release(next): the Version Packages PR #21988 (18.0.0-next.0) is red in CI — check-future-spec-major and two metadata forward-conversion tests are not major-roll aware; fix on main, never on the generated branch #23135

Activity

1. objectstack-fleet

   assigned

   zhuangjianguo

   on Oct 13, 2026
2. objectstack-fleet

   added

   bugSomething isn't workingSomething isn't working

   priority:p1High: required for production / M2High: required for production / M2

   pm:dispatched

   domain:devx

   area:devpathThe road — create, dev, verify, publish/install, connect an agent, iterateThe road — create, dev, verify, publish/install, connect an agent, iterate

   on Oct 13, 2026
3. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetbot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 4 · domain:devx seat 1 · 2026-10-13T03:02Z · maintainer-directed (the card's first line quotes the order verbatim; this seat's session, 2026-10-13)
   Session: session_012zwsLMxQBwHUYJgPyrpEvx
   Account: zhuangjianguo (the seat's linked user as GET /user answers it; the card's assignee)
   Branch: claude/issue-23135-v18-version-pr-red
   Worktree: objectstack-issue-23135
   Domain: domain:devx
   Seat: domain:devx#1
   File surface: scripts/check-future-spec-major.mjs (its published-major reading, floor and QUOTATION_EXEMPTIONS, plus its self-test); packages/metadata/src/plugin-artifact-forward-conversion.test.ts, and the forward-conversion source it pins if the producer is there. Also any further red check's producer on #21988's head, declared in the report. ⛔ No governed surface (docs/adr/**, .claude/**, skills/**, AGENTS.md). ⛔ No packages/spec/src/** non-test. ⛔ No edit to changeset-release/main / #21988 (stop on breach; explain in the report)
   Container & model: M, mode:subagent, model: opus (the default judgment tier; dispatch-gates --tier on this surface: "no path-derived mandate", floor sonnet · default opus · ceiling fable)
   Clause-②: no
   Responsibility: scripts/check-future-spec-major.mjs and the @objectstack/metadata forward-conversion tests read the tree's version as "current" or "published", and both flip at the 18.0.0-next.0 roll (seat reading; not established) | no platform path covers it: main is at 17.x, and only #21988's tree carries the roll | #21988's CI reaches it on every refresh, so the next line is blocked
   Thread-read: none
   Serial constraints cleared: no open PR touches scripts/check-future-spec-major.mjs, packages/metadata/src/plugin-artifact-forward-conversion.test.ts, docs/adr/0087-* or docs/v17-docs-sweep.md (all open PRs' file lists read at this act). #23042 (open, spec pre-mode retiredAfter guidance) is related and unclaimed here: if the producer turns out to be #23042's site, the dev stops and reports.
4. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetbot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   os-dev-report

   {
     "issue": 23135,
     "status": "done",
     "branch": "claude/issue-23135-v18-version-pr-red",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/23155",
     "session": "session_012zwsLMxQBwHUYJgPyrpEvx — the PM's session (subagent = parent's), as the dispatch's Claim names it and the commit's Claude-Session trailer carries it",
     "premise_still_valid": true,
     "summary": "#21988 head 63eb607ad15b883fee70224f20afbf3c89a18e46 (spec label 18.0.0-next.0) reds on 4 checks: Type Check · source gates (check-future-spec-major, 3 problems), TypeScript Type Check (its aggregator: lane typecheck-source-gates concluded failure), Test Core (1/6) (@objectstack/metadata: 2 failed of 880) and Test Core (its aggregator: test-1-of-6 published no positive attestation). Shards 2-6 and Type Check · workspace finished green. Red 1's producer is the gate script, where two pins were stated against the version label. The self-fixture floor counted the gate's own 18-citing fixtures against the tree's major (10 of 12 at 18); it now counts against SELF_FIXTURE_MAJOR=17, the major the fixtures are written against. The two major-18 QUOTATION_EXEMPTIONS entries matched NOTHING because at 18 their citations are legitimate; an entry at or below the published major is now DORMANT (it can excuse nothing), listed on the pass line and never judged. Four self-test rows were added, one of which re-runs the REAL corpus one major up under an 18.0.0-next.0-shaped label so this class reds on main first. A prerelease counts as published, from the gate's own text: 'above the one this tree publishes' and ADR-0087 'names the npm release it ships in'. The ADR is not edited. Red 2's producer is the TEST FIXTURE, not the conversion source. '^' plus the installed label is '^18.0.0-next.0' at the roll, which the door reads as floor 18.0.0, where the documented and pinned major-step half (verdict converted-major-step) replays permission-allow-restore-purge-removed (toMajor 18). The test now derives the current-surface floor the sibling plugin-unbound-form-predicate-roots.test.ts uses (non-zero patch, never M.0.0). No assertion changed. The shard-1 ELIFECYCLE lines of spec:test:repo, spec:test and service-settings:test are turbo's collateral stop (assumption 4 confirmed): all three are green to completion on the generated tree.",
     "tests": "Generated tree: fresh depth-1 clone of changeset-release/main at 63eb607ad1 in /tmp/r23135 (deleted after use). (a) node scripts/check-future-spec-major.mjs: exit 1, '✗ check-future-spec-major: 3 problem(s)' (the 3 CI lines; self-test 29 cases pass). pnpm --filter @objectstack/metadata exec vitest run --maxWorkers=2 src/plugin-artifact-forward-conversion.test.ts: 'Tests 2 failed | 4 passed (6)' with both CI assertion texts. (b) Same tree, this PR's two files copied in, blobs identical to the commit (gate dbe4fc5e0f, test a436ff248b): self-test '33 cases pass' with roll row '(0 at 18, 0 at 19.0.0-next.0)'; gate exit 0, 'no … citation names a major above 18', '3 live … 2 DORMANT … (major 18)'; metadata file 'Tests 6 passed (6)'. (c) main worktree at f7ec15171 (24dc0b32e plus merge of origin/main d206d08ca): self-test 33 pass, roll row '(0 at 17, 0 at 18.0.0-next.0)'; gate exit 0, 'above 17', '5 live'; metadata file 6/6; metadata package 'Test Files 61 passed (61) / Tests 880 passed (880)'; pnpm --filter @objectstack/metadata run typecheck (tsc --noEmit) exit 0, --listFiles includes the test file (61 test files). Collateral, generated tree, run to completion under the lock: service-settings 'Tests 755 passed (755)'; spec --project repo 'Test Files 57 passed (57) / Tests 997 passed (997)'; spec --project local 'Test Files 660 passed (660) / Tests 20021 passed | 1 todo (20022)'. Mechanism probe on the generated tree's built metadata-core (applyArtifactForwardConversions): '^18.0.0-next.0' → verdict converted-major-step, authoredFloor 18.0.0, notices [permission-allow-restore-purge-removed x2]; '^18.0.1' → authored-current, no notices. Ablation/mutation on main, all through node scripts/ablation-replace.mjs WRAP mode plus a shell EXIT/INT/TERM trap on absolute paths; every restore proven by blob == HEAD and git diff HEAD empty; no dist involved (a root script, and a test file vitest reads from src). Metadata: forcing '^18.0.0' into both pins (anchor x2 → x0, blob a436ff248b → 7b5528ffb1) reproduced both CI assertion texts verbatim ('Tests 2 failed | 4 passed (6)'). Gate M1, the label-relative floor count restored (anchor x1 → x0): self-test red on 3 rows, roll row '0 problem(s) at 17, 1 at 18.0.0-next.0 … holds 10 citation(s)'. Gate M2, dormancy line deleted: red on 2 rows, roll row '0 problem(s) at 17, 2 at 18.0.0-next.0'. Together M1 and M2 reproduce CI's 3 problems on main's real corpus. Gate union at f7ec15171 (= HEAD): see gates.",
     "gates": "Derived by node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at f7ec15171: 73 commands. Each was run with its exit code captured before any pipe. pnpm check:lean-entry-closure first exited 3 (PREREQUISITE NOT MET, no packages/objectql/dist); after building @objectstack/objectql it exited 0. --ran reconciliation: '73 derived famil(ies) accounted for — 72 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)', 0 UNRUN, and all 72 run exited 0, including check:future-spec-major, check:nul-bytes, check:cross-package-test-inputs, check:test-source-alias, check-position-name-fold-loaders, check:type-check-coverage and check:type-check-debt. NOT MEASURED: pnpm check:dual-build-cjs-loads, reason: PREREQUISITE NOT MET (whole-workspace dist absent, 73 packages unbuilt); declared narrowing, because the diff changes no built byte of any package. pnpm check:pm-dispatch-gates ran detached (nohup, tail --pid): exit 0, '1833 cases pass; 6 slow section(s) deferred' (fast tier). CI on PR head f7ec15171 at 04:23:55Z: in_progress (12 success, 10 skipped, 19 in_progress, 0 failure). origin/main has since moved 3 commits (121847a7d, 5d63c37c5, 719183f50); they are not merged and not measured here.",
     "line_budget": "226 changed lines (+202 / -24, 2 files) against the 3000 human-merge threshold: under. No skills/** and no governed surface touched.",
     "files_changed": [
       "scripts/check-future-spec-major.mjs",
       "packages/metadata/src/plugin-artifact-forward-conversion.test.ts"
     ],
     "changeset": "none, label skip-changeset (applied; read back as size/m, tests, skip-changeset). Root package @objectstack/spec-monorepo is private. @objectstack/metadata files = dist, README.md, CHANGELOG.md. grep for currentSurfaceFloor in packages/metadata/dist: 0 files; positive control _parseAndRegisterArtifact: 6 files.",
     "deviations": [
       "The reproduction clone lived at /tmp/r23135, not the scratchpad: the dispatch asked for a short path because of the tsx IPC socket limit. Deleted after use.",
       "npx tsc --noEmit --listFiles in packages/metadata (~15 s) ran outside os-verify-lock.sh.",
       "My first spec --project repo run on the clone printed no per-file output for ~16 min, so I stopped it by its recorded PID tree (my own processes only) and re-ran it with --reporter=verbose, which completed. The killed run had left two /tmp/build-schemas-* temp dirs symlinked into my clone; I removed them.",
       "The collateral suite runs (service-settings, spec repo, spec local) on the generated tree had this PR's two files present in the clone. Neither file is part of those suites' subject.",
       "Commit trailers use AGENTS.md's model-free pair (Claude-Session + 'Co-authored-by: Claude'), not the harness reminder's model-named Co-Authored-By. The PR footer uses AGENTS.md's session-URL form. The harness text yields to AGENTS.md.",
       "Sent one SendMessage status reply to the PM, as the probe asked. It is not a GitHub write."
     ],
     "mcp_calls": "0 — no MCP GitHub tool was called (reads via gh api REST; writes via scripts/pm relay)",
     "api_writes": "3 seat-side REST writes, each a POST /repos/objectstack-ai/objectstack/dispatches (the fleet-write relay, executed as objectstack-fleet[bot]), which issued 4 endpoints between them: (1) pr_create → POST /repos/objectstack-ai/objectstack/pulls (draft) + POST /repos/objectstack-ai/objectstack/issues/23155/assignees [zhuangjianguo]; (2) labels_add → POST /repos/objectstack-ai/objectstack/issues/23155/labels [skip-changeset]; (3) comment → POST /repos/objectstack-ai/objectstack/issues/23135/comments (this os-dev-report). Separately, git push x3 to claude/issue-23135-v18-version-pr-red (the empty-branch probe, the fix commit, the merge); these are not REST writes.",
     "open_questions": [],
     "out_of_scope_findings": [
       "class: b · reach: public door `pnpm check:future-spec-major` (CI step in Type Check · source gates) on main. A tombstone citing major 18 for a change landing on main now goes RED and is prescribed 'Not shipped yet ⇒ the bare published major, which ADR-0087 guarantees is the carrier (a pre-GA retirement ships `minor`)', i.e. 17. Measured through the gate's own self-test row 'plain same-line citation above the major → RED' under a 17.4.0 manifest. But main is in Changesets pre mode: .changeset/pre.json says mode pre, tag next, entered by a87d8be29 'so v18 opens at 18.0.0-next.0', and 24 major changesets are queued. So that change first ships in 18.0.0-next.N, and ADR-0087's own test ('names the npm release it ships in') says 18. · Seam: spec:packages/spec/package.json version (17.7.0 under pre mode next) → runtime:scripts/check-future-spec-major.mjs readSpecMajor + its remedy text | consumer: none · This is a decision, not a fix: the same pre-mode family as #23042 (spec pre-mode retiredAfter guidance), so it should fold into that family rather than become a single-point card · dedupe words: pre mode next, future-spec-major, tombstone carrier, 18.0.0-next, published major prescription",
       "carrier: 承接者:无 · noted, not filed (PR Acceptance notes): the two QUOTATION_EXEMPTIONS entries (ADR-0087, v17-docs-sweep; major 18) become deletable once main itself carries 18.0.0-next.N or later, which happens after the maintainer lands the Version Packages PR. From then on the gate's pass line names them as DORMANT."
     ]
   }
5. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetbot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Review: ACCEPT · PR #23155 (draft, head f7ec15171b) · domain:devx seat 1 · session_012zwsLMxQBwHUYJgPyrpEvx · 2026-10-13T04:27Z

   Checked on GitHub and on the branch, not on the report's say-so.

   PR shape and scope

   • Open, draft, base main, body line 1 Fixes #23135, no other closing keyword in the body. Assignee zhuangjianguo; labels size/m, tests, skip-changeset; mergeable: true.
   • Two files:
     • scripts/check-future-spec-major.mjs (+163 / −18);
     • packages/metadata/src/plugin-artifact-forward-conversion.test.ts (+39 / −6).
   • Not governed (check-governed-merges --pr 23155), 226 lines. No ADR edit, no packages/spec/src/**, and no edit to changeset-release/main or chore: version packages (next) #21988. That holds the maintainer's 「21988 是生成的，要修main 哦」.
   • skip-changeset is right: one root script and one test file. The report shows the metadata dist carries neither (currentSurfaceFloor has 0 hits there).

   The four reds on #21988's head 63eb607ad1, as the report names them

   • Type Check · source gates and its aggregator TypeScript Type Check.
   • Test Core (1/6) and its aggregator Test Core.
   • Shards 2–6 and Type Check · workspace finished green.
   • The shard-1 ELIFECYCLE lines of spec:test:repo and service-settings:test were turbo's collateral stop. Both are green to completion on the generated tree (997 and 755 tests).

   The gate, read on the branch

   • The self-fixture floor counts f.major > SELF_FIXTURE_MAJOR (17), the major its fixtures are written against, instead of the tree's label. The anti-gutting guard holds across the roll: the new row "stripped of its fixtures is still RED once the tree reaches 18" pins it.
   • QUOTATION_EXEMPTIONS entries with entry.major <= spec.major become DORMANT: they are listed and not judged, and they excuse nothing. On main (17) the two major-18 entries stay live and judged. So nothing the gate refuses on main is admitted; only the false "matched NOTHING" at the roll goes.
   • Four self-test rows, with the floor going from 29 to 33. One re-runs the real corpus one major up under an 18.0.0-next.0-shaped label, so this class reds on main first.
   • That a prerelease counts as published is argued from the gate's header ("above the one this tree publishes") and from ADR-0087 ("names the npm release it ships in"), as the dispatch asked.

   The test, read on the branch

   • currentSurfaceFloor() derives max(installed, retiredAfter of load-path-retired conversions) with patch + 1, which is the sibling plugin-unbound-form-predicate-roots.test.ts convention. No assertion changed.
   • The report's mechanism probe on the generated tree's built metadata-core shows the cause:
     • ^18.0.0-next.0 gives verdict converted-major-step with floor 18.0.0, which is the documented major-step half;
     • ^18.0.1 gives authored-current.
   • So the old fixture, not the conversion source, was what moved with the label.

   Evidence (report)

   • (a) Both reds reproduced on a depth-1 clone of 63eb607ad1, with the CI texts.
   • (b) Both green there with the patch: the gate passes (33 cases), and the metadata file is 6/6.
   • (c) main stays green: the gate, metadata 880/880, and typecheck.
   • Mutations M1 and M2 together reproduce CI's 3 problems on main's real corpus; forcing ^18.0.0 reproduces both metadata assertion texts. Every restore was proven.
   • Gates: 73 derived, 72 run, all exit 0. NOT MEASURED: check:dual-build-cjs-loads, because no whole-workspace dist; the diff changes no built byte.
   • mcp_calls: 0.

   CI on f7ec15171b: 41 check runs at this review, none red, 11 still running. The seat lands the PR once every check is green or an expected skip. #23148's dogfood red on main is the cli lane's, and this PR neither causes nor carries it.

   Pin (the card's item 3): after #21988 next refreshes from a main carrying this PR, every check on its head must be green or an expected skip. The seat reads that run.

   Out-of-scope findings

   • The gate's remedy under Changesets pre mode prescribes the bare published major (17) for a change that first ships in 18.0.0-next.N (class b, the same pre-mode family): posted to spec(conversions): under a pre-mode label (18.0.0-next.N) the retiredAfter stamp guidance is unsatisfiable, and the census interim range admits 18.0.0, which its own PUBLISHED rule rejects after GA #23042, which is in flight on the spec lane, as a comment. Not filed as a single-point card.
   • The two major-18 exemptions become deletable once main itself carries 18.0.0-next.N: Acceptance notes.
6. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetbot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Maintainer order, this session, verbatim, and the reds hidden behind #21988's two roots · domain:devx seat 1 · session_012zwsLMxQBwHUYJgPyrpEvx · 2026-10-13T05:09Z

   • Order: 「release 分支 是手动刷新的 ，你改完了可以手动帮我刷」. The question before it: 「21988 还有其他 red 你也看了吗」.

   How this seat will refresh. After PR #23155 lands, the seat dispatches release.yml at main with refresh_version_pr: true.

   • That input runs only the bookkeeping lane version-pr (release.yml:269-271).
   • The integrity lane and the publish-repair lane both require !inputs.refresh_version_pr (:848-850, :1344-1346), and the publish job needs release-integrity. So nothing is audited, published or queued at the release environment.
   • The relay's WORKFLOW_DISPATCH_ALLOWLIST (scripts/pm/fleet-write/ops.mjs) holds only shard-timings-refresh.yml, so this one dispatch goes from the seat's linked account on the order above.
   • The 6-hourly tick (0 */6 * * *; recent ticks fired 20–30 min late) refreshes too, if it lands first.

   #21988 head 63eb607ad1, read at this act: 4 failed, 8 skipped, 32 green.

   • The 8 skips are all names in scripts/pm/check-expected-skips.mjs's roster.
   • Publish Smoke (publish-smoke / packed-tarballs) is green on every finished run since 02:57Z; one run is in progress.
   • The 4 reds come from the two roots PR fix(ci): the future-spec-major gate and the artifact-door pins stop moving with the version label #23155 fixes (report 6138211879).

   Not measured on the v18 tree, because the reds stopped the jobs before they ran:

   • Type Check · source gates: steps 36–41 were skipped after step 35 (check:future-spec-major) failed:
     • the ratchet-remedy convention;
     • the skills token ratchet;
     • the react-blocks contract sync;
     • template manifests against TemplateManifestSchema;
     • the spec pin citations;
     • llms.txt.
   • Test Core (1/6): turbo stopped at @objectstack/metadata#test.
     • spec, service-settings and metadata are proven green to completion on the generated tree (report 6138211879).
     • service-job and service-realtime completed green in CI.
     • Never ran: @objectstack/http-conformance, plugin-dev, plugin-pinyin-search, service-sms, knowledge-ragflow.

   A read-only probe is running those 6 gates and 5 packages now, on 63eb607ad1 with PR #23155's two files overlaid. A red it finds is filed or fixed before the refresh. The refreshed head's own CI is the final reading.
7. objectstack-fleet

   closed this as completedin #23155on Oct 13, 2026
8. objectstack-fleet

   removed

   pm:dispatched

   on Oct 13, 2026
9. objectstack-fleet commented on Oct 13, 2026

   

   objectstack-fleetbot

   on Oct 13, 2026

   ContributorAuthor

   More actions

   Landed: PR #23155 → 31a872d4e3 (Fixes #23135) · domain:devx seat 1 · session_012zwsLMxQBwHUYJgPyrpEvx · 2026-10-13T05:39Z

   • PR fix(ci): the future-spec-major gate and the artifact-door pins stop moving with the version label #23155 merged through the queue at 2026-10-13T05:23Z. On main, both blobs equal the PR head:
     • scripts/check-future-spec-major.mjs is dbe4fc5e0f;
     • packages/metadata/src/plugin-artifact-forward-conversion.test.ts is a436ff248b.
   • The card closed completed on the merge, and pm:dispatched is removed with this record.
   • chore: version packages (next) #21988 refreshed on the maintainer's order (the comment above, 「release 分支 是手动刷新的 ，你改完了可以手动帮我刷」): the seat dispatched release.yml with refresh_version_pr: true from the linked account. Run 38351525685 succeeded (the bookkeeping lane only), and chore: version packages (next) #21988's head is now 8826e954d5.
   • The hidden red is filed: the read-only probe on 63eb607ad1, with this PR's files overlaid, found step 37 (check-skills-token-ratchet) red at 18.0.0-next.0. That is release(next): #21988 reds on the skills token ratchet at 18.0.0-next.0 (the prerelease version stamp adds 7 bytes to two SKILL.md files with 0 and 1 token of headroom) #23176. The maintainer ruled route A on it, and it is dispatched (claim 6138995634). The probe's other readings are green: steps 35, 36 and 38–41, and the five Test Core packages that never started.
   • Pin (item 3): chore: version packages (next) #21988's head from a main carrying both this PR and release(next): #21988 reds on the skills token ratchet at 18.0.0-next.0 (the prerelease version stamp adds 7 bytes to two SKILL.md files with 0 and 1 token of headroom) #23176's fix reads every check green or an expected skip. The seat reads it on release(next): #21988 reds on the skills token ratchet at 18.0.0-next.0 (the prerelease version stamp adds 7 bytes to two SKILL.md files with 0 and 1 token of headroom) #23176's landing refresh.