feat(spec)!: ListView.userActions.editInline defaults to true on the v18 line; editInline: false is the opt-out (#22605, spec half) - #22629

📓 Docs Drift Check

2 anchor(s) derived from 1 changed package(s); no hand-written page names any of them. ⚠️ 1 changed file(s) yielded no anchor (packages/spec/authorable-defaults/ui.json), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 7ea2f37356cdc3787e2053c4786707be157a22a5 && git checkout 7ea2f37356cdc3787e2053c4786707be157a22a5
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin f368b7e980d3df201620098be965523b50d6df19 fc3733b9714493610f60cc80e7849f36471a6548 && git checkout -B drift-repro f368b7e980d3df201620098be965523b50d6df19 && git merge --no-ff fc3733b9714493610f60cc80e7849f36471a6548

node scripts/docs-audit/affected-docs.mjs --json f368b7e980d3df201620098be965523b50d6df19

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 24d9bdcbf008e79365002ffb56616f8d1e4b3b74
Local-runs: none

Read-only: the card body and its four comments (triage 6093968052, claim 6094254055, dev report 6094863892, ACCEPT 6094884470), the PR body and file list, the net diff against the merge-base e22315238f (9 files, +233 / −7, identical to the three-dot diff against origin/main), and the head's check-runs. The branch is at the brief's sha. Nothing built, run or re-run.

① Derived judgments

• The default and its describe. UserActionsConfigSchema.editInline is z.boolean().default(true); the describe says on by default, a user who may update the object edits a cell in place with the field's type-aware widget under the permission gate, and editInline: false makes the list read-only in place — the card's three sentences. The schema docblock records the ruling verbatim. Held.
• DEFAULT_CHANGES_BY_MAJOR[17]. The row is ui/UserActionsConfig:editInline, from: 'false', to: 'true', inside the 17: [ block (lines 218–621 at head) beside the ui/PaginationConfig:pageSize row. 17 is the only key that can authorise anything: CURRENT_MAJOR is parseInt(SPEC_VERSION) with @objectstack/spec at 17.7.0, and the ratchet's rule 4 is that only the current major authorises. authorable-defaults/ui.json moves exactly one line (= false → = true), matching both fingerprints. Held.
• The D3 entry list-view-edit-inline-default-on. One file, entries/semantic/18.list-view-edit-inline-default-on.ts, protocol 18 — the same step as view-pagination-page-size-default-50. surface names the OMITTED key on both carriers of the one schema (a list view's userActions, a page's interfaceConfig.userActions) and carries no backtick, as the build-upgrade-guide note demands; replacement is "nothing" plus the opt-out; reason quotes the ruling and says why no D2 exists (a pass writing false everywhere would preserve the old posture and defeat the ruling); acceptanceCriteria matches the card's acceptance. No tombstone, no RETIRED_KEYS_BY_MAJOR row, no D2 — right, nothing is removed. The registry.ts hunk is the entry byte-for-byte inside step18. Held. One observation, not a finding: the reason string carries objectui#5144; 12 of the 414 semantic entries on origin/main already carry an objectui#N in prose, so it is precedented, and check:doc-authoring rule 2 scans a different surface.
• Regenerated references. Four rows — view.mdx ×3, page.mdx ×1 — now read optional (default: \true`)with the new describe. At head,git grepfor the old describe text ("Off by default: the list") hits only this PR's own changeset quoting it; no rendered row survives.TypeScript Type Check(the job that runs every spec artifact gate,check:docsandcheck:authorable-surfaceincluded) concludedsuccess` on this head, which is the CI reading that the JSON-schema and docs artifacts are current.
• Accept set unchanged, only the parsed default moves. editInline stays z.boolean(); .default() alone moves; the inferred type is boolean on both sides, which is why check:api-surface has nothing to say. The five pins hold the contract: absent parses to true on the schema and through ListViewSchema; explicit false and true keep their values; the view-level inlineEdit: true still parses with no userActions block materialised (the fold is the renderer's). The dev's ablation (.default(false) back) reads as exactly the two default pins red, blob-equal restore — consistent with the diff; not re-run here.
• Every other place that still states the old default, at head: .changeset/console-20c6d351ad74.md lines 24 and 71 say the console at pin 20c6d351ad74 reads an absent key with the spec's default, off — true of that build until objectui#12086 lands and the pin moves; carrier named in ③. Root CHANGELOG.md:1677 is release-owned history and stays. content/docs/ui/views.mdx:111 lists the view-level inlineEdit with no default; layout-dsl.mdx:667 is record:details' own inlineEdit, a different key; object.zod.ts:2043 says only that editInline is a view key. No example and no skills/** file in this repository states the old default; no in-repo metadata declares editInline at all. The permission gate and objectui are untouched, as the card requires.

② Semver level

• minor is right; major is not, even though pre mode would admit it. check-changeset-no-major.mjs ships breaking changes as minor for the launch window (end condition: GA, not reached); its RC exemption lets a major through while .changeset/pre.json reads mode: pre, tag next, so major was available — but the repository's own default-flip precedents on this line all took minor: view-pagination-page-size-default-50 (8ecbe0f, under ### Minor Changes), approval-escalation-enabled-default-flip ("shipped as minor under the repo's launch-window convention"), audience-posture-default-invite-only (same words). Protocol 12's rest-requireauth-default-flip (7c09621) sits under ### Major Changes, but that was before the window and is not the rule now. The fixed group already carries the pending major that opens the v18 line (22080-v18-line-opens.md, 22130-protocol-18.md), so this minor moves no version and rides the 18.0.0-next.N train. The BREAKING banner is present and is the only breaking signal the ADR-0087 gate reads here (no major, no bang on the changeset's first line, no narrowing arm); the marker registered list-view-edit-inline-default-on resolves at head and is new in this diff; the FROM → TO table and the one-line fix are there. Held.
• The Clause-② arm is NOT right, and this is the one finding of record. The line answers 「本卡放宽接受集或扩大公开面吗」, and the ruled definition (pr-automation.yml WHICH LEVEL, the maintainer's ruling on finding(changeset): two independent contract reviews read the repo's own history to opposite bumps for "add an exported symbol to a published index" #15294) is "a purely additive widening of a published package's public surface (a new exported symbol on an index, a new accepted key or value)". This diff adds no export, no key and no value; the changeset itself says "The accept set is unchanged" and declares the act BREAKING, and the ADR-0087 gate's own docblock says "A widening is not a break". One act cannot be a purely additive widening and a break at once. A moved default is neither a widening nor a narrowing of the accept set, so the truthful declaration is Clause-②: no with no arm, breaking-ness carried where the window puts it — the banner and the marker. The card's parenthetical ("a view declaring nothing gains an affordance it did not have") reads the console's behaviour, not the package's surface. Origin: the card and the dispatch claim 6094254055; the dev copied the line as os-dev.md requires (照抄, ⛔ 不自造), so the correction is the seat's — on the claim, then on the PR body (line 2) and .changeset/22605-spec-list-view-edit-inline-default-on.md:13. It matters because the line ships verbatim into packages/spec/CHANGELOG.md (416 Clause-② lines there today) and is what every later census reads. No gate moves on the correction: the level axis reads no as not-declared and exits 0 with minor untouched; the ADR-0087 gate still reads BREAKING from the banner and still finds the marker. The pm-dispatch instrument-discipline reference assigns the direction to this review (Clause-② 的方向由达档复核裁), so it is ruled here: no.

③ Boundary flags

• Dev deviations, each answered. spec-changes.json and the upgrade guide reverted to base: right — since [Decision] generated protocol-change records (spec-changes.json per-major section, docs/protocol-upgrade-guide.md): keep committing them, or generate them at publish like the release section #22449 B′ the PR-stage check:spec-changes / check:upgrade-guide project the registry in memory, and origin/main has made the guide a pointer stub; the entries README and the AGENTS.md regenerate table still say otherwise, which is the dev's first out_of_scope_findings entry — carrier spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485, named in the ACCEPT, answered. check:dual-build-cjs-loads, check:dts-closure, check:published-readme-exports NOT MEASURED locally: CI owns them — Build Core is success on this head, Lint & Repo Gates still running, named below. Seven pushes: AGENTS.md demands a WIP push before every minutes-long step; not a deviation. No re-merge after main moved: origin/main is now ad1495796e, four commits past this PR's merge-base (plugin-audit, plugin-approvals ×2, metadata-protocol), none touching a path of this diff; the queue rebuilds onto current main. Model-free commit trailers: what the pre-push hook requires; not a deviation. test:repo beyond three files declared to CI: Test Core still running, named below.
• out_of_scope_findings, both answered. (1) README / AGENTS.md regenerate instruction drift → spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485, as above. (2) The console pin changeset's item (4) states the spec default as false: accurate about the console at that pin, and the next pin bump's changeset is where the new reading belongs — the carrier the ACCEPT names. Note for that bump: until the pin moves past objectui#12086, this changeset and the console one ship in one train reading "spec default true" beside "the console reads the spec's default, off"; each is true of its own artefact, so the pin-bump changeset must say which build it describes.
• Landing order — escalated, not corrected. PR docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630 (the skill paragraph, Tier H) is already in the merge queue: refs/heads/gh-readonly-queue/main/pr-22630-…, third behind fix(metadata-protocol): a sealed-item refusal of a create names a free name, not the edit remedy (#22591) #22623 and feat(i18n): the engine translates a screen's title and description in the run's locale #22626, after two APPROVED reviews on its head by os-zhuang (in GOVERNED_APPROVERS), then ready_for_review, auto_merge_enabled and added_to_merge_queue by that same account — the maintainer's hand, within Tier H's allowance, and no seat act is measured to dequeue it. That reverses the seat's stated order (ACCEPT 6094884470 and the final 速读 on docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630: PR 2 waits for PR 1). The consequence is real and customer-reaching: between the two landings skills/objectstack-ui/rules/list-views.md on main teaches "on by default" while view.zod.ts:1535 on main still reads .default(false) and @objectstack/spec@latest is 17.7.0, and the published catalog is installed straight from the repository (npx skills add objectstack-ai/objectstack/skills, npm create objectstack), not from a release. The seat asks the maintainer whether to hold docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630 (hand removal from the queue) or to accept the window — and either way enqueues this PR's corrected head the moment its re-review PASSes and the two pending contexts are green, so the window is hours, not days.
• CI on this head at the time of writing. Required: Build Core, Dogfood Regression Gate, Temporal Conformance (live PG + MySQL), Governed Surface Queue Guard, TypeScript Type Check all success; Lint & Repo Gates in_progress; Test Core not yet concluded (shard 4/6 running). No advisory red.
• Re-review scope. Every section above except the arm is held. A head that differs from 24d9bdcbf0 only in the two declaration lines (PR body line 2, changeset line 13) re-reviews against this record's ① and ③ unchanged; nothing in packages/spec/src/**, the ledgers or the generated pages is asked to move.

Implemented-by: claude/issue-22605-edit-inline-default-on
Reviewed-by: session_01KNKBCRDJCu5tGy3TEbvtrF

VERDICT: FAIL

Generated by Claude Code

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: fc3733b9714493610f60cc80e7849f36471a6548
Local-runs: none

Round 2, scoped: what moved since the round-1 head 24d9bdcbf0 (record 6095058376, FAIL), and whether that record's FAIL items are cured. Read-only: the card body and its comments through the dev's patch report 6095196973 (the seat's patch order 6095101276 included), the PR body and file list, the two-point diff 24d9bdcbf0..fc3733b971, the net diff against the merge-base e22315238f (unchanged since round 1), and the head's check-runs. Nothing built, run or re-run; the dev's local readings are reported here as the dev's, never adopted as a gate's.

What moved. git diff --name-only 24d9bdcbf0 fc3733b971 is exactly the four files the patch report names — the changeset, packages/spec/scripts/lib/default-changes.ts, entries/semantic/18.list-view-edit-inline-default-on.ts, registry.ts — 4 files, +15 / −7, one commit, 24d9bdcbf0 an ancestor of the head. The net diff against the merge-base is still the PR's 9 files (+241 / −7: round 1's +233 plus this hop), identical to the three-dot diff against origin/main 1b99388505. main is seven commits past the merge-base; none touches a path of this diff, and none has touched src/migrations, default-changes.ts, authorable-defaults/ or the two projections since the round-1 merge, so no regen-merge was owed. Everything round 1 held in view.zod.ts, view.test.ts, ui.json and the two reference pages is untouched by this hop and stays held.

① Derived judgments

• The D3 entry — right. id, surface, replacement and acceptanceCriteria are byte-identical to the round-1 head, compared field by field. reason moves in one line: "the one-vocabulary rule of objectui#5144 stands" is now "the one-vocabulary rule objectui ruled for the fold stands". Scanned with the pin's own shape (# plus four or five digits, word-bounded) over every non-comment line of the entry: no hit. The only #5144 left in the file sits in the leading // comment, which now also says why it lives there ("the citation lives here, not in the printed guidance, which carries no tracker number"). That is the right home: os migrate meta prints surface, replacement, reason and acceptanceCriteria and never the comment, and the tracker ids main's other entries carry all sit comment-side. Round 1 read "12 of 414 entries carry an objectui#N" as precedent for a printed string; that was wrong — they are comments — and the pin said so on the round-1 head.
• The registry.ts hunk — right, the generator's copy. The 52-line block inside step18 is the entry's leading comment and literal rendered exactly as renderRegion renders them (build-migration-registry.ts): every comment line indented four spaces, the declaration line as {, the closing }; as },, every literal line indented four — compared line for line against the entry file at this head, no difference. It sits in id order, before list-view-navigation-view-retired. The hunk's +6 / −2 against the round-1 head is the comment run growing by four lines plus the one reason line, which is what the generator copies by design; the dev's deviation note on this is answered. The CI reading of generator identity is check:migration-registry, a step of Lint & Repo Gates, named under ③.
• The DEFAULT_CHANGES_BY_MAJOR[17] row — right. key, from, to unchanged; the row still sits under 17: [, the only major that authorises a hop (CURRENT_MAJOR is the package major), and still names the D3 entry by id. Its reason takes the same one-sentence rewrite; scanned with the same shape: no hit. No gate demanded this one — the pin reads the registry, not this ledger, and main's own ledger still carries tracker ids in four of its earlier reason strings — but the row's reason is printed into the build log on every run (build-schemas.ts, the "declared default change(s) accepted" block), so the rewrite follows the rule that text an author is shown carries the lesson and not the number, and it loses nothing. check:authorable-surface accepted the moved fingerprint on this head (Type Check · source gates concluded success), as on the round-1 head.
• The changeset's Clause-②: line — right at no, judged under ②. Nothing else in the changeset moved: the **BREAKING** banner, the FROM → TO table, the one-line fix, the measured reach and the adr-0087 marker registered list-view-edit-inline-default-on are byte-identical to the round-1 head, and the marker still resolves to the entry at this head.

② Semver level

• minor — held from round 1, unchanged on this head: the launch-window convention ships a breaking change as minor (check-changeset-no-major), the three default-flip precedents on this line took minor, and the fixed group already carries the pending major that opens the v18 line, so the level moves no version on its own.
• Clause-②: no — right, and round 1's one finding of record is cured. The line answers 「本卡放宽接受集或扩大公开面吗」, and the ruled definition of a widening (pr-automation.yml WHICH LEVEL, the maintainer's ruling on finding(changeset): two independent contract reviews read the repo's own history to opposite bumps for "add an exported symbol to a published index" #15294) is "a purely additive widening of a published package's public surface (a new exported symbol on an index, a new accepted key or value)". This diff adds no export, no key and no value — editInline is still z.boolean(), only .default() moves, and the changeset itself says "The accept set is unchanged". A moved default narrows nothing either, so no arm: no (narrowing) would declare an accept-set shrink that does not exist, and no (widening) is the malformed pair. Breaking-ness rides where the window puts it: the **BREAKING** banner, which check-adr-0087-registration's breakingDeclaration reads as a signal of its own with no help from the arm (its self-test pins a bare Clause-②: no as NOT a breaking signal and the banner as one), plus the registered marker, which resolves. The level axis of check-changeset-no-major reads no as not-declared and stands down with minor untouched. Check Changeset, the job that runs both gates against the merge-base, concluded success on this head — the gate reading for both.
• The PR body's line 2 agrees: Clause-②: no, bare, on a line of its own, which readClause2Line returns as declared / no / no arm. The correction chain round 1 asked for is complete on every machine-read carrier: the claim 6094254055 (corrected by the seat), the PR body's line 2 (the seat), the changeset's line 7 (the dev). One residue, named in ③. Round 1 and the patch order both numbered the changeset's declaration "line 13"; it is line 7, and the dev edited the right line — the misnumbering was round 1's.

③ Boundary flags

• Round 1's FAIL items, each. (1) The Clause-② arm: cured, above. (2) Test Core (4/6) red on 24d9bdcbf0 — confirmed by that head's check-runs (Test Core (4/6) failure, the Test Core fan-in failure), the failing assertion being migrate-meta-engine-guidance.test.ts's "the printed guidance cites a tracker id" on this PR's entry. Cured on this head by its own check-run: Test Core (4/6) concluded success, and the required Test Core fan-in with it (both named below). The diff-side reading is that the head's printed fields are clean under the pin's own regex; the dev reports red reproduced before the patch and green after it, with the CLI closure build between the registry change and the reading on both legs (the pin resolves @objectstack/spec/migrations through the built dist), a shape consistent with the diff and not re-run here.
• The dev's deviations, each answered. Line 7, not 13: the right line moved, above. Worktree recreated at 24d9bdcbf0 with no merge of origin/main: right — nothing on main since touches a path of this diff or a registry input, and the queue rebuilds onto current main. The changeset's prose still citing objectui#5144 and objectui#12086: not owed a change. The pin's surface is the printed migrate-meta guidance, the registry's four fields; check:doc-authoring's roots are .claude, docs, skills and content, not .changeset/, and its tracker rule is for runtime strings and refusal prose. A changeset ships to packages/spec/CHANGELOG.md, where 304 lines on main already cite an objectui#N and 59 of the 259 .md files under .changeset/ carry a #NNNN: a changelog citing the sibling repository's ruling record is this repository's norm, and here the citation is the provenance of the rule the changeset says still stands. The same holds for the objectui#5144 in the view.zod.ts docblock and the view.test.ts comment, both held from round 1 and neither printed. The registry hunk carrying the comment run: by design, above.
• One residue on the PR body, owed to the seat, non-blocking. Line 17 of the body (the "What changed" bullet for the changeset) still describes the changeset as carrying Clause-②: yes (widening) and argues minor from "Clause-②: yes takes at least minor". No gate reads it: the declaration reader takes the first key-initial line (line 2) and keeps a mid-line backticked mention only as a residue to quote when no declaration exists, which is why Check Changeset and the claim guard are green on this head. It ships nowhere — a PR body is not a changelog. But it contradicts line 2 of the same body and the changeset it describes, so the seat corrects that bullet (and may note the citation move in the D3 bullet) in the act that enqueues; a body edit moves no head and asks for no re-review.
• The landing-order window round 1 escalated is now open, and this PR is its only remedy. PR docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630 landed first by the maintainer's hand (f368b7e980; the seat's reopen note 6095085980): the published skill on main teaches that a list edits in place by default, while view.zod.ts on main still reads .default(false) and the published spec is unchanged. The seat's order calls the patch urgent for that reason, rightly; the window closes when this head lands, and nothing else in this diff is asked to move. The seat enqueues on this record once every required context is green.
• Carriers held from round 1, unchanged: the console pin changeset's "spec default false" line (carrier: the next pin bump), the README / AGENTS.md regenerate-table drift (spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485), spec-changes.json and the upgrade guide not regenerated (right, since [Decision] generated protocol-change records (spec-changes.json per-major section, docs/protocol-upgrade-guide.md): keep committing them, or generate them at publish like the release section #22449 B′).
• CI on this head, read at 2026-10-10T07:55Z. Required (the ruleset's seven): TypeScript Type Check success (Type Check · source gates, · workspace, · debt ledger, · consumer gates all success beneath it), Build Core success, Dogfood Regression Gate success, Temporal Conformance (live PG + MySQL) success, Governed Surface Queue Guard success, Lint & Repo Gates success (the job that runs check:migration-registry — the generator-identity reading for the registry hunk — and check:doc-authoring), Test Core success with every shard success, Test Core (4/6) among them. The last three concluded while this record was being written and were read again before it was posted: 35 check-runs on the head, none open, none red. Advisory: every concluded check success or skipped (Check Changeset, the three PR-context guards, Spec property liveness, Check PR Size, Console Pin Gate skipped, the packed-tarball smoke skipped); no advisory red.
• State of the PR: draft, needs:contract-review carried, no auto-merge, no governed path in the file list — the review is owed by the claim's own term, not by Tier S.

Implemented-by: claude/issue-22605-edit-inline-default-on
Reviewed-by: session_01KNKBCRDJCu5tGy3TEbvtrF

VERDICT: PASS

Generated by Claude Code

Landing pre-checks at fc3733b971, by the owning seat: every check green, queued

domain:spec seat 3 (#18883) · zhuangjianguo · session session_01KNKBCRDJCu5tGy3TEbvtrF · 2026-10-10T07:59Z · holder of claim 6094254055 on #22605.

• The review: contract review round 2 PASS 6095355885 names this head. It judges round 1's FAIL items cured: the changeset's Clause-②: no is right for a moved default, and Test Core (4/6) is green. The D3 entry's printed text carries no tracker id, and its id, surface, replacement and acceptanceCriteria are byte-identical to round 1's head.
• The seat's own read of the patch: git diff 24d9bdcbf0 fc3733b971 is the 4 files the order named (+15 / −7), and nothing else moved. In each, only the Clause-②: line or the one-vocabulary sentence's citation changed; the citation moved into the entry file's leading comment.
• The review's one residue, fixed in this act: the PR body's "What changed" bullet for the changeset still said Clause-②: yes (widening) and argued minor from it. It now reads Clause-②: no. Line 2 already agreed, and no head change follows.
• CI: 42 check-runs: 38 success, 4 skipped. check-expected-skips --pr 22629 reads every skip in the roster.
• Governed: check-governed-merges --pr objectstack-ai/objectstack#22629 reads NOT governed; 248 changed lines.
• Closing keywords: the body carries Part of #22605 alone, and no commit message carries one. The seat closes spec: ListView.userActions.editInline defaults to true in v18 and editInline: false is the opt-out; objectui reads an absent key as on (maintainer ruling 2026-10-10, after objectui#5144) #22605 with its landing note, since its skill half landed as PR docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630.
• main drift since the merge base e22315238f: main (now 1b99388505) moved none of the PR's 9 paths. GitHub reports mergeable: true (clean). It is a D3-entry PR under the temporary rule-A relaxation (6091886885 on spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485), so no pre-queue sync is owed.
• Why now: PR docs(skills): a list view edits in place by default; a read-only list declares editInline: false (#22605, skill half) #22630 put the skill paragraph on main first, so until this lands, the published skill says inline editing is on by default while the spec parses an absent editInline to false.

needs:contract-review comes off; pr_ready and automerge_enable follow.

Generated by Claude Code