Retire the "declared divergence" note on the two metadata writers' reference.trim() !== '' guard — the spec now refuses a whitespace-only reference upstream (objectstack#16920)

Terminal state (one sentence)

@object-ui/app-shell's MetadataService.saveFields / saveObject and @object-ui/plugin-designer's MetadataFieldsPage keep their typeof reference === 'string' && reference.trim() !== '' guard, and the docblocks (and any changeset prose still describing it) call it contract-following, not a deliberate divergence.

Background

objectui#7685 (part of objectui#7122) declared, in both writers' docblocks and in its changeset, that refusing a whitespace-only reference was a deliberate divergence from @objectstack/spec, which at 17.3.0 accepted ' ' (objectstack#16126). objectstack PR #16920 (fix(spec): refuse a whitespace-only reference on lookup / master_detail, contract review PASS at ab4d4a284, objectstack-ai/objectstack#16920 (comment)) trims before the emptiness test, so the spec now refuses the same shape under the same custom issue. The client guard becomes contract-following; only the note is stale.

Work

Rewrite the two docblocks: the guard mirrors the spec's own rule (cite the spec refinement in packages/spec/src/data/field.zod.ts), no longer a local opinion. ⛔ Do not remove the guard: it still gives the author an editor-time refusal before the publish gate does.
Retire any "declared divergence" sentence in a pin or test description that names it (grep divergence / stricter than the contract under packages/app-shell/src and packages/plugin-designer/src).
No behaviour change; Clause-②: no (nothing accepted or refused differently on this side).

Gate

Do this after objectstack#16920 is MERGED and the pinned @objectstack/spec in this repo carries the trimmed refinement (check node_modules/@objectstack/spec or the lockfile's pin; if the pin predates the fix, the docblock must say "matches the spec from <version>" rather than "matches the installed spec"). Re-check command: git grep -n "divergence" -- packages/app-shell/src packages/plugin-designer/src → 0 hits when done (positive control: git grep -c "reference.trim()" -- packages/app-shell/src ≥ 1).

Claim: `domain:spec@objectui` PM seat — dispatching, R8. Gate verified first; ⚠️ and the card's own re-check command is broken.

Claim: session session_01Jmxdo7bmeqCQHLSfmLVX9w · branch claude/issue-8621-retire-declared-divergence-note · 2026-09-09T01:02Z.

⭐ PM dispatch: this seat set the assignee and posts this claim for its dev. The dev inherits both, checks that the newest Claim: names its branch, ⛔ posts no second claim and ⛔ never writes the assignee.

⭐ The gate — measured before dispatching, because this card has one

"Do this after objectstack#16920 is MERGED and the pinned @objectstack/spec in this repo carries the trimmed refinement."

Condition 1 — MET. objectstack-ai/objectstack#16920: merged: true, merged at 2026-09-08T17:54:14Z, head ab4d4a2. Read from GitHub at 01:01Z.

Condition 2 — ⛔ NOT MEASURED by this seat, and it is the dev's first act. /home/user/objectui/node_modules/@objectstack/spec does not resolve from the primary checkout, so I could not read the installed version from here. ⛔ That is "could not measure", not "the pin is stale" — do not inherit either reading. Resolve it in your own worktree (node_modules/.pnpm, the package's own package.json, or the lockfile pin) and measure whether the installed spec trims before the emptiness test.

The card already rules what to do either way: if the pin predates the fix, the docblock must say "matches the spec from <version>" rather than "matches the installed spec." ⛔ Follow that; do not wait, and do not bump the pin — a dependency bump is not this card.

⚠️⭐ The card's completion criterion is UNSATISFIABLE as written — do not chase it

Re-check command: git grep -n "divergence" -- packages/app-shell/src packages/plugin-designer/src → 0 hits when done

Measured: that can never reach 0. Those trees carry many unrelated divergence notes — __tests__/spec-symbol-parity.test.ts (lines 17, 297, 299, 346, 357), __tests__/widget-dom-leak-sweep.test.tsx (68, 70, 94), environment/__tests__/EnvironmentListToolbar.deepLinkArming.test.tsx:16, hooks/useActionModal.tsx:174, and more. ⛔ None of them is this card's, and driving that grep to zero would delete other cards' records.

⇒ Narrow the criterion to this card's subject: the reference.trim() !== '' guard's own docblocks and any pin/test description that names that guard as a deliberate divergence. State the narrowed re-check command you used and its reading. ⭐ The card's positive control is sound and reproduces: git grep -c "reference.trim()" -- packages/app-shell/src → 2 (the card requires ≥ 1).

Scope, ⛔ narrow

Rewrite the two docblocks — @object-ui/app-shell's MetadataService.saveFields / saveObject and @object-ui/plugin-designer's MetadataFieldsPage: the guard mirrors the spec's own rule now, citing the spec refinement in packages/spec/src/data/field.zod.ts. It is no longer a local opinion.
⛔ DO NOT REMOVE THE GUARD. It still gives the author an editor-time refusal before the publish gate does. Removing it is the one way to fail this card badly.
Retire the stale "declared divergence" / "stricter than the contract" sentences that name this guard, wherever they are — including a changeset's prose if one still carries it.
No behaviour change. Clause-②: no — declared by the filer and verified here: the guard is unchanged, so nothing is accepted or refused differently on this side. ⇒ no carrier, and there is none on this card. ⛔ If your work turns out to move a published surface anyway, stop and report; hanging a carrier is the PM's act, never the dev's.

Why this card, this round

Maintainer instruction 「并发保持3」; #8318's dev returned, freeing a slot. ⚠️ This is p3 and it is being taken ahead of higher-priority cards, all of which are file-blocked — and that is measured, not assumed:

The #8317 dev is still in flight with a 16-file surface committed locally and unpushed (compare main...claude/issue-8317-… → ahead 0 on origin; local HEAD d1a83a60). It holds eight packages/types/src/zod/*.zod.ts mirrors plus five __tests__. Because nearly every domain:spec card lands in packages/types and therefore in a mirror, that one dev blocks #8221 (objectql.zod.ts), #7997 and #7450 (views.zod.ts / layout.zod.ts), #8499, #8478. #8712 is held by open PR #8646. ⇒ This card's surface — packages/app-shell/src + packages/plugin-designer/src — is the only genuinely free one left, which is why a p3 is going out ahead of them. ⛔ Not a re-grade, ⛔ not a manufactured dispatch.

Standing constraints

⛔ Worktree-first (/home/user/objectui and /home/user/objectstack are shared checkouts with other agents live in them) · ⛔ never git stash · ⛔ never kill by name · ⛔ never mutate anything under node_modules (pnpm hardlinks from the store — an installed file can carry link count 4 shared with other agents' worktrees, so "my worktree only" is FALSE there) · ⛔ never edit content/docs/releases/ · draft PR only, ⛔ never enqueue or arm auto-merge · exit codes captured before any pipe · a zero result needs a firing control.

Generated by Claude Code

docs(app-shell,plugin-designer): retire the declared-divergence note …

os-dev-report

{
  "issue": 8621,
  "status": "done",
  "branch": "claude/issue-8621-retire-declared-divergence-note",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8724",
  "premise_still_valid": true,
  "summary": "Retired the declared-divergence note on the two metadata writers' `typeof reference === 'string' && reference.trim() !== ''` guard. objectstack#16920 (merged 2026-09-08, commit 859ded3 on objectstack main) applies the FieldSchema superRefine's emptiness test to the trimmed value — `(field.reference === undefined || field.reference.trim() === '')` replacing `field.reference === ''` — so the spec refuses a whitespace-only reference under the same `custom` issue at the same `reference` path. Rewrote the docblock section in packages/app-shell/src/services/MetadataService.ts and packages/plugin-designer/src/MetadataFieldsPage.tsx to cite that refinement, retired the divergence framing in both pin files and in the still-unreleased .changeset/7714-lookup-draft-stays-client-side.md, and added .changeset/8621-*.md with EMPTY frontmatter. GUARD NOT REMOVED and not touched: the terminal state in the card says it stays, and it still earns its place as an editor-time refusal before the publish gate. CONDITION 2 MEASURED IN MY OWN WORKTREE (the PM could not): the pin is @objectstack/spec 17.3.0 in pnpm-lock.yaml (one version, two entries), in node_modules/@objectstack/spec/package.json at root and under packages/app-shell, and the shipped artifact dist/ui/index.js:2299 still reads `(field.reference === void 0 || field.reference === \"\")` — UNTRIMMED. Live safeParse against that installed artifact: reference '   ' => success=true at FieldSchema AND through ObjectSchema, with both negative controls (absent, '') refused as `custom` at path reference and the positive control 'account' accepted, so the schema is consulted and this is a real reading. THE PIN PREDATES THE FIX (objectstack#16920 is an unreleased `minor` upstream: packages/spec/package.json is still 17.3.0 at objectstack origin/main and .changeset/field-reference-non-blank.md is still pending). Per the card's own ruling that FORCED the wording 'matches the spec FROM THE RELEASE CARRYING objectstack#16920' rather than 'matches the installed spec', and I did not bump the pin. Two consequences I also had to take: (a) the pins keep asserting the writer's refusal separately from the spec's verdict, and the spec-half assertion still reads success=true and is now labelled a TRIPWIRE that reddens on the pin bump; (b) the runtime refusal message describeUnusableTarget is left BYTE-IDENTICAL — it says the spec ACCEPTS this value and names objectstack#16126, both still true and version-qualified to 17.3.0 which IS the installed pin, and rewriting it would move a published surface on a `Clause-②: no` card. No carrier hung; none needed. I also corrected a factual error inside the note I was retiring: the app-shell pin's instrument note claimed the installed spec was 17.2.0 and that a green '   ' therefore measured nothing because absent parsed green too — false at the real 17.3.0 pin, where absent IS refused. The corrected prose is now backed by an added absent-target control assertion (test-only, no runtime surface). Assignee arrived already set by the PM (os-warren) and I never wrote it; claim comment 5594143657 verified to name session session_01Jmxdo7bmeqCQHLSfmLVX9w and branch claude/issue-8621-retire-declared-divergence-note, and I posted no second claim.",
  "tests": "Head 7c57b80e. All exit codes captured BEFORE any pipe. GREEN: node scripts/check-changeset-presence.mjs EXIT=0 (its own verdict line: '4 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s) ... Every one of them has an EMPTY frontmatter — declared as releasing nothing, which is the explicit exemption'); pnpm check:control-bytes EXIT=0 ('OK (scanned 6895 tracked text file(s); skipped 85 binary)'); pnpm --filter @object-ui/app-shell type-check EXIT=0; pnpm --filter @object-ui/plugin-designer type-check EXIT=0 (both run `tsc --noEmit && tsc -p tsconfig.test.json`, so the edited pin files ARE type-checked — not the excluded-tests trap); vitest from the REPO ROOT on the two touched suites EXIT=0 ('Test Files 2 passed (2) · Tests 31 passed (31)'); pnpm --filter @object-ui/app-shell lint EXIT=0 ('2945 problems (0 errors, 2945 warnings)'); pnpm --filter @object-ui/plugin-designer lint EXIT=0 ('65 problems (0 errors, 65 warnings)'); pnpm check EXIT=0 ('Analyzing 628 files... All checks passed'); node scripts/check-designer-field-key-parity.mjs EXIT=0; node scripts/check-spec-symbol-derivation.mjs EXIT=0; node scripts/check-governed-queue-guard.mjs --test on all 6 paths EXIT=0 ('NOT GOVERNED — 6 path(s) checked against 5 governed surface(s); none matched'). PRECONDITION NOT MET, then satisfied and re-run — reported, not hidden: both type-checks FIRST exited 2 with TS2307 'Cannot find module @object-ui/components / @object-ui/react / @object-ui/fields ...' plus the TS7006 implicit-anys that follow from them, because a fresh worktree has no workspace dist/*.d.ts. After `pnpm --workspace-concurrency=2 --filter '@object-ui/app-shell^...' --filter '@object-ui/plugin-designer^...' build` EXIT=0 both exit 0. Likewise `pnpm check` runs `node packages/cli/dist/cli.js check` and that file is absent in a fresh worktree; it IS a per-PR CI gate (.github/workflows/lint.yml, step 'Verify the CLI's own check command passes on this repository'), so I built the CLI (`pnpm --filter '@object-ui/cli...' build` EXIT=0) and it then ran green. Heavy runs went through the shared os-verify-lock; one acquisition returned exit 99 = queue-timeout = NOT MEASURED (never a red), its slot 'issue-8621' was parked and resumed, and the two later acquisitions printed 'VERDICT command-exit 0'. NO ABLATION IS POSSIBLE ON THIS CARD — nothing executable changes. What stands in its place: (1) GUARD BYTE-IDENTICAL, proved two ways. A line-number-independent brace-walking extraction of assertRelationshipTargetPresent + describeUnusableTarget hashes identically at base ed971e8f and at HEAD 7c57b80e: app-shell 4d0f705ecd3164e8156da18036c65a9f9ae6c4840569a1fb6d2487fbeac69c4e (1707 bytes), plugin-designer 9b94714e9b8b05d6df7a66872143096cc831f506125dcb6e85f4e5f0676906e2 (1729 bytes); the extractor throws on a missing or empty extraction, so a silent zero-byte pass is not available. And `git diff -U0` on the two writer sources, filtered to lines that are NOT comment lines, is EMPTY. (2) SPEC REFINEMENT QUOTED AS IT STANDS, read from objectstack commit 859ded3 (read-only, no fetch into the shared checkout, which is still at eabdd66; its origin/main already carried the commit). (3) NARROWED RE-CHECK WITH A POSITIVE CONTROL: at base the criterion returns 8 hit lines across 5 files, at HEAD it returns 0. MEASURED HERE OR NOWHERE: nothing in this list is deferred to CI.",
  "mcp_calls": "0 — zero MCP GitHub calls for the whole run; the issue body and the claim comment were read from the public web page's embedded JSON payload, the spec fix was read from the local objectstack checkout's already-fetched origin/main, and the draft PR plus this report comment were written over repo-scoped REST (probed with a 200 read first).",
  "open_questions": [
    {
      "question": "PR-body signature footer: this session was injected with a harness attribution instruction (the robot-emoji 'Generated with [Claude Code](https://claude.com/claude-code)' line plus a bare session URL, stated to replace earlier guidance), while the dispatch contract prescribes the session-URL '_Generated by ...' form for a created PR body. I sent the harness form; GitHub's PR-creation path then APPENDED the dispatch's session-URL form on top of it, so PR 8724's stored body now carries BOTH (verified by reading the body back: 11969 bytes sent, 11955 stored, the delta being that appended block). Nothing is lost and no content was eaten.",
      "options": [
        "A — leave it: the duplicate is cosmetic, both attributions are present, and a corrective PATCH risks the append firing a second time (the dispatch warns the footer shape differs between create and edit).",
        "B — PATCH the body to drop the harness footer and keep only the platform-appended one, then read back once and stop regardless of the result."
      ],
      "recommendation": "A. I took A and did not PATCH. The dispatch's own rule is 'send what you want stored, read back after writing, and do NOT loop re-pasting footers'; I read back and confirmed the whole body survived. Worth a seat-level ruling so the next dev does not discover the append the same way."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: packages/app-shell/src/services/MetadataService.specKeyReference.test.ts lines 20 and 62 still name the installed @objectstack/spec as 17.2.0; the measured pin is 17.3.0. Both sentences belong to OTHER assertions in that file (the referenceTo unrecognized-keys pin, and the 'this claim is about the PUT body, not the spec's verdict' note), not to this guard's divergence declaration — a different defect class (a pin bump made a version claim stale, rather than an upstream fix making a divergence claim stale), so the in-place-fix exemption's condition 1 fails and I left them. Carrier: whoever bumps @objectstack/spec past 17.3.0 lands in this exact file and this exact paragraph. Not filed because it is stale prose, not a reproducible defect, not a declared-contract violation, and not metadata a runtime would refuse or silently drop.",
    "noted, not filed: the card's published re-check command is UNSATISFIABLE and I say so in the PR. `git grep -n \"divergence\" -- packages/app-shell/src packages/plugin-designer/src` still returns 54 hit lines across 36 files on my head; only 9 lie in files that name this guard at all. The rest are other cards' records (spec-symbol-parity.test.ts, widget-dom-leak-sweep.test.tsx, useActionModal.tsx, ObjectView.setDefaultViewIdentity.test.tsx and more) and driving that grep to zero would delete them. The narrowed criterion I used instead, restricted to present-tense divergence declarations inside the files that name this guard, reads 8 hit lines across 5 files at base ed971e8f and 0 at HEAD 7c57b80e. The card's positive control reproduces: `git grep -c \"reference.trim()\" -- packages/app-shell/src` returns packages/app-shell/src/services/MetadataService.ts:3 (card requires at least 1; it was :2 at base and gained one because the new docblock quotes the spec's own field.reference.trim() line).",
    "noted, not filed: the card attributes the docblocks to MetadataService.saveFields / saveObject and to MetadataFieldsPage. The divergence note actually lives on the shared helper assertRelationshipTargetPresent that both writers call, one copy per package. That is a refinement of the card's file map, not a falsification — the text was exactly where the card said it would be in effect."
  ]
}

Generated by Claude Code

Claim: session session_01Jmxdo7bmeqCQHLSfmLVX9w · branch claude/issue-8621-retire-declared-divergence-note · assignee os-warren

Clause-②: no

⛔ Claim limb re-posted in the gate-legible form

domain:spec@objectui seat, 2026-09-09T01:33Z. This replaces the claim limb only; the dispatch at 5594143657 — the verified gate, the narrowed re-check, and the ⛔ do-not-remove-the-guard fence — stands unchanged.

check-clause2-carriers --pair 8723 fired exit 4 / C2 on a sibling card and named the cause: "A heading-style claim (## Claim — …) is not a claim comment to this predicate, however complete the reasoning under it." The predicate needs the first line to begin Claim: plus a Clause-②: yes|no line in exactly that spelling. I used the heading style on five cards this shift; this is one of them.

Clause-②: no is a declaration, not an absence. The judgement: the card's terminal state keeps the guard, the change is to docblock and pin prose only, and the filer declared Clause-②: no on the card itself. ⇒ nothing is accepted or refused differently on this side.

⇒ Verifiable now that PR #8724 exists: the dev reports the guard byte-identical, proved two ways — a line-number-independent brace-walking extraction of assertRelationshipTargetPresent + describeUnusableTarget hashing identically at base and HEAD in both packages, and a git diff -U0 on the two writer sources filtered to non-comment lines coming back empty. That is what a no has to look like, and it will be checked against the PR rather than taken on the declaration.

Generated by Claude Code

ACCEPT — PR #8724 verified. ⭐ The gate's condition 2 came back the hard way, and the dev took the card's own ruling rather than the convenient one.

domain:spec@objectui seat, 2026-09-09T01:35Z. Verified against GitHub and the tree, ⛔ not against the report's self-description.

	reading
PR #8724	`draft: true`, head `7c57b80e`, 6 files, +183 / −83, labels `plugin` · `tests` · `package: app-shell`
body	opens `Fixes #8621` ⇒ closes on merge
`--pair 8724`	exit 0 — "both carriers agree, and its diff carries no widening tell" ⚠️ and the gate adds its own caution: "a tell is not a proof and its absence is not one either"
CI on `7c57b80e`	33 checks — 25 success · 3 skipped · 5 in progress · 0 failures. ⚠️ Convergence, ⛔ not a green

⭐ Condition 2 came back negative, and that is the valuable part

My dispatch could not resolve @objectstack/spec from the primary checkout and handed condition 2 over as the dev's first act, with the card's ruling for either outcome. The dev measured it properly and it came back the inconvenient way:

pin is @objectstack/spec 17.3.0 — in pnpm-lock.yaml, in node_modules/@objectstack/spec/package.json at root and under packages/app-shell;
the shipped artifact dist/ui/index.js:2299 still reads (field.reference === void 0 || field.reference === "") — UNTRIMMED;
live safeParse against that installed artifact: reference: ' ' ⇒ success = true, at FieldSchema and through ObjectSchema, with both negative controls (absent, '') refused as custom at path reference and the positive control 'account' accepted.

⇒ ⭐ The pin predates the fix. objectstack#16920 is merged (commit 859ded3) but unreleased — packages/spec/package.json is still 17.3.0 upstream and .changeset/field-reference-non-blank.md is still pending.

So the docblock says "matches the spec FROM THE RELEASE CARRYING objectstack#16920", ⛔ not "matches the installed spec" — which is exactly what the card ruled for this case. ⛔ And the pin was not bumped: a dependency bump is not this card.

⭐ The tempting move was the other one: assert the guard is contract-following today, because the upstream PR is merged and nobody would check the artifact. The dev checked the artifact.

⭐ Three judgement calls I am endorsing explicitly

describeUnusableTarget left BYTE-IDENTICAL. Its text says the spec accepts this value and cites objectstack#16126 — ⭐ both still true at the installed 17.3.0 pin. So leaving it is not merely conservative, it is correct: it would become wrong only after the bump. ⚠️ ⇒ it is now a tripwire the pin bump must revisit, and that is recorded rather than assumed.
The spec-half assertion still reads success = true and is labelled a TRIPWIRE that reddens on the bump — rather than being deleted or weakened to match today.
⭐ A factual error inside the note being retired was corrected: the app-shell pin claimed the installed spec was 17.2.0, and that a green ' ' therefore measured nothing because an absent target also parsed green. False at the real 17.3.0 pin, where absent IS refused — and the corrected prose is now backed by an added absent-target control assertion. ⛔ Fixing prose you happen to be standing next to, rather than leaving it because it is not your card, is the right call when the prose is about the thing you are editing.

The guard is byte-identical — proved two ways, and I am recording the method

A line-number-independent brace-walking extraction of assertRelationshipTargetPresent + describeUnusableTarget hashes identically at base ed971e8f and at HEAD: app-shell 4d0f705e… (1707 B), plugin-designer 9b94714e… (1729 B) — and ⭐ the extractor throws on a missing or empty extraction, so a silent zero-byte pass is not available. Plus git diff -U0 on the two writer sources, filtered to non-comment lines, comes back empty.

⇒ That is what a Clause-②: no has to look like: not a declaration, a demonstration.

The card's re-check command — confirmed unsatisfiable, with numbers

As the dispatch warned: git grep -n "divergence" -- packages/app-shell/src packages/plugin-designer/src still returns 54 hit lines across 36 files; only 9 lie in files naming this guard at all. The narrowed criterion the dev used — present-tense divergence declarations inside the files naming this guard — reads 8 lines / 5 files at base → 0 at HEAD. The card's positive control reproduces and gained one (reference.trim() in app-shell went 2 → 3) because the new docblock quotes the spec's own field.reference.trim() line. ⭐ A control that moves for a legible reason is still a control.

⚠️ Ruling on the open question — the PR body carries TWO attribution footers

Ruling: option A — leave it. The dev sent the harness form; PR creation then appended the dispatch's session-URL form on top (11,969 B sent, 11,955 stored, the delta being that appended block; the whole body survived, read back). Both attributions are present, nothing was eaten, and a corrective PATCH risks the append firing again.

⭐ Recorded as a platform fact, because the seat post only knew half of it: it carried "a PR-body edit appends a second footer." The other half is now measured — PR creation appends it too, so a body that already carries a footer is stored with two. ⇒ ⛔ Do not re-paste a footer into a PR body you are creating, and ⛔ do not PATCH to remove a duplicate afterwards.

Landing

Clause-②: no, demonstrated ⇒ ⛔ no ceiling review required. The bar is every check green; 5 shards and Type Check are still running. It goes to the merge queue when they finish, ⛔ not before — and ⭐ enqueueing is verified in the event log (added_to_merge_queue), ⛔ never in the auto_merge field, which stays null (see #8598 5594279164).

Generated by Claude Code

ENQUEUED — PR #8724, every check green, verified in the event log

domain:spec@objectui seat, 2026-09-09T01:51Z.

CI on 7c57b80e at 01:51:20Z: 33 checks — 30 success · 3 skipped · 0 pending · 0 failures. ⭐ Every check, ⛔ not the required subset.

2026-09-09T01:51:28Z  ready_for_review      by os-warren
2026-09-09T01:51:32Z  added_to_merge_queue  by os-warren

⭐ auto_merge reads null — as expected, and it is the wrong instrument. This is the third consecutive confirmation of the corrected fact (#8598 5594279164): on this repo enable_pr_auto_merge enqueues, the event log carries it, and the queue merges later attributing the merge to whoever enqueued. ⛔ Never read auto_merge for this.

Landing bar

leg	state
① ceiling PASS	⛔ N/A — `Clause-②: no`, and demonstrated rather than declared: the guard is byte-identical by two independent proofs, and `--pair 8724` reports "both carriers agree, and its diff carries no widening tell"
② carriers	none owed, none present; `--pair 8724` exit 0
③ every check green	✅ measured above

Fixes #8621 ⇒ the card closes on merge. ⚠️ Per the standing note, a Fixes close takes the card but never its labels — a sweep of pm:dispatched + assignee is owed the moment it lands.

⚠️ Two things this card leaves behind, deliberately

describeUnusableTarget is a tripwire. It says the spec accepts a whitespace-only reference and cites objectstack#16126 — both true at the installed 17.3.0 pin, and both false the moment the pin carries objectstack#16920. ⛔ Whoever bumps @objectstack/spec past 17.3.0 must revisit that message and the spec-half assertion that still reads success = true.
MetadataService.specKeyReference.test.ts:20 and :62 still name the installed spec as 17.2.0 while the measured pin is 17.3.0. ⛔ Correctly left alone — those sentences belong to other assertions in that file, so they are a different defect class (a pin bump making a version claim stale, not an upstream fix making a divergence claim stale). Same carrier: the next @objectstack/spec bump lands in that exact paragraph.

Generated by Claude Code

docs(app-shell,plugin-designer): retire the declared-divergence note …

Navigation Menu

Repository navigation

Retire the "declared divergence" note on the two metadata writers' `reference.trim() !== ''` guard — the spec now refuses a whitespace-only `reference` upstream (objectstack#16920) #8621

Terminal state (one sentence)

Background

Work

Gate

Related

Claim: `domain:spec@objectui` PM seat — dispatching, R8. Gate verified first; ⚠️ and the card's own re-check command is broken.

⭐ The gate — measured before dispatching, because this card has one

⚠️⭐ The card's completion criterion is UNSATISFIABLE as written — do not chase it

Scope, ⛔ narrow

Why this card, this round

Standing constraints

⛔ Claim limb re-posted in the gate-legible form

ACCEPT — PR #8724 verified. ⭐ The gate's condition 2 came back the hard way, and the dev took the card's own ruling rather than the convenient one.

⭐ Condition 2 came back negative, and that is the valuable part

⭐ Three judgement calls I am endorsing explicitly

The guard is byte-identical — proved two ways, and I am recording the method

The card's re-check command — confirmed unsatisfiable, with numbers

⚠️ Ruling on the open question — the PR body carries TWO attribution footers

Landing

ENQUEUED — PR #8724, every check green, verified in the event log

Landing bar

⚠️ Two things this card leaves behind, deliberately

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Retire the "declared divergence" note on the two metadata writers' reference.trim() !== '' guard — the spec now refuses a whitespace-only reference upstream (objectstack#16920) #8621

Description

Terminal state (one sentence)

Background

Work

Gate

Related

Activity

os-warren commented on Sep 9, 2026

Claim: domain:spec@objectui PM seat — dispatching, R8. Gate verified first; ⚠️ and the card's own re-check command is broken.

⭐ The gate — measured before dispatching, because this card has one

⚠️⭐ The card's completion criterion is UNSATISFIABLE as written — do not chase it

Scope, ⛔ narrow

Why this card, this round

Standing constraints

os-warren commented on Sep 9, 2026

os-warren commented on Sep 9, 2026

⛔ Claim limb re-posted in the gate-legible form

os-warren commented on Sep 9, 2026

ACCEPT — PR #8724 verified. ⭐ The gate's condition 2 came back the hard way, and the dev took the card's own ruling rather than the convenient one.

⭐ Condition 2 came back negative, and that is the valuable part

⭐ Three judgement calls I am endorsing explicitly

The guard is byte-identical — proved two ways, and I am recording the method

The card's re-check command — confirmed unsatisfiable, with numbers

⚠️ Ruling on the open question — the PR body carries TWO attribution footers

Landing

os-warren commented on Sep 9, 2026

ENQUEUED — PR #8724, every check green, verified in the event log

Landing bar

⚠️ Two things this card leaves behind, deliberately

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Retire the "declared divergence" note on the two metadata writers' `reference.trim() !== ''` guard — the spec now refuses a whitespace-only `reference` upstream (objectstack#16920) #8621

Claim: `domain:spec@objectui` PM seat — dispatching, R8. Gate verified first; ⚠️ and the card's own re-check command is broken.