Flow designer: a time-relative start node (config.timeRelative, no triggerType) shows Trigger "—" and flags every sweep field "Kept, not in effect — its controlling field is off", though the sweep is live #12672

Activity

1. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   Triage: first grade, bug · priority:p2 · domain:ui · area:studio · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-16T14:58Z

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

   • Why p2: an admin is told that a live sweep's fields are "not in effect", which misstates the product.
   • Lane: domain:ui. The landing is packages/app-shell/src/views/metadata-admin/inspectors/flow-node-config.ts (the start group).
2. objectstack-fleet

   added

   area:studioChanging a running app without code — authoring, publish, docs and the portalChanging a running app without code — authoring, publish, docs and the portal

   bugSomething isn't workingSomething isn't working

   domain:uiobjectui ui stream: fix lands on the published library or apps — objectui execution seatobjectui ui stream: fix lands on the published library or apps — objectui execution seat

   pm:queue

   priority:p2

   pm:dispatched

   and removed

   pm:queue

   on Oct 16, 2026
3. objectstack-fleet

   assigned

   os-elon-musk

   on Oct 16, 2026
4. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 2
   Session: session_01Vk15bz8xHtKGUfXfCm16h4
   Account: os-elon-musk (the seat's linked user as GET /user answers it; the card's assignee)
   Branch: claude/issue-12672-time-relative-start-gates
   Worktree: objectui-issue-12672
   Domain: domain:ui
   Seat: domain:ui#3
   File surface: packages/app-shell/src/views/metadata-admin/inspectors/flow-node-config.ts: the start group's triggerType select and the gates of timeRelative.* and schedule.expression, plus isFieldVisible / inactiveRetainedKind / controllerAdmits only where a gate must read the resolved kind; FlowNodeInspector.tsx only where it hands the resolved flow kind to those functions or shows the Trigger select's value; tests beside each; one .changeset/12672-*.md (stop on breach; explain in the report)
   Container & model: S, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui over the named paths: no path-derived mandate)
   Clause-②: no (designer only; no export, prop or type changes, and the stored flow shape does not change)
   Responsibility: @object-ui/app-shell's flow node inspector gates the time-relative sweep fields and the schedule expression on triggerType: 'time_relative', a key the canonical shape never carries; the spec's resolveFlowTriggerKind decides by config.timeRelative's presence | none covers it: the flowKind gate exists but covers only secret (api) | every admin who opens a time-relative flow's start node in the Studio, measured on the stock showcase flow showcase_task_due_reminder (QA, objectstack-ai/objectstack#23878); the gates stand on main 8296c6c (flow-node-config.ts:499, :513, :522, :531, :540, :548)
   Thread-read: 6218262626
   Serial constraints cleared: objectui#12696 (os-tesla) changes the metadata-admin designer's own visibleOn dialect in anchors.ts, SchemaForm.tsx and form-spec.ts, with prose-only edits elsewhere; none of those is on this surface. objectui#12723 and #12721 (other seats) are PackagesPage.tsx and DiagnosticsPage.tsx. No in-flight domain:ui claim and no open objectui PR names flow-node-config.ts or FlowNode*, read across the lane's pm:dispatched claims and all 9 open PRs' file lists 2026-10-16T18:41Z.

   ---

   Generated by Claude Code
5. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   os-dev-report

   {
   "issue": 12672,
   "status": "done",
   "branch": "claude/issue-12672-time-relative-start-gates",
   "pr": "#12777",
   "session": "session_01Vk15bz8xHtKGUfXfCm16h4 — mode:subagent, so this is the dispatching seat's session (domain:ui seat 3), as the claim names it",
   "premise_still_valid": true,
   "summary": "Reproduced on unchanged main (base 7ead049), through the real FlowNodeInspector and in a real console. The console was objectstack 66d456d01 app-showcase plus this branch's apps/console dev server. On main, Task Due Reminder's start node shows Trigger reading an em dash and 4 inactive-retained banners: Sweep object, Date field, Offset days, Extra filter. Fix, in flow-node-config.ts and FlowNodeInspector.tsx: the timeRelative.* fields gain flowKind ['time_relative'], schedule.expression gains ['schedule','time_relative'] and condition gains FLOW_TRIGGER_KINDS, each beside its existing showWhen. controllerAdmits now admits a field carrying both gates when EITHER admits it; before, it needed both. The new unsetTriggerReading(field, node, flowKind) gives the option the Trigger select shows in its existing placeholder slot (muted, data-placeholder). It writes nothing: mount and unmount send zero patches, and an edit elsewhere adds no triggerType. The host hands the resolved kind only to the first top-level start node, which is the node the resolver and the engine read. Measured: FlowSchema parses a second start node, and before this change that node borrowed the first one's kind. PM hypotheses: (1) controllerAdmits DID need a change, since the old both-gates rule kept the canonical node flagged; ablation A1 shows it. (2) flowKind was the first start node's kind, now guarded. (3) the select reads through the placeholder slot fed by the field's defaultValue, on a render-only copy of the field. (4) no fork. Deviation from the claim's named gates: the Entry condition gate was fixed in place under the bounded exemption. Same defect class; the shape is fixed by the engine, which evaluates the start condition for every trigger type; same file and same tests. Without it, the select's new reading contradicted a sweep's condition banner on screen. The seat should amend the claim's file surface to name the condition gate. Conflict in the dispatch, called out rather than silently resolved: item 4 says a node whose triggerType IS set keeps today's behaviour, while the stale-block pin says the designer must agree with the spec. Measured: a schedule, manual or api token beside a timeRelative block resolves to time_relative, so that block is live, not stale. I followed the governing text: no banner there, and the banner is kept under a record-* token. See open_questions[0].",
   "tests": "Reproduction before any source edit, on base 7ead049: pnpm exec vitest run --maxWorkers=2 on the new FlowNodeInspector.timeRelativeStart-12672.test.tsx gave exit 1, 8 failed and 7 passed of 15. First failures read 'Sweep object: expected the banner element to be null' and "expected '—' to be 'Time-relative (date sweep)'". Final pins at head 1f02bfb, two files (that inspector file plus flow-node-config.timeRelativeStart-12672.test.ts): 48 passed, exit 0. The same 48 pins with main's two source files swapped in under a trap: 30 failed, 18 passed. 3 of the 30 are only the absent helper ('unsetTriggerReading is not a function'); 27 are behavioural. The 18 that pass on main are the guards: no write on open, explicit pick writes, record-* outranks a descriptor, secret gate unchanged, second start node. Restore checked: blob equals HEAD and git diff HEAD is empty. Ablations at 1f02bfb, each through objectstack scripts/ablation-replace.mjs (anchor hit as declared, marker landed on disk, restore checked blob-equal to HEAD). The subject is imported from src, so no dist sits in the path and no rebuild was needed. Failing pins out of 48: A1 either-gate reverted to both-gates 27; A2 sweep flowKind removed at 6 sites 13; A3 cron flowKind removed 6; A4 no Trigger reading 4; A5 write-on-open mutant 4, including 'a mount and unmount sends no patch'; A6 kind handed to every node 1, the second-start pin; A7 condition flowKind removed 8. All went red, the usual direction. The first runs of A2, A3 and A5 were no-ops: the tool refused each because its replacement overlapped the anchor, and no test ran. They were re-anchored and rerun. Real console, Chromium, before (main source swapped in) then after: Task Due Reminder 'Daily Sweep' went from Trigger '—' with 4 banners to Trigger 'Time-relative (date sweep)' as a placeholder with 0 banners and all six sweep fields plus Cron schedule shown. Scheduled Project Digest 'Every 60s (demo)' went from Trigger '—' with no fields to 'Schedule (cron)' as a placeholder with Cron schedule shown empty (finding 0).",
   "gates": "Local, objectui; the list was hand-derived by the seat because objectui has no dispatch-gates.mjs, and all six were run. Exit codes captured before any pipe; heavy runs went through os-verify-lock with slot 12672. (1) pnpm --workspace-concurrency=2 --filter '@object-ui/app-shell^...' run build: exit 0, scope 29 of 47 projects, run at d2d9bc9. Later commits touch only app-shell, which is outside that closure. (2) pnpm exec vitest run over all 66 test files that git grep finds for 'flow-node-config\|FlowNodeInspector', plus metadataError.heldEdit-11786.test.ts (another isFieldVisible reader): at 1f02bfb exit 0, 67 files, 1124 passed, 1 skipped. The skip is the existing skipIf in flow-node-config.spec-reconciliation.test.ts. (3) pnpm --filter @object-ui/app-shell run type-check: exit 0 at 1f02bfb. Its tsconfig.test.json step covers the new tests; an earlier run caught a missing readOnly prop in them, fixed in 7f6b05d. (4) eslint on the 4 changed .ts/.tsx files: exit 0 with 0 errors and 3 warnings. The same 3 warnings appear on base's FlowNodeInspector.tsx (two no-explicit-any, one set-state-in-effect), shifted 10 lines. (5) node scripts/check-changeset-presence.mjs: exit 0, 1 changeset declared, .changeset/12672-time-relative-start-gates.md (patch, @object-ui/app-shell). (6) pnpm check:new-line-citations: exit 0, 'VERDICT new-cross-file-line-citations: 0 new citation(s)'. Plus a control-byte scan of the changed files: no match. CI check-runs on 1f02bfb at PR open (one read, no polling): success — Build Docs, Changeset Bump Policy, Changeset Claim Re-read, Changeset Declaration, Changeset Fixed Group Check, Governed Surface Queue Guard, Line Citation Gate, label; skipped — Test (coverage), Test (coverage shard N/8), dependabot; in_progress — Build & E2E, Bundle Analysis, Changeset Overwrite Report, Lint, Live E2E (informational), Repo Gates, Test (dist pins), Test shards 1-16, Type Check.",
   "mcp_calls": "0",
   "api_writes": "2 relay strokes, each POST /repos/objectstack-ai/objectstack/dispatches, executed as objectstack-fleet[bot]. Stroke 1, pr_create: POST /repos/objectstack-ai/objectui/pulls opened #12777 as a draft, then POST /repos//issues/12777/assignees added os-elon-musk; read back 10223 bytes sent, 10223 stored, identical. Stroke 2: POST /repos//issues/12672/comments, this os-dev-report. Also 4 git pushes of the branch (empty probe, d2d9bc9, 7f6b05d, 1f02bfb). No label, assignee or claim writes on the card.",
   "open_questions": [
   {
   "question": "Item 4 ('a start node whose triggerType IS set keeps today's behaviour') conflicts with the stale-block pin ('the designer agrees with the spec'). For a schedule, manual or api token beside a timeRelative block, resolveFlowTriggerKind answers time_relative and the engine sweeps. Which wins?",
   "options": [
   "A, as shipped: follow the spec. No banner on the sweep fields under a schedule, manual or api token; the banner stays under a record-* token. The stored token still shows in the select. Cost: none further.",
   "B: restore the banners whenever any token is set. Cost: the designer again calls a live sweep 'not in effect' for those three tokens, contradicting the governing text."
   ],
   "recommendation": "A, because the governing text makes resolveFlowTriggerKind the designer's only reader of the kind, and the engine binds through it (deriveTriggerBinding)."
   },
   {
   "question": "The cron field now appears, empty, on schedule start nodes that store no triggerType, wherever the cadence is a form it cannot read (interval or bare string). Measured on the stock showcase's Scheduled Project Digest. Keep it in this PR or hold it back?",
   "options": [
   "A: land as is and file finding 0 for the read. Cost: an empty Cron schedule over an interval or bare-string cadence until the read lands. It is the same field explicit-Schedule nodes already show today.",
   "B: drop schedule.expression's flowKind from this PR until the read is fixed. Cost: the canonical cron-object schedule node keeps its false banner, and the seat's pinned item 2 goes undone.",
   "C: widen this PR to read a bare cron string. Cost: a claim-surface amendment (getFieldValue is outside it), and the interval form still needs a decision about an interval control."
   ],
   "recommendation": "A, because the gate is now correct against the spec. The read gap already exists and needs its own decision for the interval form."
   }
   ],
   "out_of_scope_findings": [
   "class: a · reach: public door, Studio flow designer on the stock showcase flow showcase_scheduled_digest, start node 'Every 60s (demo)', cadence { type: 'interval', intervalMs: 60000 }: Cron schedule reads empty, measured in a real console at this PR's head; on main the field is hidden. app-todo task_reminder / overdue_escalation (schedule: '0 8 * * *') read value '' in the inspector harness · evidence: the start group's schedule.expression field reads only config.schedule.expression or the flat config.cron (getFieldValue walks into a string and returns undefined). Its own code comment claims 'a bare-string or object-shaped schedule renders through it'. trigger-schedule's normalizeSchedule accepts a bare cron string, { cron }, { expression } and interval forms, and typing a cron into an interval cadence merges expression beside type 'interval', which normalizeSchedule resolves as the interval · Seam: spec:config.schedule (start node, shape open) → runtime:trigger-schedule normalizeSchedule | renderer:flow-node-config start schedule.expression · dedupe words: cron schedule field empty interval, bare cron string start node, normalizeSchedule designer, schedule.expression getFieldValue",
   "class: c · reach: named producers — the designer's own Object field on a schedule start node (its help reads 'Target object for record / scheduled-scan triggers'), and objectstack examples/app-todo task_reminder and overdue_escalation, which store objectName: 'todo_task' on schedule start nodes · evidence: the start group's objectName showWhen list names schedule and api, but the engine's schedule binding lifts no object and trigger-schedule's ScheduleTrigger never reads objectName. The list omits time_relative, where deriveTriggerBinding reads config.objectName as the sweep object's fallback. On a token-less schedule node, app-todo's objectName now reads 'Kept, not in effect — its controlling field is off' beside a select that shows Schedule (harness reading); with an explicit Schedule token it reads as live. Same family as this card (a start-node gate versus the resolved kind): fold it into that family's close-out card if one exists rather than a single-point card · Seam: spec:resolveFlowTriggerKind → runtime:AutomationEngine.deriveTriggerBinding (schedule case lifts no object; time_relative case falls back to config.objectName) | renderer:flow-node-config start objectName showWhen · dedupe words: objectName schedule trigger ignored, start node Object field schedule, scheduled-scan objectName, objectName time_relative fallback",
   "carrier: 承接者:无 · noted in the PR's Acceptance notes, not filed: (i) the Entry condition's select list omits manual, though the engine also evaluates the start condition on a manual execute(); a read-only inference, not reproduced in the UI. (ii) the Trigger select's schedule, time_relative and manual tokens are written by the designer and read by neither resolveFlowTriggerKind nor the engine; they record authoring intent only."
   ]
   }

   ---

   Generated by Claude Code
6. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   Seat answers to the dev's open questions · domain:ui seat 3 · session_01Vk15bz8xHtKGUfXfCm16h4 · 2026-10-16T20:12Z

   Report 6223747416 (PR #12777). Both answers fall in the seat's no-escalation class: they settle the order and shape of a technical task under governing text the dispatch already cited.

   Question 1 — answer A, as shipped.

   • The conflict is the seat's own. The dispatch's item 4 ("a start node whose triggerType IS set keeps today's behaviour") was worded wider than its Governing text ("resolveFlowTriggerKind is the spec's one reader of a flow's trigger kind").
   • The dev measured that a schedule, manual or api token beside a timeRelative block resolves to time_relative, and that the engine sweeps it. So that block is live, and the designer must not call it "not in effect".
   • The Governing text wins. Under a record-* token the stored sweep fields keep their banner, and under any token the select shows the stored value.

   Question 2 — answer A: land as is.

   • The empty Cron schedule over an interval or bare-string cadence is the field's existing read gap. On main it already shows the same way on a node that stores an explicit schedule token.
   • Option B would keep the canonical cron-object node's false banner, which the dispatch's item 2 fixes. Option C needs an interval control, and that is a product call.
   • The read gap is filed as objectui#12779, for triage, with the interval question named in it.

   The Entry condition gate. It is the same defect class, in the same file, under the same gates, and its shape is pinned by the engine, which evaluates the start condition for every kind. The claim is re-stated to name it. That is an in-place fix, with all four conditions read in the re-statement.

   Out of scope, where each went.

   • The objectName gate disagreeing with the engine's bindings (offered on schedule, missing on time_relative) and the condition list's manual → objectui#12780, the family card after this one, with an enumeration pin over the start group.
   • The Trigger select's schedule / time_relative / manual tokens, which nothing reads → the PR's acceptance notes.

   ---

   Generated by Claude Code
7. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 2 (re-stated by the same session on the same branch: the file surface gains the start group's Entry condition gate as an in-place fix of the same defect class, after report 6223747416; nothing else changes)
   Session: session_01Vk15bz8xHtKGUfXfCm16h4
   Account: os-elon-musk (the seat's linked user as GET /user answers it; the card's assignee)
   Branch: claude/issue-12672-time-relative-start-gates
   Worktree: objectui-issue-12672
   Domain: domain:ui
   Seat: domain:ui#3
   File surface: as claim 6222333691 declared, plus the start group's condition (Entry condition) gate in packages/app-shell/src/views/metadata-admin/inspectors/flow-node-config.ts, which is gated on the resolved kind like the others (stop on breach; explain in the report)
   Container & model: S, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui over the named paths: no path-derived mandate)
   Clause-②: no (designer only; no export, prop or type changes, and the stored flow shape does not change)
   Responsibility: unchanged from claim 6222333691. The seat's answers to the report's two open questions are 6223829818.
   Thread-read: 6223829818
   Serial constraints cleared: in-place fix, all four conditions read by the seat:

   1. the same defect class: a start-node field gated on the select alone, so a live value reads "not in effect" on a node that stores no triggerType;
   2. a pinned shape: the engine evaluates a start node's condition for every trigger kind, so the gate admits every kind the spec resolves;
   3. no other claim or open objectui PR names flow-node-config.ts, read across the lane's pm:dispatched claims and all 6 open PRs' file lists 2026-10-16T20:13Z;
   4. the same gate family and the same test files.

   ---

   Generated by Claude Code
8. objectstack-fleet commented on Oct 16, 2026

   

   objectstack-fleetBot

   on Oct 16, 2026

   ContributorAuthor

   More actions

   ACCEPT → PR #12777 comment 6223950275 · domain:ui seat 3 · session_01Vk15bz8xHtKGUfXfCm16h4 · 2026-10-16T20:20Z. The seat lands it through the relay (pr_ready + automerge_enable); on MERGED, Fixes closes this card.

   ---

   Generated by Claude Code
9. objectstack-fleet

   closed this as completedin #12777on Oct 16, 2026
10. objectstack-fleet

    removed

    pm:dispatched

    on Oct 16, 2026
11. objectstack-fleet commented on Oct 16, 2026

    

    objectstack-fleetBot

    on Oct 16, 2026

    ContributorAuthor

    More actions

    Landed · domain:ui seat 3 · session_01Vk15bz8xHtKGUfXfCm16h4 · 2026-10-16T20:33Z

    • PR fix(app-shell): a time-relative start node's sweep, cron and entry-condition fields follow the trigger the spec resolves (objectui#12672) #12777 merged through the merge queue as 8cc02b4 on main (2026-10-16T20:31Z). Head 1f02bfb: ACCEPT 6223950275. The seat's answers are 6223829818, and the claim re-statement is 6223842891.
    • Read by content on main at 8cc02b4:
      • packages/app-shell/src/views/metadata-admin/inspectors/flow-node-config.ts carries flowKind: ['time_relative'] on the six sweep fields, ['schedule', 'time_relative'] on the cron field and FLOW_TRIGGER_KINDS on the Entry condition (:499, :523), and defines unsetTriggerReading (:1851);
      • FlowNodeInspector.tsx hands the resolved kind to the first top-level start node only (kindStart, :364) and feeds the select's reading (:731).
    • This card closed completed on the PR's Fixes. The seat strips pm:dispatched in the same stroke as this note.
    • Out of scope, as the ACCEPT records: the Cron schedule's cadence read is objectui#12779, and the start group's remaining gates are objectui#12780.

    ---

    Generated by Claude Code
12. akarma-synetal

    added a commit that references this issue on Oct 28, 2026

    fix(app-shell): a time-relative start node's sweep, cron and entry-co…

    ...

    8cc02b4