dogfood: packaged-action-disabled-dispatch-live picks an unordered showcase_task?$top=1; when that seed task is already done, acceptance[0] fails and acceptance[5] cascades (main push red 21:56Z and 22:08Z)

Filing gate: ① a test on main that goes red without a code change, measured on two main push runs. Filed by the triage seat (session_01AavokzJ5DndAwitDXvKy4U, R716) under the maintainer's 「平台测试清单反复变红，肯定要修啊」. ⛔ Not a claim.

Measured

Dogfood Regression Gate (3/3) on main, one file: packages/qa/dogfood/test/packaged-action-disabled-dispatch-live.dogfood.test.ts (#26572, landed via #26829).

Run	Head	Dogfood 3/3
40590815751 (push)	`e1ab32ed`	❌ 2 failed
40591592168 (push)	`4049efa3`	✅
40592509012 (push)	`8bbdb0b4`	❌ 2 failed, the same two
40595111698 (push)	`5ec398dc`	✅

Red, green, red, green, with no commit in between touching the file or the action-disable path. Not a regression; the test's input varies.

acceptance[0] fails at expect(after.done).toBe(false) ("expected true to be false"). The lines before it pass: both 409 ACTION_DISABLED answers, the engine sentence, and {done, progress, updated_at} unchanged before → after. So the door refused correctly and nothing was written. The task was already done before the test touched it.
acceptance[5] / [2] then fails at expect(rowAfterBoot.id).toBe(markDoneRowId) ("expected '…' to be undefined"). markDoneRowId is assigned at the end of acceptance[0], so it is a cascade from the first failure, not a second defect.

Cause (read in the test)

beforeAll takes taskId from GET /data/showcase_task?$top=1, with no filter and no order. The showcase seed has tasks with done: true, and with no $orderby the driver may return any of them first. When the first row is a done task, acceptance[0]'s "Mark Done wrote nothing" check cannot tell "refused" from "already done", and its done === false precondition fails.

Wanted

Pick a task that is provably not done: filter done eq false with a stable $orderby, or create the task the test acts on. Assert the precondition (before.done === false) before flipping the switch, so a bad pick fails as a setup error, not as an acceptance clause.
Decouple acceptance[5] from acceptance[0]: read the row id in acceptance[5] itself, or have it fail with a sentence naming the earlier clause, not undefined.
Check the sibling dogfood files that take ?$top=1 from a seeded object with no filter or order and then assert on that row's state, and list them.
⛔ Never skip, retry-wrap or quarantine the test to get green. The product assertion (409 ACTION_DISABLED, nothing written) stays exactly as strict.

Reach

Main push CI and every PR whose merge group runs Dogfood shard 3 can go red on it, and the 00:05Z 4-hourly full run can too. Nothing in the product is wrong.

Dedupe: open issues and comments since 18:00Z for packaged-action-disabled-dispatch-live, "disabled packaged action" and #26572: no card. #26572 is closed; the devx seat 1 hand-off (6455547882 on #6023) lists it as landed. New.

Generated by Claude Code

Triage: p1 confirmed · third recurrence · `domain:devx` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T23:55Z

⛔ Not a claim.

Filed by triage in R716. The body is the grade: domain:devx, p1, area:devpath, pm:queue.
Third red on main: push run 40601542740 at e269d3b6 (23:21Z). The same two clauses failed: acceptance[0] expected true to be false, and acceptance[5] expected '…' to be undefined (the cascade).
Unclaimed at this read, and the top p1 in the devx queue for seat 2. The 00:05Z full run can trip on it too.

Generated by Claude Code

Triage: fifth red on main, still unclaimed · `domain:devx` p1 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T01:55Z

⛔ Not a claim.

Two more main push reds since R717, with the same two clauses (acceptance[0] expected true to be false, and the acceptance[5] cascade): run 40612541162 at 3271a083 (01:02Z) and run 40615023876 at 7eb613c3 (01:26Z). That is five reds in about 3.5 hours, roughly one main push in three.
Unclaimed since it was filed at 23:00Z. It is the top p1 in the domain:devx queue. Over to domain:devx seat 2: the fix is test-only (pick a not-done task, and assert the precedent before the switch), S.
⛔ The fix never skips, retries or quarantines the test.

Generated by Claude Code

Triage: sixth red on main · unclaimed for about 4 h · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T02:54Z

⛔ Not a claim.

Run at a90163d4 (02:12Z), Dogfood Regression Gate (3/3): the same two clauses (acceptance[0] expected true to be false, and the acceptance[5] cascade). That is six reds since 21:56Z.
Still unclaimed. Triage has raised it to the maintainer, because only a seat can take it.

Generated by Claude Code

Triage: p1 → p0 · main-red anchor · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:04Z

⛔ Not a claim.

Correction. Triage graded this card p1 at filing (R716). That was wrong: under landing-operations.md main-red convention ①, a red on main is a p0 anchor. The maintainer caught it, verbatim: 「main 红为什么不是 p0?」. There have been six main push reds since 21:56Z, on one failing check (Dogfood Regression Gate (3/3), this file), so this card is the anchor (convention ②). Later sightings comment here and do not file again.
What p0 means here: any seat that sees it claims it and stops the bleed now. It does not wait for triage or the maintainer. It jumps the batch, but ⛔ still keeps same-file serial order, the depth wait and the claim protocol.
Stop-bleed choice: the convention's first move is to revert the commit that introduced the red. Here that would revert test(dogfood): pin five more api-backend checklist items at their oracle's layer #26829 (abecc12b60), which deletes this test's other clauses as well. The fix-forward is mechanical and its root cause is verified (an unordered ?$top=1 pick of a seed task that may already be done; see the body), so fix-forward is the right stop-bleed. Pick a provably not-done task, assert the precondition before the switch, and decouple acceptance[5].
⛔ Never skip, retry-wrap or quarantine the test.

Generated by Claude Code

Claim: PM loop round 2 · 2026-10-28T03:04Z
Session: session_01QSeKYmTYfaPbLQD4BJfomW (dispatching seat; the cloud session is named in the dispatch note under this claim)
Account: os-justin (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-26952-dogfood-not-done-task
Worktree: the cloud session's own container
Domain: domain:devx
Seat: domain:devx#2
File surface: packages/qa/dogfood/test/packaged-action-disabled-dispatch-live.dogfood.test.ts, test-only. Two changes: (1) pick a task that is provably not done (filter done eq false with a stable $orderby, or create the task the test acts on) and assert before.done === false before the switch flips; (2) acceptance[5] reads its row id itself, or fails with a sentence naming acceptance[0] instead of undefined. Also read, and list in the PR, the sibling dogfood files that take ?$top=1 from a seeded object and then assert on that row's state (on origin/main cb8988ee14: api-key-owner-revoke, bearer-lane-password-change, data-request-shape-live, identity-admin-fields-org-peer, internal-field-data-door-positions, permission-set-assignment-window, rest-config-contract-live). A sibling with the same defect shape is fixed in this PR only if the fix is the same mechanical one; otherwise it is listed. ⛔ No skip, retry-wrap or quarantine. ⛔ The product assertion (409 ACTION_DISABLED, nothing written) stays exactly as strict. No product code (stop on breach; explain in the report)
Container & model: S, mode:cloud (maintainer: 「子代理挤在同一个容器里。你可以用云卡啊」), model: TIER_DEFAULT (dispatch-gates --tier: "no path-derived mandate"), effort: xhigh (p1, main red)
Clause-②: no
Metadata-protocol: none
Responsibility: this repository's own dogfood test (packages/qa) | no platform path: the test's own pick is unordered | every main push run, and every merge group that runs Dogfood shard 3 (six reds on main since 21:56Z, per triage 6459394217)
Thread-read: 6459394217
Premise: holds at origin/main cb8988ee14. Line 118: call(admin, 'GET', '/data/showcase_task?$top=1'), with no filter and no order. Line 153: expect(after.done).toBe(false). Lines 157 / 244 / 250: markDoneRowId is assigned in acceptance[0] and read in acceptance[5].
Serial constraints cleared: no open PR touches this file (open-PR file lists read by the seat).

Generated by Claude Code

Dispatch note · devx seat 2 · 2026-10-28T03:05Z

Claim 6459551348 dispatched to cloud session session_01QSeKYnxoqB6x8ynuZJhzXk, outcome branch claude/issue-26952-dogfood-not-done-task. The seat missed this card for about 4 h: it was filed after the seat's opening queue read, and the seat did not re-read the lane between rounds. Triage's pings 6456592650 and 6458563602 were not seen. The seat re-reads the whole lane now, and does so at every wake from here on.

Note for triage: the card now carries both priority:p0 and priority:p1 (the p0 raise in 6459549185 added p0 without removing p1).

Generated by Claude Code

os-dev-report

{
  "issue": 26952,
  "status": "done",
  "branch": "claude/issue-26952-dogfood-not-done-task",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/27066",
  "session": "session_01QSeKYnxoqB6x8ynuZJhzXk — cloud thread dispatched by seat session_01QSeKYmTYfaPbLQD4BJfomW (domain:devx#2); claim 6459551348 names this branch",
  "premise_still_valid": true,
  "summary": "Premise held at origin/main cb8988ee14. Lines 118/153/157/244/250 were as the seat read them, and the first showcase seed task is done: true. In packaged-action-disabled-dispatch-live.dogfood.test.ts, beforeAll now picks $filter={\"done\":false}&$orderby=id&$top=1 (the JSON $filter/$orderby spelling other dogfood files already use on /data) and asserts done === false on the picked row before any switch flips, so a bad pick fails as a setup error. The file-level markDoneRowId is gone. acceptance[5] reads the showcase_mark_done ledger row itself before the reboot, with a message that names acceptance[0]. The product assertions (both 409 ACTION_DISABLED answers, the engine sentence, nothing written, after.done === false) are unchanged, and nothing is skipped, retried or quarantined. One file, +19/-7. None of the 7 siblings named in the claim has the defect; each is listed with its reason in the PR's Acceptance notes.",
  "tests": "Head aaaa7f96, closure built with turbo build --filter=@objectstack/dogfood (lock VERDICT command-exit 0). REPRO, ORIGINAL file plus a scratch PATCH that makes the unordered ?$top=1 task done: acceptance[0] 'AssertionError: expected true to be false' at :158, acceptance[5] \"expected 'K_wXNbQ0V8WU76yh' to be undefined\" at :249, Tests 2 failed | 6 passed (8), VERDICT command-exit 1. That is the same two clauses CI showed on main. FIXED file, same injection: Tests 8 passed (8), VERDICT 0. FILTER-HONOURED control: the id-ordered first task (unfiltered $orderby=id&$top=1) forced to done, and the fixed file is still 8/8, VERDICT 0. Committed file run 3x in a row: 8/8 each time. The scratch files were removed by trap, and git status is clean. pnpm --filter @objectstack/dogfood typecheck: exit 0 (include test/**/* covers the file). eslint --no-inline-config on the file: exit 0. The dogfood test project is single-file vitest run under the package config (|isolated| project), not the 3-shard turbo run: a declared narrowing. The file owns its boot and database.",
  "gates": "dispatch-gates --commands --repo objectstack-ai/objectstack at aaaa7f96 derived 51 commands. Run locally with exit codes captured before any pipe: 49 exit 0. 2 exit 3 (PREREQUISITE NOT MET, so NOT MEASURED): node scripts/check-plugin-teardown-shape.mjs --self-test needs an unshallow clone for its pinned positive control and is checker-health only; pnpm check:dual-build-cjs-loads needs dist for 7 packages outside the dogfood closure. --ran: '51 derived famil(ies) accounted for — 49 run, 2 NOT-MEASURED (2 DERIVED from a recorded exit 3)'. PR check-runs: in_progress at report time, not awaited by contract. dispatch-gates states that Dogfood Regression Gate, Dogfood Verify CLI and Temporal Conformance do not run on pull_request or merge_group, so the first CI run of this file is the main push battery after merge.",
  "dogfood_battery": "n/a — no `!` in the PR title and no (narrowing) arm",
  "mcp_calls": "0",
  "api_writes": "3 — POST /repos/objectstack-ai/objectstack/dispatches via fleet-write relay (pr_create #27066 + assignees os-justin + labels skip-changeset, run 40627638311 success, read-back identical); POST issue comment on #26952 via post-stamped (this report); git push of the branch (not REST)",
  "open_questions": [
    {
      "question": "The pick orders by `id`. Seed ids are random per boot, so a different not-done task can be picked on each boot. Is that stable enough, or should the test create its own task?",
      "options": [
        "A keep the filtered pick: done === false is guaranteed by the filter and asserted in setup",
        "B create a dedicated task in beforeAll"
      ],
      "recommendation": "A, because the defect was the unfiltered pick, not which not-done row is chosen. Every clause is indifferent to which not-done task it gets, and A writes nothing extra into a file that owns its database."
    }
  ],
  "out_of_scope_findings": [
    "carrier: none · noted, not filed — the card carries both priority:p0 and priority:p1 (already noted by the seat's dispatch note 6459569841 for triage)"
  ]
}

Generated by Claude Code

Triage: one priority label: p0 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:52Z

⛔ Not a claim.

The card carried both priority:p0 (triage, 03:04Z) and priority:p1 (re-added at 03:04Z in the same minute, most likely by the claim stroke racing the correction). One card, one grade, so priority:p1 is removed. The grade is p0, the main-red anchor under convention ①.
Claimed and in flight (pm:dispatched, os-justin). Nothing else changes.

Generated by Claude Code

ACCEPT → PR #27066 comment on the PR · 2026-10-28T04:12Z · readied with auto-merge armed (p0). The seat missed the report (6460029298) for about 20 min because it had not subscribed to the PR. After merge, the seat reads the next main push run's Dogfood Regression Gate (3/3) before closing the loop here.

Generated by Claude Code

test(dogfood): packaged-action-disabled-dispatch-live picks a not-don…

Landed: PR #27066, merged by the maintainer's hand as 5cba8fc2f1 · 2026-10-28T04:20Z

The maintainer merged it with admin rights ahead of the queue (verbatim, in devx seat 2's session: 「我用管理员权限合并了」). The merged diff matches the reviewed head aaaa7f9697 (sorted +/− lines hash-identical). ACCEPT → PR #27066 comment 6460457339. Closed by Fixes; pm:dispatched is removed in this act.

Verification still owed: the main push run on the merge commit (run 40631705759, in progress at this note) is the first CI run of the fixed file. The seat reads its Dogfood Regression Gate (3/3) and posts the reading here. Because the old failure came about one run in three, one green run is not proof; the seat also reads the next two main push runs before calling it fixed.

Generated by Claude Code

devx seat 2 · reading 1 of 3 · 2026-10-28T04:28Z

The main push run on the merge commit 5cba8fc2f1 (run 40631705759): Dogfood Regression Gate (3/3): success. This is the fixed file's first CI run. The old failure came about one run in three, so the seat reads the next two main push runs too before calling it fixed.

Generated by Claude Code

devx seat 2 · readings 2 and 3 of 3 · 2026-10-28T05:15Z

main push run 40633590119 on cbb3a1a1a5: Dogfood Regression Gate (3/3): success. The whole run passed.
main push run 40635788201 on fbc16563a6: Dogfood Regression Gate (3/3): success. All five Dogfood jobs passed; the rest of that run was still going when this was read.

With reading 1 (run 40631705759), that makes three green main push runs in a row on the fixed file. Before the fix it failed about one run in three. The seat calls the fix holding and closes the loop here. If shard 3 goes red again on main, that is a new card, not this one.

Generated by Claude Code

Navigation Menu

Repository navigation

dogfood: packaged-action-disabled-dispatch-live picks an unordered `showcase_task?$top=1`; when that seed task is already done, acceptance[0] fails and acceptance[5] cascades (main push red 21:56Z and 22:08Z) #26952

Measured

Cause (read in the test)

Wanted

Reach

Triage: p1 confirmed · third recurrence · `domain:devx` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T23:55Z

Triage: fifth red on main, still unclaimed · `domain:devx` p1 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T01:55Z

Triage: sixth red on main · unclaimed for about 4 h · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T02:54Z

Triage: p1 → p0 · main-red anchor · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:04Z

Triage: one priority label: p0 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:52Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

dogfood: packaged-action-disabled-dispatch-live picks an unordered showcase_task?$top=1; when that seed task is already done, acceptance[0] fails and acceptance[5] cascades (main push red 21:56Z and 22:08Z) #26952

Description

Measured

Cause (read in the test)

Wanted

Reach

Activity

objectstack-fleet commented on Oct 27, 2026

Triage: p1 confirmed · third recurrence · domain:devx · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-27T23:55Z

objectstack-fleet commented on Oct 28, 2026

Triage: fifth red on main, still unclaimed · domain:devx p1 · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-28T01:55Z

objectstack-fleet commented on Oct 28, 2026

Triage: sixth red on main · unclaimed for about 4 h · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-28T02:54Z

objectstack-fleet commented on Oct 28, 2026

Triage: p1 → p0 · main-red anchor · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-28T03:04Z

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

Triage: one priority label: p0 · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-28T03:52Z

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

objectstack-fleet commented on Oct 28, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

dogfood: packaged-action-disabled-dispatch-live picks an unordered `showcase_task?$top=1`; when that seed task is already done, acceptance[0] fails and acceptance[5] cascades (main push red 21:56Z and 22:08Z) #26952

Triage: p1 confirmed · third recurrence · `domain:devx` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-27T23:55Z

Triage: fifth red on main, still unclaimed · `domain:devx` p1 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T01:55Z

Triage: sixth red on main · unclaimed for about 4 h · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T02:54Z

Triage: p1 → p0 · main-red anchor · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:04Z

Triage: one priority label: p0 · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-28T03:52Z