms)

• Each key by its own measurement. Does the block's renderer actually read it?
  • Declare what the renderer honours.
  • Refuse or retire what it does not.
  • ⛔ Never batch-declare an input a renderer does not honour, since declared-but-not-delivered is the failure decision 3 = B exists to avoid.
• The refused arms narrow what is accepted to what the spec accepts. The four lazy blocks get loaded, so the guard judges them.
• Declaring inputs widens the public surface (Clause ②): each slice owes a contract-review record at CONTRACT_REVIEW_TIER.
• Slice freely by block family. Each landing strikes its own ledger entries in chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086's two guards.

[objectstack-fleet]

Pointer (one more item this card owns), from the objectui#11073 contract review (5903181734 on PR #11086, section ③). domain:devx seat 2, session_01TdiauJaVCHuj45EzZGUxHh, 2026-09-30T03:03Z.

• The ObjectTree tree DECLARE verdict is pinned as owed in the 8655 header, and no card carries it. It belongs with this card's object-tree slice: object-tree is one of the four lazily registered blocks booked here.
• The other un-filed observations from that review stay recorded as acceptance notes on objectui#11073 (5903204435, carrier 承接者:无); they are not this card's.

[objectstack-fleet]

Unblocked: Blocked-by: #11073 is closed (PR #11086 MERGED at 2026-09-30T04:00Z, merge commit 81f8498). main now installs @objectstack/spec 17.5.0, which is what this card's measurements need. pm:blocked → pm:queue for the domain:spec seat. Unlock scan by domain:devx seat 2 (objectui#10917), session_01TdiauJaVCHuj45EzZGUxHh, 2026-09-30T04:03Z.

• The ledger entries this card owns are live on main: 62 in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts, each "Expires 2026-10-30, or when objectui#11168 lands". The five caps in OBJECTUI_11111_LEDGER_CAPS are the counters to lower slice by slice.
• Also this card's: the ObjectTree tree pointer (5903232431).

[objectstack-fleet]

Claim: PM loop round 13 (slice 1 of this card: the action:* family)
Session: session_012UwY3ahMixEFkfTUxMVkYm
Account: huangyiirene (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-11168-action-family-inputs
Worktree: objectui-issue-11168
Domain: domain:spec
Seat: domain:spec#1. The maintainer, 2026-09-30, in this seat's chat, chose 「恢复，3 张一批 (Recommended)」 for the batch this card leads.
File surface: slice 1 takes the four action:* blocks' rows in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts: the 46 unpublished spec keys (action:button 22, action:icon 20, action:group 2, action:menu 2), the off-spec input action:group.name, the refused arms (action:group.actions as an object, action:group.size 'md', action:menu.actions as an object), and the member-shape pins action:group.actions / action:menu.actions. Each key is decided by its own measurement (does the block's renderer read it?), per decision 3 = B: declare what the renderer honours, refuse or retire what it does not. The work lands in the blocks' registrations and renderers (packages/core/src/registry/public-blocks.ts, packages/components/src/renderers/action/), the matching packages/types declarations and zod mirror where the key is typed there, this card's own ledger rows and the OBJECTUI_11111_LEDGER_CAPS / owner-count figures they move, tests, and one changeset. ⛔ Not element:definition-list, element:repeater, object-form.layout or the four lazily registered blocks (later slices). ⛔ Not objectui#8652's or objectui#8649's rows. If the slice is too large to review as one PR, stop at a block boundary and hand the rest back as a worklist. Stop on breach and explain in the report.
Container & model: L, mode:subagent, model: opus (dispatch-gates --tier --repo objectstack-ai/objectui: no path-derived mandate; the maintainer, 2026-09-29: 「你可以派 opus 开发，然后等项目总监契约复审，」)
Clause-②: yes — publishing a spec key as a block input widens the published authoring surface; refusing an arm the spec refuses narrows what is accepted. The PR waits as a draft for the director seat's contract review.
Thread-read: 5903802870
Serial constraints cleared: objectui#8649 (dispatched in the same batch) also edits registry-inputs-spec-parity.test.ts: its own nine rows, plus the two shared lines both cards move (the unpublishedKeys cap and the owner-count pin, { 'objectui#11168': 62, 'objectui#8652': 2, 'objectui#8649': 9 }). That file is not a SINGLE_CLAIM_PATHS path: whichever PR lands second merges main and re-derives those two lines. None of the five open PRs touches packages/core/src/registry/, packages/components/src/renderers/action/ or apps/console (files read at main 0ffc423b1). PR #11125 (objectui#6152) owes a round that edits objectql.zod.ts; this slice does not take object-form.layout, so it stays off that file.

[objectstack-fleet]

os-dev-report
{
"issue": 11168,
"status": "done",
"branch": "claude/issue-11168-action-family-inputs",
"pr": "#11185",
"session": "session_012UwY3ahMixEFkfTUxMVkYm — the dispatching domain:spec seat's session (subagent = parent's)",
"premise_still_valid": true,
"summary": "Slice 1 (the four action:* blocks) measured every booked key through the real SchemaRenderer, the renderer's forward and the real ActionRunner against the installed 17.5.0 rows, and landed as draft PR #11185 (head 05d9cfd, four commits). Declared: 20 keys on action:button (all but endpoint and undoable) and 19 on action:icon (all but endpoint), location/visible on action:group, size/visible on action:menu; retired action:group.name; narrowed both actions inputs to a list of objects and action:group.size to the spec's default/sm/lg/icon; registered member pins for both actions lists and the ten array/object-armed keys on each leaf. Held with measurement (entries stay booked, reason text now states it): endpoint on both leaves (the console's own api handler reads target, never endpoint) and undoable on action:button (Undo needs a host _rowRecord stash the block never writes) — see open_questions. Ledger: offSpecInputs 1 to 0, unpublishedKeys 57 to 14, refusedArms 5 to 2, memberPins 6 to 4, MEMBER_PIN_EXEMPTION_CEILING 7 to 5, owner count objectui#11168 62 to 13 (objectui#8652 2 and objectui#8649 9 untouched). One consequence outside the guard: public-block-binding-reach derives its population from blocks publishing objectName, so action:button/action:icon joined it and are ledgered in NO_DATA_REACH (on an action block the key is a dispatch target, not a data binding). Premise checks: the installed spec is 17.5.0 and its four rows declare exactly the card's key lists; the action-definition premise held (20 of action:button's 22 keys and 18 of action:icon's 20 are forwarded and acted on by the runner or the handler it dispatches to; visible and disabled are consumed by the renderer); no document in the repo writes an object actions or size: 'md' on these blocks, so no document needed correcting.",
"tests": "All from the repo root through os-verify-lock. Components suite 341 files in three batches: renderers/ui/notifications 'Test Files 151 passed (151) / Tests 1466 passed | 17 skipped'; src/tests halves 'Test Files 94 passed | 1 skipped (95)' and 'Test Files 95 passed (95)'. Console suite 134 files in three batches plus scripts/tests/check-action-forward-parity.test.ts and app-shell widget-dom-leak-sweep: 45, 47 and 44 files passed. Final union at HEAD 05d9cfd (guard, binding-reach, packages/components/src/renderers/action/, action-group.test.tsx, forward-parity script tests): 'Test Files 33 passed (33) / Tests 775 passed (775)'. New pins: action-button-icon-inputs-11168 47 passed, action-group-menu-inputs-11168 20 passed. type-check: @object-ui/components (tsc --noEmit && tsc -p tsconfig.test.json) exit 0 and @object-ui/console (tsc --noEmit && tsc -b tsconfig.node.json --force) exit 0 after building each dependency closure; --listFilesOnly lists both pin files in the components test project and the guard + reach test in the console project. Lint (declared narrowing): eslint --format json on the 8 touched TS files, 0 errors; warnings are pre-existing no-explicit-any / only-export-components on untouched lines (diff adds no any); eslint.config.js has no parserOptions.project/projectService, so not type-aware. Ablations via objectstack scripts/ablation-replace.mjs, predictions written first: (A) delete action:button's confirmText input — predicted and observed exactly 2 red ('every declared key is a published input …' with 'action:button does not publish confirmText', and the guard's 'action:button publishes every top-level key its spec props schema declares'), 275 green, anchor x1 to x0, blob fce9b14a0a47 to 97074e6ff9e1; (B) re-admit md on action:group.size — exactly 2 red (the size pin and 'action:group declares no arm the spec refuses outright'), 248 green, blob 26b31ec9a251 to 01c4266dad38; (C) drop confirmText from action:button's forward — 1 of 47 red (action:button's confirmText row) and check:action-forward-parity red ('does not forward 1 key the runtime reads: confirmText'), blob fce9b14a0a47 to 09421448fe69. Every leg restored: blob == HEAD and git diff HEAD empty; tree status empty afterwards. Gates at 05d9cfd, all exit 0: check-changeset-presence/no-major/fixed/overwrite/claims, pending-changeset-literals, check:control-bytes, check:new-line-citations (0 new), check:spec-symbols, check:action-forward-parity, check:doc-snippets ('679 of 679 block(s) judged, 0 failed'), check:doc-types, check:prompt-keys, check:handler-key-reads, check:installed-pin-claims, check:test-path-roots, check:phantom-deps, check:unreferenced-sources, check:doc-example-readers, check:i18n-keys, check:element-data-source-declaration, check:esm-specifiers, check:side-effects-array, check:self-import, check:component-surface-parity (report-only, no inputs finding on the four blocks); check-governed-queue-guard --test on the 10 changed paths: NOT GOVERNED; control-byte self-scan: no hit. NOT MEASURED: check-spec-range-floors, reason: needs a whole-workspace build and the diff touches no package.json; check:sdui-registration-pins, reason: needs a console bundle and the diff touches no sideEffects array; repo-wide pnpm type-check / pnpm test / pnpm lint are CI's.",
"mcp_calls": "0 — no MCP GitHub tool called",
"api_writes": "3 relay strokes (objectstack scripts/pm tools, each a POST /repos/objectstack-ai/objectstack/dispatches executed by fleet-write as objectstack-fleet[bot]): (1) fleet-write/dispatch.mjs pr_create — POST /repos/objectstack-ai/objectui/pulls (draft forced); (2) label-write.mjs — POST /repos//issues/11185/labels (needs:contract-review) and POST /repos//issues/11185/assignees (huangyiirene), read back MATCHES; (3) post-stamped.mjs — POST /repos//issues/11168/comments (this report). git push is not a REST write. Reads only otherwise (card, comments 5904860606, 5902351047, 5891483523, the PR read-backs).",
"open_questions": [
{
"question": "action:button.endpoint and action:icon.endpoint — the spec rows declare the key ('API endpoint for an api action, forwarded to the runner'). Measured: the block forwards it and the runner's built-in api executor reads it (executeAPI: action.api || action.endpoint || action.target; a probe on a runner with no api handler sent the request to the endpoint, control with target sent it to the target). But both console runtimes register their own api handler, which reads const target = action.target || action.name and never endpoint (forward-parity census: no read in useConsoleActionRuntime or RecordDetailView). On the product, an api action written with endpoint never calls it. The spec's own ActionSchema alias table already maps endpoint to target. The dispatch did not carry the four-axis framework, so the options below carry costs and a recommendation, not an axis analysis.",
"options": [
"A — keep it unpublished (as now) and ask upstream to refuse endpoint on the four action:* rows with the target prescription ActionSchema already gives. Cost: one spec change; objectui changes nothing; the booked entries stay until the spec moves.",
"B — teach the console api handlers to read endpoint beside target, then publish it. Cost: a behaviour change plus a second spelling of one concept (AGENTS.md #0.1).",
"C — publish it as the runner reads it. Cost: declared-but-not-delivered on the console, the failure decision 3 = B exists to avoid."
],
"recommendation": "A, because it keeps one spelling (target), matches the spec's own ActionSchema alias, and publishes nothing the product drops."
},
{
"question": "action:button.undoable — the spec row declares it ('Offer an Undo affordance after an update action'). Measured: the block forwards it, but the runner's operation: update path (the spec's declared form of an update) captures Undo only under if (action.undoable && rowRecord && writtenFields.length > 0), where rowRecord is a host _rowRecord stash this block never writes (probe: the def the block forwards gets undo null; the same def plus a stash gets an undo). The console runtime reads it under the same guard. Only RecordDetailView's own api handler honours it without a stash (action.undoable && isThisRecord && pageRecord), for a logical target on a record page.",
"options": [
"A — keep it unpublished (as now) pending a ruling on the fork below. Cost: the entry stays booked on this card.",
"B — a behaviour change so the runner's update path can capture Undo on the block path (for example from the record page's own record instead of a row stash), then publish. Cost: runtime change in @object-ui/core with its own pins.",
"C — ask upstream to narrow undoable off the action:button row. Cost: one spec change; the record-page api path loses a declared key it honours today.",
"D — publish it with a description naming the one path that honours it. Cost: declared-but-not-delivered for the spec's canonical update form."
],
"recommendation": "A now, then B or C by ruling. B if the product wants Undo on page-authored update buttons; C otherwise. D is the failure mode the ruling names."
}
],
"out_of_scope_findings": [
"carrier: 承接者:无 · scripts/check-action-forward-parity.mjs says in its header that locations 'is read only off the AUTHORED action … never off the forwarded def', but isRecordScopedAction(action) in packages/core/src/actions/serverActionHandler.ts and the console flowHandler read it off the forwarded def, and RUNTIME_CONSUMERS does not list that file (the extractor's documented behind-a-helper blind spot). Dormant: the gate is one-directional and every surface forwards the key. Noted in PR #11185 Acceptance notes, not filed.",
"carrier: 承接者:无 · the same gate's JUSTIFIED entries for undoable on action:icon/action:group/action:menu reason only from the console runtime's rowRecord guard; RecordDetailView's apiHandler reads action.undoable && isThisRecord && pageRecord with no stash. Read-only inference, reach unmeasured. Noted, not filed.",
"carrier: 承接者:无 · action-forward-parity.test.tsx's header calls recordIdField inert on these renderers; this slice measured it honoured on the script path with a single grid selection (resolveServerActionRecordId). Stale prose. Noted, not filed.",
"carrier: objectui#10872 · packages/types/src/zod/public-blocks.zod.ts still says the action:* blocks have 'no spec row yet'; 17.5.0 carries them. Noted, not filed.",
"carrier: 承接者:无 · @object-ui/types exports an ActionGroup interface whose required name is the key this slice retired; nothing imports it. Dormant. Noted, not filed."
]
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT, slice 1: PR #11185 at 05d9cfdfa (the action:* family), ready for the director seat's review. From the domain:spec @ objectui seat, session session_012UwY3ahMixEFkfTUxMVkYm.

Checked against GitHub, not the report (5907196304):

• Form. Draft, 10 files (+1389/−85), four commits with the model-free trailer pair only, needs:contract-review on, and assignee huangyiirene. git merge-tree against today's main: clean.

• Decision 3 = B, key by key. Declared:

  • action:button: 20 of 22 keys.
  • action:icon: 19 of 20.
  • action:group: location / visible.
  • action:menu: size / visible.

  Retired: action:group.name. Narrowed: both actions inputs, to a list of objects, and action:group.size, to the spec's four values. Each declared key was measured honoured through the real renderer, forward and runner. The three ablations each reddened exactly the predicted pins, and check:action-forward-parity went red on the forward leg. No repository document needed correcting.

• Ledger. offSpecInputs 1 → 0, unpublishedKeys 57 → 14, refusedArms 5 → 2, memberPins 6 → 4, and owner count objectui#11168 62 → 13. The other owners' rows are untouched.

• Overlap. PR feat(types,plugin-detail): declare the field-security triple on record:details / record:highlights / record:related_list and read it un-cast (objectui#8649) #11184 (objectui#8649) moves the same unpublishedKeys cap line and owner-count pin in registry-inputs-spec-parity.test.ts, as stated at dispatch. git merge-tree of the two heads conflicts there. Whichever lands second merges main and re-derives those lines, and owes a new record.

Answers:

1. endpoint on action:button / action:icon: A, keep it unpublished. The entries stay booked with the measured reason: the console's api handlers read target, never endpoint, and the spec's own ActionSchema alias table maps endpoint to target. That is a contract divergence inside the spec, so the fix is upstream: refuse endpoint on the four action:* rows with the target prescription. The objectstack card for it is this card's next step, filed with its own dedupe by the next claimant. It is not filed from a closing shift.
2. undoable on action:button: A, keep it unpublished for now. The entry stays booked. Which way to go is a decision: a runtime change so the block path can capture Undo (B), or narrowing the spec row (C). It stays on this card for a ruling. D (publish with a caveat) is the declared-but-not-delivered failure decision 3 = B exists to prevent.

Noted, not filed (each recorded in the PR's Acceptance notes):

• the forward-parity header's locations sentence;
• the undoable JUSTIFIED reasoning on the other three blocks;
• the stale "recordIdField inert" prose;
• public-blocks.zod.ts's 「no spec row yet」 (carrier objectui#10872);
• the dormant ActionGroup.name type.

Remaining on this card after this slice: element:definition-list, element:repeater and object-form.layout; the four lazily registered blocks and the ObjectTree tree pointer; and the endpoint / undoable entries above. The PR lands with Refs #11168.

[objectstack-fleet]

Booked on this card's remaining slices: action:button.size. From the domain:spec @ objectui seat, session session_012UwY3ahMixEFkfTUxMVkYm, 2026-09-30T10:07Z. This carries the director's ③ from record 5908412638 on PR #11185, which PASSed slice 1 at 05d9cfdfa.

• action:button publishes size as sm / md / lg. At 17.5.0 the spec declares default / sm / lg / icon / md on that block, so the manifest refuses default and icon where the contract accepts them. This predates slice 1 and is outside its claim, so it is not fixed there.
• It belongs with the remaining slices, beside the held endpoint / undoable entries: publish the enum the spec declares, measured against what the renderer honours (decision 3 = B).

Landing order for slice 1: PR #11184 (objectui#8649) is in the merge queue and edits the same registry-inputs-spec-parity.test.ts cap line and owner-count pin. PR #11185 waits for it to land, then merges main and re-derives those lines. At the same push, its Clause-② line becomes the one-arm form Clause-②: yes (narrowing), as the record suggests. It then owes a new record before the queue.

[objectstack-fleet]

Pointer from objectui's domain:ui seat 2 (session_011p7ikEivgXefNDaE5S5Uec). ⛔ Not a claim, and nothing here is relabelled: this card is its seat's.

For this card's object-tree measurement: PR objectui#11200 (objectui#8348, ACCEPT on that card) moves the tree's record-source arm to view-data, per the installed 17.5.0 ComponentPropsMap['object-tree'] row: objectName optional, data = ViewData, staticData = array. Two faces still say otherwise, and neither is moved by that PR:

• the registration inputs: plugin-tree's treeInputs declares objectName required, and no data / staticData. That is this card's half, by its own body.
• the types mirror: @object-ui/types ObjectTreeSchema (TS and zod) does the same. No card carries it yet. It is read in source only, with no public-door reach measured, so the seat did not file it (filing gate ①).

If this card's object-tree pass reads the mirror too, one measurement settles both: for example, objectui validate on a staticData-only tree with no objectName. A refusal there is the reach a mirror card needs.

_{domain:ui seat 2 · cross-seat pointer · 2026-09-30T10:37Z}

---

Generated by Claude Code