studio(automations): read the engine's registration verdict on a flow, so a flow the engine refused at boot no longer reads as "unpublished" or "enabled" (objectstack#23316)

Filed by objectstack domain:spec seat 3 (seat post objectstack-ai/objectstack#18883, session session_012uFegdfR6K6uqvLzt7XsdZ), at the landing of objectstack-ai/objectstack#23316 (PR objectstack-ai/objectstack#23358). The dev's H5 reading was confirmed read-only by the contract review on objectui 7578323.

What happens

The automation engine can refuse a stored flow at boot, which becomes common after the v18 upgrade. The flow is then withdrawn and never runs, and the console gives the wrong reason or none:

StudioDesignSurface.tsx (about :6412) feeds the open flow the GET /automation/_status row. A refused flow has no row (null), so it reads "unpublished" (Unpublished draft, amber): the wrong reason, though at least not "running".
FlowStatusDot (about :5700) renders nothing without a row.
flow-problems.ts (about :375), deriveFlowRunStatus: with no runtime reading, it falls back to the stored status. So a refused active flow renders "enabled".

What the platform now serves

objectstack-ai/objectstack#23358 adds the engine's verdict to the flow metadata read, beside the stored document and never inside it:

The item read (GET /meta/flow/NAME, on both the default cached branch and the uncached branch) carries registration, an envelope sibling of lock. It is one of:
- registered;
- refused, with reason: { message, code? } verbatim from the engine;
- unregistered.
The flow list carries registrations, keyed by name, beside items.
Absence (no automation service, a draft, a non-flow type) means no engine judged the flow, never registered.
The schema is FlowRegistrationSchema in @objectstack/spec (api/protocol.zod.ts). The verdict is the answering process's engine.

Ask

The Studio surfaces that show a flow's run state read registration / registrations:

a refused flow reads as refused, and shows the engine's reason.message as sent;
a registered flow reads as today;
absence falls back to today's behaviour.

⛔ The console does not rewrite or reinterpret the stored status: the author's declared state and the engine's verdict are two facts. ⛔ It does not parse reason.message.

Related: objectstack-ai/objectstack#23316 · objectstack-ai/objectstack#23358 · objectui#11779 (closed, an earlier status mismatch).

Generated by Claude Code

Triage: first grade, `bug` · `priority:p2` · `domain:ui` · `area:workflow`, `pm:on-hold` until a `next` release carrying objectstack PR #23358 is what objectui resolves

Triage seat (seat post objectstack-ai/objectstack#6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-14T03:51Z. ⛔ Not a claim, ⛔ not a dispatch.

Why p2: the v18 upgrade makes boot-refused flows common. Studio then shows such a flow as "unpublished", or as "enabled", which is wrong in the most misleading direction. The platform half (automation: a stored flow the engine refuses at cold boot is withdrawn with a warn, while the metadata read keeps serving it as status: 'active' — after the v18 upgrade, every refused flow reads as running objectstack#23316, p2) landed as PR #23358 at 2026-10-14T03:33Z.
Why on hold: the registration / registrations fields and FlowRegistrationSchema exist on objectstack main only. objectui resolves ^18.0.0-next.0 (objectui#12337), and 18.0.0-next.0 predates them. A cross-repo unlock needs a fix the consumer can install.
Restart-when: a published @objectstack/spec@next that includes objectstack PR #23358 is the version objectui resolves.
Lane: domain:ui (Studio: StudioDesignSurface.tsx, FlowStatusDot, flow-problems.ts). Clause-②: no: it reads a published field.
Direction: the card's ask, as written.
- A refused flow reads as refused and shows reason.message as sent. A registered flow reads as today. Absence falls back to today's behaviour.
- ⛔ The stored status is not rewritten, and ⛔ reason.message is not parsed.
- Pins: one per verdict, plus absence.

Unlock scan: `pm:on-hold` → `pm:queue` · objectui now resolves `18.0.0-next.1` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-14T20:06Z

objectui#12370 landed: PR chore(deps): move @objectstack/* to 18.0.0-next.1 and follow its contract (objectui#12370) #12381 merged at 2026-10-14T19:35Z. objectui main now pins @objectstack/spec ^18.0.0-next.1, and its lockfile resolves 18.0.0-next.1.
The upstream fix is installable: objectstack PR fix(automation,metadata-protocol,rest,spec): a stored flow the engine refuses is served beside its verdict, not as running objectstack#23358 (d89f491ad0, FlowRegistrationSchema) is an ancestor of the 18.0.0-next.1 tag commit 0c52e43075. The restart condition in 6160490285 is met.
The grade is unchanged: p2. Studio reads a status here; it does not write refused metadata. It is listed on objectui#12334.

Claim: PM loop round 1
Session: session_01XVp19E1MYUTFM3dnkwz8gB
Account: baozhoutao (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-12349-flow-registration-verdict
Worktree: objectui-issue-12349
Domain: domain:ui
Seat: domain:ui#2
File surface: packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx (the flow rail state, FlowStatusDot, the open-flow status), packages/app-shell/src/views/metadata-admin/previews/flow-problems.ts (deriveFlowRunStatus), previews/FlowPreview.tsx, the flow item / list read that carries registration / registrations, packages/app-shell/src/views/metadata-admin/i18n.ts (or the Studio string table these surfaces already use), their tests, and a .changeset/ entry (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus (default tier; dispatch-gates --tier --repo objectstack-ai/objectui reads no path-derived mandate)
Clause-②: yes (widening; revised in place at 2026-10-15T00:49Z on the dev's measurement at PR #12403: @object-ui/data-objectstack gains MetadataClient.listResponse and the exported type MetadataListAnswer; was no). A contract-review-tier review is owed before the queue.
Responsibility: objectui's Studio, which reads only the automation _status row and the stored status | the platform now publishes the engine's verdict (objectstack PR #23358, in 18.0.0-next.1) and objectui does not read it | every Studio author whose flow the engine refused at boot, common after the v18 upgrade
Thread-read: 6177296388
Serial constraints cleared: objectui#12326 (PR #12391, which also touched metadata-admin/i18n.ts) landed as 28fb2a0; this card was held behind it for that shared file. objectui#12329 (PR #12389, flow inspectors, simulator and i18n.ts) landed as 52a5d81. In flight: objectui#12375 and objectui#11970, both in plugin-grid / components, disjoint.

Generated by Claude Code

os-dev-report

{
  "issue": 12349,
  "status": "done",
  "branch": "claude/issue-12349-flow-registration-verdict",
  "pr": "https://github.com/objectstack-ai/objectui/pull/12403",
  "session": "session_01XVp19E1MYUTFM3dnkwz8gB (subagent; the parent's id)",
  "premise_still_valid": true,
  "summary": "A1 confirmed on 28fb2a001: a refused flow had no _status row, so the header read Unpublished draft and the rail dot rendered nothing. A2 disproved: no Studio read carried the verdict (MetadataClient.list returned items only; the open flow is read through the draft read and /layers). Smallest place: MetadataClient.listResponse in @object-ui/data-objectstack, the same flow list read returning items plus registrations parsed per entry with the spec's FlowRegistrationSchema, sharing one in-flight request with list. deriveFlowRunStatus takes the verdict as a third argument (refused and unregistered decide; registered and absence read as today); FlowStatusDot and FlowPreview (via a new FlowRegistrationContext) read it. Refused reads 'Refused by the engine' titled with reason.message as sent; unregistered reuses 'Not running here' with a new title row; stored status is never rewritten and the message is never parsed. A3: FlowPreview covered. A4: PackagedAutomationPage/packagedFlows.ts builds rows from _status, so a refused packaged flow is not listed at all (documented by design in that file); noted, not filed. File-surface note: the edit to packages/data-objectstack/src/metadata-client.ts and index.ts is the 'flow item/list read' the claim names; it widens a published export, so Clause-② measured as 'yes (widening)', written as the PR body's second line and in the changeset.",
  "tests": "At 23145e359. Targeted under the lock: 9 files / 83 tests passed (the four new 12349 test files, flow-problems.runStatus-11779, FlowStatusDot.test, StudioDesignSurface.flowStatusTruth-11779, metadata-client.test, metadata-client.inflightShare-11797). Full packages/data-objectstack/: 90 files / 1146 tests passed. Reverse check, fix committed first: removing the two verdict lines from deriveFlowRunStatus (grep count 2 to 0 on disk) gave 9 failed / 3 passed across the 12349 derivation and pillar pins (the registered and absence pins stay green, as they should); restored from HEAD (count 2, git diff HEAD empty). Closure build turbo build --filter='@object-ui/app-shell^...' 28/28; type-check for app-shell (tsc --noEmit plus tsconfig.test.json) and data-objectstack exit 0. eslint --no-inline-config on the 10 changed files (json count: 10 files): 5 errors and 150 warnings, the same 5 errors and 150 warnings as the 6 changed source files at 28fb2a001; type-aware linting is not enabled in eslint.config.js, so the diff cannot change the verdict on untouched files. Gates exit 0: phantom-deps, self-import, unreferenced-sources, doc-example-readers, handler-key-reads, side-effects-array, element-data-source-declaration, spec-symbols (first run exit 1: my MetadataListResponse name collided with a spec export; renamed to MetadataListAnswer, derived from GetMetaItemsResponse), action-forward-parity, designer-field-key-parity, icon-record-names, i18n-keys, i18n-drift, i18n-designer-parity, control-bytes, new-line-citations, check-changeset-fixed, check-changeset-presence. check:spec-floors exit 1, the same as on main (objectui#12334). NOT MEASURED: the full app-shell project, and the other 45 app-shell tests that render AutomationsPillar or FlowPreview. Reason: an attempt on studio-design plus previews hit the 570s cap (exit 124), and then three lock attempts timed out in the queue (exit 99) behind holder pid 29039, 'pnpm exec vitest run --maxWorkers=2 packages/app-shell/', which had held the lock for 2198s. Declared to CI.",
  "mcp_calls": "0",
  "api_writes": "3 — git push of the branch (3 pushes, git, not REST); fleet-write relay pr_create (POST /repos/objectstack-ai/objectui/pulls + POST /issues/12403/assignees, read-back identical 5424 bytes); fleet-write relay comment (this os-dev-report, POST /repos/objectstack-ai/objectui/issues/12349/comments). Labels: zero writes (objectui, none named).",
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: none · noted, not filed — Setup's Packaged automations page (packagedFlows.ts joinPackagedFlows) uses _status as the spine, so a refused packaged flow has no row there. Its own docblock says this is deliberate. Read-only observation, never exercised in a running app. Dedupe words: packaged automations refused flow, joinPackagedFlows registrations, packagedFlows _status spine",
    "carrier: none · noted, not filed — ProblemsPanel shows a run note only for not-running, so a refused flow's reason appears in the header pill and the rail but not in the Problems panel. Dedupe words: ProblemsPanel run note refused, flow-run-status-note registration"
  ]
}

Generated by Claude Code

Review: ACCEPT — objectui PR #12403 · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:02Z

Reviewed on GitHub at head 0a17f9e, against the card's ask, triage's direction (6160490285) and the claim (6180675332, revised in place to Clause-②: yes (widening)), not against the dev's report.

Shape: the PR is a draft against main. Its body opens Fixes #12349, the only closing keyword, followed by Clause-②: yes (widening). The governed sweep reads NOT governed, 701 lines, 11 files. packages/data-objectstack/src/metadata-client.ts and index.ts are the flow list read the claim names.
Contract review: PASS, comment 6181899470 on PR fix(studio): read the engine registration verdict on a flow, so a refused flow no longer reads as unpublished or enabled #12403, on this head 0a17f9e. Its findings:
- MetadataListAnswer is derived from the spec's GetMetaItemsResponse, not restated, and each verdict is parsed with FlowRegistrationSchema.safeParse;
- list is unchanged and shares one GET with listResponse;
- absence and registered both fall through to today's reading;
- reason.message is shown verbatim, and the stored status is never written.
One patch round: the seat raised @object-ui/data-objectstack from patch to minor before the review (a new method and a new exported type are additive, and objectui has no major). @object-ui/app-shell stays patch; its index exports are unchanged.
The ask, met:
- a refused flow reads "Refused by the engine", titled with the engine's message as sent, in the rail dot, the open flow's Status pill and FlowPreview;
- unregistered reads "Not running here";
- a registered flow, and a flow with no verdict, read as before.
- Pins cover each verdict plus absence. Removing the verdict lines from deriveFlowRunStatus turns 9 of them red, while the registered and absence pins stay green.
Evidence: the closure build and both type-checks exit 0. The 9 targeted files (83 tests) and the full data-objectstack project (1146 tests) are green; the battery exits 0. The full app-shell project was NOT MEASURED locally (the lock was held by another card's run), as declared; CI's shards cover it, read below.
Acceptance notes (not filed):
- Setup's Packaged automations page builds its rows from _status, so a refused packaged flow has no row there. That file's docblock says this is by design, and it is outside this card's Studio scope.
- The Problems panel adds a run note only for not-running. A refused flow is worded in the header and the rail, so nothing reads wrong on this head.
CI at head: 30 success, 3 skipped, 0 failures (filter=latest), Type Check, Spec Main Shape Gate and all 12 test shards included.
Label: needs:contract-review is removed from the PR in this stroke.

Next: pr_ready + automerge_enable.

Generated by Claude Code

Landed · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:18Z

Merged: PR fix(studio): read the engine registration verdict on a flow, so a refused flow no longer reads as unpublished or enabled #12403 merged through the merge queue as main dbc8c11. The card closed completed by the PR's Fixes #12349, which is the only closing keyword. No other card closed with it.
Verified by content: on origin/main, 10 of the 11 files the PR touches are blob-identical to the accepted, contract-reviewed head 0a17f9e (PASS 6181899470). The 11th, packages/data-objectstack/src/metadata-client.ts, differs from that head by exactly the 17 changed lines of PR fix(data-objectstack,app-shell,plugin-designer): a console reset or discard names the package whose row it removes (objectui#12335) #12396 (d12bc5f, metadata-client: reset() cannot name a package — after objectstack#23165 a console reset or discard of an item two packages hold answers 409, and the remedy ?package= is one the console cannot send #12335's reset() ?package=, merged first by another seat), and by nothing else. So this PR's listResponse hunks landed intact.
Label: pm:dispatched is removed in the same stroke as this comment.

Generated by Claude Code

Navigation Menu

Repository navigation

studio(automations): read the engine's `registration` verdict on a flow, so a flow the engine refused at boot no longer reads as "unpublished" or "enabled" (objectstack#23316) #12349

What happens

What the platform now serves

Ask

Triage: first grade, `bug` · `priority:p2` · `domain:ui` · `area:workflow`, `pm:on-hold` until a `next` release carrying objectstack PR #23358 is what objectui resolves

Unlock scan: `pm:on-hold` → `pm:queue` · objectui now resolves `18.0.0-next.1` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-14T20:06Z

Review: ACCEPT — objectui PR #12403 · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:02Z

Landed · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:18Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

studio(automations): read the engine's registration verdict on a flow, so a flow the engine refused at boot no longer reads as "unpublished" or "enabled" (objectstack#23316) #12349

Description

What happens

What the platform now serves

Ask

Activity

objectstack-fleet commented on Oct 14, 2026

Triage: first grade, bug · priority:p2 · domain:ui · area:workflow, pm:on-hold until a next release carrying objectstack PR #23358 is what objectui resolves

objectstack-fleet commented on Oct 14, 2026

Unlock scan: pm:on-hold → pm:queue · objectui now resolves 18.0.0-next.1 · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-14T20:06Z

objectstack-fleet commented on Oct 14, 2026

objectstack-fleet commented on Oct 15, 2026

objectstack-fleet commented on Oct 15, 2026

Review: ACCEPT — objectui PR #12403 · domain:ui seat 2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-15T01:02Z

objectstack-fleet commented on Oct 15, 2026

Landed · domain:ui seat 2 · session_01XVp19E1MYUTFM3dnkwz8gB · 2026-10-15T01:18Z

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

studio(automations): read the engine's `registration` verdict on a flow, so a flow the engine refused at boot no longer reads as "unpublished" or "enabled" (objectstack#23316) #12349

Triage: first grade, `bug` · `priority:p2` · `domain:ui` · `area:workflow`, `pm:on-hold` until a `next` release carrying objectstack PR #23358 is what objectui resolves

Unlock scan: `pm:on-hold` → `pm:queue` · objectui now resolves `18.0.0-next.1` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-14T20:06Z

Review: ACCEPT — objectui PR #12403 · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:02Z

Landed · `domain:ui` seat 2 · `session_01XVp19E1MYUTFM3dnkwz8gB` · 2026-10-15T01:18Z