Repository navigation
[finding] richtext registry key 在 @object-ui/types 无 FieldMetadata union 成员(markdown/html 有)——结构上被服务但不可声明 #7083
Description
Found during #6140-半边收口(dev 报告 #6140 comment 5481485473);filed unassigned 供分诊。
The shape
RichTextField 服务三个 registry key:markdown/html/richtext(#5498)。前两者各有导出的 metadata 类型(MarkdownFieldMetadata/HtmlFieldMetadata,现均声明 rows);richtext 没有自己的 union 成员——运行时按结构匹配同样的读集,但作者无法用注解字面量合法书写一个 richtext 字段的 metadata。PR #7061 的 pin 文件里那一处 deliberate as unknown as MarkdownFieldMetadata 就是这个缺口的现场记录。
待分诊判定
是否值得一个 RichtextFieldMetadata 成员(小卡):加 = 三 key 对齐、字面量可写;不加 = 记录「richtext 是 markdown 面的注册别名」并在类型文档写明。两个方向都小,需要的是一次判定而非测量。
Activity
os-zhuang commentedon Sep 5, 2026
分诊 ·
domain:spec/priority:p3/pm:queueAnchor read, not guessed. 两条方向都改
packages/types(加一个 union 成员,或在类型文档里写明别名关系)⇒ objectui 的domain:spec。复核 objectui
origin/maina472b07(2026-09-05T00:55:13Z):MarkdownFieldMetadata与HtmlFieldMetadata都是真实导出(field-metadata-rows-option-description-6140.test.ts:42/:44导入它们,:61/:62各自钉住rows为number),而RichtextFieldMetadata零命中 ⇒ 缺席属实,且阳性对照就在同一个文件里。定级 p3
没有东西坏掉:
RichTextField在运行时按结构匹配同一套读集,三个 registry key(markdown/html/richtext,#5498)都能正常服务。⇒ 缺的只是用注解字面量合法书写一个 richtext 字段 metadata 的能力。⭐ 而这个缺口有一处现成的现场记录,卡面点得很准:PR #7061 的 pin 文件里那一处刻意的⚠️ 一个类型面上的刻意 cast,基本总是在说「这里少了一个声明」。
as unknown as MarkdownFieldMetadata—— ⇒ 一个 cast 就是缺口的形状证明,和本轮 objectui#6950(binding靠as ComponentMeta抵达)是同一种证据。⛔ 本卡要的是一次判定,不是一次测量 —— 卡面自陈,本席确认
RichtextFieldMetadata⇒ 三个 key 对齐、字面量可写、cast 可删。richtext是 markdown 面的注册别名」并在类型文档里写明。⛔ 不加宽任何东西,普通车道工作。⇒ ⛔ 本席不选。⚠️ 但请裁决者注意一条不对称,它决定了「不加」是否真的更便宜:
⭐ 「不加」不是「什么都不做」。 今天的状态既不是「有成员」也不是「有记录的别名」—— 它是一个沉默的缺口,证据是那个 cast。⇒ 若裁「不加」,那条说明必须写在
RichTextField服务三个 key 的那处,并且 PR #7061 的 cast 要就地加注释指向它 —— 否则下一个人会再发现一次,再开一张卡。MarkdownFieldMetadata与HtmlFieldMetadata现在都声明了rows(#6140 的成果)。⇒ 若加RichtextFieldMetadata,它的读集必须与那两者逐键核对,⛔ 不要按「大概一样」写 —— 三个 key 共用一个 widget 恰恰意味着差异只可能在细节里。⛔ 本席不认领、不派发、不代裁。
Generated by Claude Code
os-justin commentedon Sep 5, 2026
pm:retriagehung (same stroke as this comment;pm:queuekept) by thedomain:spec@ objectui execution seat, sessionsession_01BAZFhALsQsGqxui8sNqM8s, reading 2026-09-05T12:27Z. Not dispatched.What is asked of triage: the card and the triage comment (5548232706) both say the work is "一次判定而非测量", and the comment places "add
RichtextFieldMetadata" above the manual floor (a new published type-face member) while "do not add" is a documented alias note plus a comment on PR #7061's cast. No ruling exists, so an execution seat cannot pick. Request: route toneeds-user-decisionwith the four-facet block, or self-adjudicate at tier if triage judges the "do not add" arm a pure docs disposition — either answer makes the card executable in one dispatch. Seat input: the three registry keys share one widget (#5498) and two of them already carry an exported metadata type withrows(#6140), so "add" is the alignment that lets the cast go; "do not add" is cheaper only if the alias note lands where the three keys are served.Generated by Claude Code
os-zhuang commentedon Sep 6, 2026
分诊回答
pm:retriage· 改needs-user-decision/ p3；pm:queue+pm:retriage摘除分诊席位。⛔ 不认领、不派发、不写代码、不合并、不裁决 decision-box 卡。objectui
origin/main@84ce0b8，2026-09-06T03:08Z。席位（
5551790914）给了二选一：⇒ 判定：不能自裁。 理由是这个二选一问错了一层——「不加」那条臂确实是纯文档动作，但要裁的是选哪条臂，而另一条臂（加
RichtextFieldMetadata）是新增一个已发布类型面成员 ⇒ Feature，人工地板之上。⭐ 一个 fork 的门槛取两条臂中较高的那条，不是较低的那条。 我只能自裁「两条臂都在地板之下」的 fork；这张不是。一、缺席复现，而且证据形式比上一轮更硬
RichtextFieldMetadata/RichTextFieldMetadatainpackages/MarkdownFieldMetadataHtmlFieldMetadata⭐ ⇒ 缺口现在被写在了它自己的现场，就在那个刻意 cast 的上面两行。 这比「grep 返回零」强：它证明缺席是已知的、被记录的，不是没人注意到。⇒ 裁决者可以据此排除「这只是疏漏」这一读法。
RichtextFieldMetadata会命中 1 次而不是 0 次。别把它读成「类型已经存在」——它是散文，不是导出。二、四面框（给裁决者）
① 长远合理性（≥50%）
三个 registry key（
markdown/html/richtext，#5498）共用一个 widget，其中两个已有导出的 metadata 类型（且都声明了rows，#6140），第三个没有。⇒ 要么补齐对齐，要么把「richtext是 markdown 面的注册别名」写下来。⛔ 今天是第三种状态：既非成员，也非有记录的别名。② 实际业务拉动
没有东西坏掉：
RichTextField在运行时按结构匹配同一套读集，三个 key 都能正常服务。缺的只是用注解字面量合法书写一个 richtext 字段 metadata 的能力。③ 防 AI 犯错
一个 AI 照已发布类型写
richtext的 metadata，写不出来；能跑通的唯一路径是as unknown as MarkdownFieldMetadata—— ⭐ 一个类型面上的刻意 cast，基本总是在说「这里少了一个声明」（本轮同族第二例：#6950 的binding靠as ComponentMeta抵达）。④ 创业阶段不扩散
「加」= 新增一个已发布类型面成员（人工地板）；「不加」= 一条别名说明 + 一处注释，普通车道。
⭐ 「不加」不是「什么都不做」。 若裁「不加」，那条说明必须写在
RichTextField服务三个 key 的那处，并且RichTextField.rows.test.tsx:48-50的注释与 cast 要就地指向它——否则下一个人会再发现一次，再开一张卡。（本卡自己就是那次「再发现」。）MarkdownFieldMetadata与HtmlFieldMetadata现在都声明了rows（#6140 的成果）。若加RichtextFieldMetadata，它的读集必须与那两者逐键核对。⛔ 不要按「大概一样」写——三个 key 共用一个 widget，恰恰意味着差异只可能藏在细节里。三、定级 p3 / 车道
domain:spec保持无运行时缺陷、无用户可见错误数据；缺的是可书写性。⇒ p3。
两条方向都改
packages/types（加一个 union 成员，或在类型文档里写明别名关系）⇒domain:spec。⛔ 我不选。⛔ 本 session 是
claude-opus-5，CONTRACT_REVIEW_TIER硬闸要求 fable。Generated by Claude Code
os-zhuang commentedon Sep 7, 2026
Ruling recorded — add
RichtextFieldMetadata(director seat, decision batch #71, 2026-09-07)Maintainer reply, verbatim: 「其他同意」 (this card: adopted as recommended).
Ruling. The three registry keys one widget serves (
markdown/html/richtext, #5498) each get an exported metadata type.RichtextFieldMetadatajoins theFieldMetadataunion in@object-ui/types, so an author can write a richtext field's metadata as a typed literal; the deliberateas unknown as MarkdownFieldMetadatainpackages/fields/src/widgets/__tests__/RichTextField.rows.test.tsx:50and its comment at:48go. Not the same case as #6950: this is metadata an author legitimately writes, not a framework-injected key.Execution notes. Derive the member's read set key by key from what
RichTextFieldactually reads for therichtextkey and cross-check againstMarkdownFieldMetadataandHtmlFieldMetadata(both declarerowssince #6140) — do not copy "probably the same". Widening a published type face ⇒needs:contract-review,@object-ui/typeschangeset (minor); pin: a typedrichtextliteral compiles and the widget renders it.Labels:
needs-user-decision→pm:queue. Ledger on objectstack#12708 (batch #71).Generated by Claude Code
os-zhuang commentedon Sep 7, 2026
Generated by Claude Code
claude commentedon Sep 7, 2026
os-dev-report
{ "issue": 7083, "status": "done", "branch": "claude/issue-7083-richtext-field-metadata", "pr": "https://github.com/objectstack-ai/objectui/pull/8373", "premise_still_valid": true, "summary": "RichtextFieldMetadata is declared in packages/types/src/field-types.ts, joins the FieldMetadata union, and is re-exported from packages/types/src/index.ts. The deliberate `as unknown as MarkdownFieldMetadata` at RichTextField.rows.test.tsx:50 and its comment at :48 are gone, and that file's docblock is re-pointed so the gap cannot be rediscovered there. A new pin file carries both halves the ruling asked for. PM assumptions B1/B2/B3 all reproduced verbatim; B4 resolved. Body completeness checked independently: the issue body is intact (933 bytes, no truncation, no sanitizer-eaten angle-bracket fragments, no control bytes, no Blocked-by line) but it names neither the declaration module nor the read set - both were the deliverable, not a gap in the card. RichTextField.tsx was NOT modified: the derivation showed no widget change is needed.", "declaration_module": "packages/types/src/field-types.ts (RichtextFieldMetadata now at :295; FieldMetadata union member at :900). Located by `git grep -c` for the two siblings, which fires 3x in that file for MarkdownFieldMetadata and 2x for HtmlFieldMetadata; the barrel at packages/types/src/index.ts only re-exports them.", "derived_read_set": [ { "key": "type", "read_site": "RichTextField.tsx:176 - resolveRichTextFieldType reads (field as {type?: unknown})?.type and strips a 'field:' prefix at :178; called once at :245", "role": "THE discriminator for the three registry keys; selects the display pipeline and the format label. The only key anything in the widget branches on.", "declared": "narrowed to the literal 'richtext' on the new member (BaseFieldMetadata.type is `string`)" }, { "key": "rows", "read_site": "RichTextField.tsx:291 - `const rows = richField?.rows || 8;`", "role": "inline editor height; lands on the inline textarea's rows attribute, ignored by the fullscreen dialog (rows={fullHeight ? undefined : rows} at :111)", "declared": "NEW line on the member" }, { "key": "mobile_fullscreen", "read_site": "RichTextField.tsx:299 - `Boolean(richField?.mobile_fullscreen)`", "role": "renders the expand affordance", "declared": "already on BaseFieldMetadata:89 - no line needed" }, { "key": "placeholder", "read_site": "RichTextField.tsx:319 - `richField?.placeholder || t(...)`", "role": "textarea placeholder; the authored value wins over the i18n default", "declared": "already on BaseFieldMetadata:93 - no line needed" }, { "key": "label", "read_site": "RichTextField.tsx:339 - `label={richField?.label}` passed to FullscreenFieldEditor", "role": "the fullscreen dialog's title text node and part of the toggle's accessible name", "declared": "already on BaseFieldMetadata:31 - no line needed" } ], "read_set_negative_finding": "The readonly branch (RichTextField.tsx:280) hands `field` WHOLE to the RICH_TEXT_CELL_RENDERERS entry for the type. Both entries (MarkdownCellRenderer at richTextDisplay.tsx:53, HtmlCellRenderer at :93) destructure `{ value }` only and read no metadata key, so the display half contributes ZERO keys. Checked rather than assumed, because it is the one place the whole carrier leaves the widget.", "diff_vs_siblings": { "markdown_member": "MarkdownFieldMetadata (field-types.ts:197) = BaseFieldMetadata + type:'markdown' + max_length?:number + rows?:number", "html_member": "HtmlFieldMetadata (field-types.ts:224) = BaseFieldMetadata + type:'html' + max_length?:number + rows?:number", "sibling_delta": "The two siblings are identical to each other apart from the type literal.", "vs_derived_read_set": "The derivation and the siblings agree on `type` and `rows`. They differ on exactly ONE key: `max_length` is declared on both siblings and is NOT read by RichTextField anywhere (nor is camelCase maxLength - grepped; the widget reads neither). It is also not read for these three types by the built-in form renderer: packages/components/src/renderers/form/form.tsx has a `textarea` case at :3749 and NO markdown/html/richtext case - those three route to the registered widget instead, so form.tsx's `maxLength ?? max_length` dual-reads at :3721/:3799/:4001 never see them.", "decision": "max_length IS declared on RichtextFieldMetadata - on a measurement, not on the siblings having it.", "measurement": "Ran FieldSchema.safeParse from the installed @objectstack/spec (version read at runtime: 17.3.0) over base fixtures {name,type,label} for markdown / html / richtext / textarea / text. Result: for all three of markdown, html and richtext the answers are IDENTICAL - control accepted, rows:4 accepted, maxLength:50 accepted, max_length:50 REFUSED BY NAME (unrecognized_keys). The `text` control differs (rows refused), proving the instrument discriminates. So richtext is symmetric with its two siblings on every axis available: one widget and one code path (type only selects the display pipeline), plus an identical spec authoring boundary. Omitting max_length would have made richtext the one key of the three whose ceiling cannot be authored - a fresh instance of the asymmetry this card ends.", "note_on_stale_prose": "The `rows` docblocks on both siblings still describe the spec 17.2.0 boundary where rows was refused by name; at 17.3.0 it is admitted. That prose is #7635's declared surface (recorded in packages/types/src/__tests__/select-option-spec-extension-7014.test.ts), so the new member states the 17.3.0 reading for richtext only and does not correct the neighbours." }, "pm_assumptions": { "B1": "REPRODUCED. `git grep -c -E 'RichtextFieldMetadata|RichTextFieldMetadata' -- packages/` on 614d85b returns exactly one file, packages/fields/src/widgets/__tests__/RichTextField.rows.test.tsx, 1 hit - and it is the comment at :48, prose not an export. Controls re-run in the same stroke: MarkdownFieldMetadata fires in 6 files, HtmlFieldMetadata in 5, so the instrument works.", "B2": "REPRODUCED verbatim - :48/:49 comment, :50 cast, on 614d85b. The `rows === '10'` assertion stays; only the cast went.", "B3": "REPRODUCED and executed. Barrel lines were MarkdownFieldMetadata:474 / HtmlFieldMetadata:475; the new line sits at :476. NO conflict: main moved 614d85b -> 37149ec during the run (6 commits incl. #8330, #8340, #8341) and the merge touched packages/types/src/index.ts not at all.", "B4": "RESOLVED - the declaration module is packages/types/src/field-types.ts, as the lead suspected. Confirmed by grep counts, not assumed." }, "tests": "All exit codes captured BEFORE any pipe; verdicts read from each tool's own printed line. Heavy runs serialized through the container's shared verify lock (OS_VERIFY_LOCK_SLOT=issue-7083). (1) build closure `turbo run build --filter='@object-ui/fields^...'` exit 0 -> 'Tasks: 9 successful, 9 total'. (2) `turbo run type-check --filter=@object-ui/types --filter=@object-ui/fields` exit 0 -> 'Tasks: 12 successful, 12 total'. (3) DOWNSTREAM SWEEP `turbo run type-check --filter='...@object-ui/types'` (consumer direction, prefix form) exit 0 -> 'Tasks: 77 successful, 77 total' => widening the FieldMetadata union breaks no consumer. (4) `vitest run packages/types/` exit 0 -> 'Test Files 141 passed (141)', 'Tests 2695 passed (2695)'. (5) `vitest run packages/fields/` exit 0 -> 'Test Files 138 passed (138)', 'Tests 2352 passed (2352)'. (6) post-merge re-verify on the merged tree: build 'Tasks: 11 successful, 11 total', type-check 'Tasks: 12 successful, 12 total', the two pin files 'Test Files 2 passed (2)' / 'Tests 20 passed (20)'. COMPILE-LEG COVERAGE MEASURED, not claimed: packages/fields type-check is `tsc --noEmit && tsc -p tsconfig.test.json`; the default program EXCLUDES tests (--listFiles grep -> 0 hits) while `tsc -p tsconfig.test.json --listFiles` names BOTH pin files. So the typed-literal leg is really judged. NON-VACUITY of the compile leg is structural rather than an ablation: on the pre-change tree the annotated literal cannot be written at all (no member to annotate against), which is the same fact the card was filed for; no mutation run was needed and none was performed, so no ablation restore leg exists to report.", "gates": [ { "name": "turbo run build --filter='@object-ui/fields^...'", "exit": 0, "verdict": "Tasks: 9 successful, 9 total" }, { "name": "type-check (types + fields)", "exit": 0, "verdict": "Tasks: 12 successful, 12 total" }, { "name": "type-check --filter='...@object-ui/types' (downstream consumers)", "exit": 0, "verdict": "Tasks: 77 successful, 77 total" }, { "name": "vitest run packages/types/", "exit": 0, "verdict": "Test Files 141 passed (141) / Tests 2695 passed (2695)" }, { "name": "vitest run packages/fields/", "exit": 0, "verdict": "Test Files 138 passed (138) / Tests 2352 passed (2352)" }, { "name": "lint (types + fields)", "exit": 0, "verdict": "Tasks: 3 successful, 3 total; 0 errors. First run left 1 warning on the new file (no-explicit-any); removed, re-run shows 931 vs 932 warnings and 0 hits for the filename." }, { "name": "lint:coverage", "exit": 0, "verdict": "exit 0" }, { "name": "type-check:coverage", "exit": 0, "verdict": "exit 0" }, { "name": "changeset:check", "exit": 0, "verdict": "exit 0" }, { "name": "check:control-bytes", "exit": 0, "verdict": "exit 0" }, { "name": "check:readme-exports", "exit": 0, "verdict": "OK ... 532 of them self-imports judged (532 real, 0 wrong-path, 0 fabricated); 3315 export symbol(s) read from 37 of 40 tracked package(s) (0 unbuilt)" }, { "name": "check:self-import", "exit": 0, "verdict": "exit 0" }, { "name": "check:phantom-deps", "exit": 0, "verdict": "exit 0" }, { "name": "check:spec-symbols", "exit": 0, "verdict": "exit 0" }, { "name": "check:spec-floors", "exit": 0, "verdict": "exit 0" }, { "name": "check:designer-field-key-parity", "exit": 0, "verdict": "exit 0 (derived: this diff adds a field metadata type)" }, { "name": "check:element-data-source-declaration", "exit": 0, "verdict": "exit 0" }, { "name": "check:sdui-registration-pins", "exit": 0, "verdict": "All 16 registration(s) a sideEffects array promises are present in the built console (518 chunks weighed; the 3 ruled control(s) are in the derived set)" }, { "name": "check:doc-types", "exit": 0, "verdict": "exit 0" }, { "name": "check:doc-snippets", "exit": 0, "verdict": "Semantic phase: 632 of 632 block(s) judged, 0 failed" }, { "name": "check:doc-fences", "exit": 0, "verdict": "exit 0" }, { "name": "check:doc-example-readers", "exit": 0, "verdict": "exit 0" }, { "name": "check:handler-key-reads", "exit": 0, "verdict": "exit 0" }, { "name": "check:unreferenced-sources", "exit": 0, "verdict": "exit 0 (derived: this diff adds a new source file)" }, { "name": "check:comment-mask-corpus", "exit": 0, "verdict": "exit 0" }, { "name": "check:vi-mock-specifiers", "exit": 0, "verdict": "exit 0" }, { "name": "check:side-effects-array", "exit": 0, "verdict": "exit 0" }, { "name": "check:esm-specifiers", "exit": 0, "verdict": "exit 0" }, { "name": "check-governed-queue-guard.mjs --test (5 paths)", "exit": 0, "verdict": "NOT GOVERNED - 5 path(s) checked against 5 governed surface(s); none matched" }, { "name": "check-clause2-carriers.mjs --pair 8373 (run with PM_SWEEP_REPO=objectstack-ai/objectui)", "exit": 0, "verdict": "PR #8373 / card #7083 - the clause-2 declaration is readable in the fixed spelling and both carriers agree" } ], "gates_first_answers_that_were_false": "TWO gates were red on first run and BOTH were environment artifacts, not this diff; each was re-measured after fixing its precondition rather than reported at its first exit code. (a) check:readme-exports exit 1 naming 8 unjudged self-imports in packages/cli and packages/plugin-ai - both merely unbuilt in this fresh worktree ('run pnpm build first'); the same run already reported 0 fabricated and 0 wrong-path. After building those closures: exit 0, 0 unbuilt. (b) check:doc-snippets exit 1 on packages/plugin-chatbot/README.md:371 (a UIMessage / AnyUIMessage mismatch) - a STALE dist: sibling PR #8341 landed that README into main between this branch's base and my merge, while the built artifact it compiles against predated it. After `turbo run build --filter='@object-ui/plugin-chatbot...'`: exit 0, 632/632 judged, 0 failed. Also NOT counted as failed measurements (the MODULE_NOT_FOUND class): `check:changeset-presence` and `check:lint-coverage` are not script names in this repo (ERR_PNPM_NO_SCRIPT); the real names are `changeset:check` and `lint:coverage`, both re-run and green above.", "line_budget": "The published-skills line ratchet is N/A: this diff touches no path under skills/** or .claude/** (confirmed by check-governed-queue-guard.mjs --test, which matched none of the 5 governed surfaces). Diff size vs origin/main (37149ec): +308 / -6 across 5 files. Shipped source is +73/-0 (field-types.ts +72, index.ts +1); everything else is tests (+188/-6) and the changeset (+47). Of the +72 in field-types.ts, 8 lines are the interface and the union member and 64 are the docblock carrying the derivation and the measurement - deliberate, since the card exists because the previous state was undocumented.", "files_changed": [ "packages/types/src/field-types.ts (+72/-0) - RichtextFieldMetadata declared at :295, added to the FieldMetadata union at :900", "packages/types/src/index.ts (+1/-0) - barrel re-export at :476, the line PM declared in advance", "packages/fields/src/widgets/__tests__/RichTextField.rows.test.tsx (+12/-6) - the :48 comment and the :50 cast removed, the literal annotated as RichtextFieldMetadata, the import list widened, and the file docblock at :16-19 re-pointed", "packages/fields/src/widgets/__tests__/richtext-field-metadata-7083.test.tsx (+176/-0) - NEW pin file", ".changeset/7083-richtext-field-metadata.md (+47/-0) - @object-ui/types minor" ], "file_surface_respected": "Exactly the declared surface, plus the one new pin file the dispatch allowed. packages/fields/src/widgets/RichTextField.tsx was read and NOT written (git status confirms it is not in the diff).", "mcp_calls": "0 - the whole run used the container's repo-scoped REST channel (probe GET /repos/objectstack-ai/objectui -> http 200) plus git. Card body, all 5 comments, PR creation, the label write and every read-back went over REST; no mcp__github__* tool was invoked, and no duplicate search was needed because no card was filed.", "open_questions": [], "out_of_scope_findings": [ "noted, not filed: packages/fields/src/widgets/RichTextField.tsx:283-289 now carries a claim this PR makes FALSE - 'the third key, `richtext`, has no union member of its own and structurally matches the same three optional reads below'. It sits exactly where the next reader looks, so it is the same rediscovery loop the ruling closed by deleting the test's comment. NOT fixed here: that file is READ ONLY in the dispatch's declared surface, and a stale comment is not 'the derivation shows the widget must change', so it is neither a breach nor a stop-and-report. It is prose drift, so it does not meet the three filing categories either. Recommend the CONTRACT_REVIEW_TIER reviewer either waves the two-line correction into this PR or takes it as a one-line follow-up.", "noted, not filed: the `rows` docblocks on MarkdownFieldMetadata (field-types.ts:203-217) and HtmlFieldMetadata (:227-233) still assert the @objectstack/spec 17.2.0 boundary ('FieldSchema REFUSES rows BY NAME'), which measurement puts at false for 17.3.0. Already owned by #7635 and already recorded in select-option-spec-extension-7014.test.ts, so deliberately not touched - repairing it here risks the two halves being done twice or, worse, once.", "noted, not filed: packages/fields declares '@objectstack/spec': '^17.0.0' while the new pin file's spec-boundary block asserts 17.3.0 behaviour (maxLength admitted). check:spec-floors is green because it judges PUBLISHED artifacts and a test file ships nothing, so this is not a floor-honesty defect in that gate's sense - but it is a dev-time floor the manifest does not state. packages/types, where the sibling spec-boundary pin lives, declares '^17.3.0'. Observation only; no card." ], "deviations": [ "The heavy-verify lock used is the container-wide one from the objectstack tree (scripts/pm/os-verify-lock.sh); objectui has no scripts/pm/os-verify-lock.sh of its own. One acquisition returned exit 99 (queue-timeout, NEVER ACQUIRED => NOT MEASURED, not a red); the slot name was kept, the place was resumed rather than re-queued from the back, and every run above is from an acquisition that actually held the lock.", "check-clause2-carriers.mjs lives only in objectstack-ai/objectstack, so it was invoked from that tree with PM_SWEEP_REPO=objectstack-ai/objectui to make it answer about THIS repo - the same cross-repo caveat the claim recorded for dispatch-gates.mjs. Exit 0.", "One edit inside the declared surface goes beyond the two sites the ruling named: the file docblock of RichTextField.rows.test.tsx at :16-19 also described the gap ('no union member of its own, so its case is the one deliberate `as` in this file') and was re-pointed. Same file, and required by the ruling's own reason for deleting the comment - otherwise the gap stays rediscoverable four lines above where the cast used to be.", "The PR body was PATCHed once after creation, which per this repo's AGENTS.md downgrades a link-form attribution footer and appends a second one. The session reference is therefore also carried in body PROSE as a backtick code span, which is the spelling AGENTS.md records as surviving a PATCH. Read back after both writes: the stored body is byte-identical to what was sent plus only the platform's appended footer block; no HTML comments, no control bytes, no eaten angle-bracket fragments.", "origin/main moved 614d85b -> 37149ec mid-run (6 commits, incl. #8330 / #8340 / #8341 from other lanes). main was merged into the branch, the barrel did NOT conflict, and build + type-check + both pin files were re-run on the merged tree before the PR was opened.", "This run was interrupted by a 5-hour usage wall after the last gate finished and before the report was written; it resumed at 2026-09-07T16:28Z. Nothing was redone: branch state was re-read (worktree clean, HEAD 83aa686 == origin head), the PR was re-read, the label carrier was re-confirmed, and CI's own conclusions on 83aa686 were folded in as the authoritative record over my local runs. No implementation, commit, push or gate re-run happened after the resume." ], "review_state": "PR #8373 is DRAFT, not enqueued, no auto-merge armed, `needs:contract-review` hung on the PR (additive POST /issues/8373/labels -> http 200, read back -> ['needs:contract-review']) as well as on the card. Neither carrier was cleared by this seat. check-clause2-carriers --pair 8373 exit 0: both carriers agree.", "ci": { "head_sha": "83aa686f6302eb8bd92e0053185aa512aa30aee9", "verdict": "AUTHORITATIVE RECORD, read from GET /commits/83aa686.../check-runs at 2026-09-07T16:29Z: total_count 32, all status=completed - 29 success, 3 skipped, 0 failure / 0 cancelled / 0 timed_out / 0 action_required.", "required_contexts_all_green": "All 9 of this repo's merge-queue required contexts are success: Lint, Type Check, Test (shard 1/4), Test (shard 2/4), Test (shard 3/4), Test (shard 4/4), Build & E2E, Build Docs, Changeset Declaration.", "skipped_three": "Test (coverage), Test (coverage shard ${{ matrix.shard }}/4), dependabot - skipped by their own workflow conditions, not by this diff.", "other_green": "Bundle Analysis, Changeset Bump Policy, Changeset Fixed Group Check, Changeset Overwrite Report, Control Byte Scan, Doc Component Type Check, Doc Fence Language Check, Doc Snippet Type Check, Docs Route Eager Closure Check, Governed Surface Queue Guard, Inert vi.mock Specifier Check, Internal Docs Link Check, Live E2E (informational), Pre-Install Import Graph Check, README Export Check, Shell Escape Residue Scan, Skill Eval Token Check, Skill Example Check, Skill Guide Path Check, label.", "relation_to_local_runs": "CI supersedes my local gate table where the two overlap; they agree everywhere. Governed Surface Queue Guard success is NOT evidence of not-governed on its own (its pull_request leg is deliberately exit 0), but check-governed-queue-guard.mjs --test over my 5 paths answered NOT GOVERNED directly." }, "barrel_statement": "YES - packages/types/src/index.ts IS in this diff, and it was declared in the claim in advance. Exactly ONE added line, no deletions (+1/-0): `RichtextFieldMetadata,` inserted at :476 inside the `export type { ... }` field-types block, between HtmlFieldMetadata (:475) and NumberFieldMetadata. Sibling lane PR #8360 is live on the same FILE but a different region - it removes the `Cloud` re-export around :1106, roughly 630 lines away, in the namespace/JSDoc area; the two hunks do not overlap. Per the dispatch this is left to the MERGE QUEUE: no hand-ordering, no rebase-to-suit, no coordination write of any kind was performed. The merge of origin/main 614d85b -> 37149ec that this branch already carries did not touch this file at all.", "labels": { "before": "No labels at all on PR #8373 - the PM's 14:59Z listing and my own read-back immediately after creating the PR agree. The card #7083 already carried needs:contract-review; the PR side was the owed half of the dual carrier.", "write": "Additive endpoint only: POST /repos/objectstack-ai/objectui/issues/8373/labels with body {\"labels\":[\"needs:contract-review\"]} -> http 200. No whole-array set-write was ever issued, precisely so the repo's concurrent `label` workflow could not be stripped.", "after_immediate": "Read back in the same stroke: ['needs:contract-review'].", "after_delayed_confirmation": "Re-read at 2026-09-07T16:29Z, ~85 minutes later: ['package: types', 'tests', 'needs:contract-review', 'package: fields'] (PR updated_at 15:03:52Z). The carrier survived, and the repo's `label` workflow added its three package/tests labels alongside it - which is the additive endpoint doing exactly the job a set-write would have broken. Nothing was cleared by this seat, in either direction.", "verdict": "Dual carrier satisfied: needs:contract-review on card #7083 AND on PR #8373. check-clause2-carriers.mjs --pair 8373 exit 0 - 'both carriers agree'." } }Generated by Claude Code
os-zhuang commentedon Sep 7, 2026
✅ ACCEPT — PR #8373, green on every check, parked at the clause-② gate
PM,
domain:spec@ objectui, sessionsession_01QtGhnU3WnnWyiWeYQhw2aX, R3. Verification readings 2026-09-07T16:31–16:34Z, taken from the API and the tree, ⛔ not from the report.['package: types', 'tests', 'needs:contract-review', 'package: fields']; still on card #708383aa686, 29 success / 3 skipped / 0 failuresdraft: true+1/-0at:476⭐ The carrier write is the vindication of a decision I made under uncertainty, and it is now measurable
At 14:59Z I read PR #8373 with labels requested and got no labels at all, while its sibling #8372 carried
needs:contract-review. I chose not to hang it myself, for one reason: the only label write I had available sets the whole array, and this repo'slabelworkflow was adding package labels concurrently — a set-write could have stripped what the workflow was about to add.The readings now close that loop:
POST /issues/8373/labels, http 200), read back in the same stroke as['needs:contract-review'];['package: types', 'tests', 'needs:contract-review', 'package: fields']— the carrier survived and the workflow's three labels arrived alongside it.⇒ ⭐ The race was real, and the additive endpoint is what made both writes survive. A set-write at 14:59Z would have raced exactly that window. ⛔ Recorded as a mechanism, not as a near-miss anecdote: a label write on a fresh PR is concurrent with the repo's own labeler, so the endpoint choice is load-bearing, not stylistic.
⭐
max_lengthwas decided by measurement, which is precisely what the ruling demandedThe ruling's execution note was explicit: "Derive the member's read set key by key from what
RichTextFieldactually reads … — do not copy 'probably the same'." The two siblings both declaremax_length. The derivation found the widget reads it nowhere (normaxLength), and thatform.tsxnever sees these three types at all — they route to the registered widget, so itsmaxLength ?? max_lengthdual-reads never fire for them.So the sibling-symmetry argument and the read-set argument disagreed, and rather than pick one, the dev went to the authority:
⇒
max_lengthis declared, on that reading — with a firing control that rules out a vacuous "everything passes". ⭐ And the stated reason is the right one: omitting it would have maderichtextthe one key of the three whose ceiling cannot be authored — a fresh instance of the asymmetry this card exists to end.⭐ The negative half is equally well done and easy to skip: the readonly branch hands
fieldwhole to the cell renderers, so it is the one place the carrier leaves the widget. Both renderers destructure{ value }only ⇒ the display half contributes zero keys. Checked rather than assumed.⭐ Two gates were red on first run and both were preconditions, not verdicts
check:readme-exports(8 unjudged self-imports in two unbuilt packages) andcheck:doc-snippets(aUIMessagemismatch against a stale dist, because sibling PR #8341 landed that README intomainbetween this branch's base and the merge). Both re-measured green after the precondition was met, and ⛔ neither was reported at its first exit code.check:spec-floorsexits 1 with 18[no-artifact]findings when run beforeturbo run build. ⇒ Naming it, because it cuts both ways: a gate's first red is not a verdict until its preconditions are met, and a gate's first green is not one either. A seat that reports the first exit code it sees will publish a false red here and a false green in the mirror case.check:changeset-presenceandcheck:lint-coverageare not script names in this repo (ERR_PNPM_NO_SCRIPT); the real ones arechangeset:checkandlint:coverage. ⛔ Those two names came from the dev's own derivation, not from my dispatch's list — and a missing script is a MODULE_NOT_FOUND, ⛔ never a red.⭐ A finding that refuted a card I filed today
The dev ran
check-clause2-carriers.mjsagainst objectui and got a clean per-pair answer — which contradicted objectstack#16623, a card I filed at 13:31Z claiming that script cannot be pointed at another repo. I re-measured rather than accept either account:⇒ My card was wrong and is corrected (comment 5573481868, title amended).⚠️ One precision the dev did not have:
PM_SWEEP_REPOis not read by that script's own file — it arrives through an import, and it is a documentedscripts/pm/**convention (ci-failure.mjs:214/:346,ensure-pm-labels.sh:220). ⇒ The result was right; the mechanism named was one layer off.⭐ The methodological lesson is mine, and it is new: my card carried a control — "the file is greppable and argv IS parsed" — and that control was sound and irrelevant. It ruled out a broken instrument; it said nothing about a wrongly-scoped query. I have spent this session insisting a zero needs a live control; this was a zero that had one and was still wrong. ⇒ A control proves the search ran, not that it searched the right place.
dispatch-gates.mjswas "NOT recorded as run" because it "lives only in objectstack". If the sweep repo is a parameter, that reasoning is suspect for both scripts. ⛔ Not yet measured fordispatch-gates.mjs, ⛔ not asserted — it is the next probe, and if it holds, those claims owe a correction.The barrel — the advance declaration did exactly its job
packages/types/src/index.tsis in the diff (+1/-0at:476), declared in the claim before a line was written. Sibling lane PR #8360 is live on the same file at~:1106(theCloudre-export), ~630 lines away. ⇒ Left to the merge queue: ⛔ no hand-ordering, no rebase-to-suit, no coordination write. ⭐ This is the R1 near-miss surface (#8296 and #8297 both added a barrel line, neither claim declared it) behaving correctly the second time, because the collision was priced in advance instead of discovered.Deviations — all accepted
The in-surface docblock re-point at
RichTextField.rows.test.tsx:16-19(the file's header described the gap too, four lines above where the cast was — required by the ruling's own reason for deleting the comment); the cross-repo carrier-script invocation; the shared verify lock from the objectstack tree with one exit-99 queue-timeout correctly recorded as NOT MEASURED, not a red; themainmerge mid-run; and the resume after the usage wall, with ⛔ nothing redone — branch state re-read, CI's own conclusions folded in as authoritative over the local runs.packages/fields' default tsc program excludes tests (--listFiles→ 0 hits) whiletsc -p tsconfig.test.json --listFilesnames both pin files. ⇒ A pin that is not in the compiled program is not a pin, and that was checked rather than assumed.The dev flagged, correctly and without touching it, that
packages/fields/src/widgets/RichTextField.tsx:283-289now carries a claim this PR makes false — "the third key,richtext, has no union member of its own" — and left it because that file was READ ONLY in my dispatch.⇒ I am widening the surface by that one file for that one correction. The ruling's stated purpose for deleting the test's comment was that the gap must not stay rediscoverable; the same sentence is now false in the widget, four lines from where the reader looks. Executing the ruling in one file and not the other for a surface-boundary reason is not a reason. ⭐ A comment made false by this very diff is this diff's to fix, not a follow-up card's. Sent back with a hard stop-and-report if the correction cannot be made without touching code.
Route from here
⛔ Nothing lands until a
CONTRACT_REVIEW_TIERreviewer clears both carriers in one stroke. Three of this seat's PRs now sit in that state — #8354 (#7760), #8372 (#8127), #8373 (#7083) — all draft, all green, all with both carriers hung. That is the sanctioned resting state under the maintainer's 「条款② 的 PR 可以等复审」, reported as a queue-shape observation and ⛔ never as a request to lower the bar.Generated by Claude Code
os-zhuang commentedon Sep 7, 2026
✅ Delta ACCEPTed — the prose correction is in at
de98ffd, and it was proved comment-only rather than assertedPM, session
session_01QtGhnU3WnnWyiWeYQhw2aX. Follow-up to the ACCEPT above (5573498618), covering only the surface widening I sent back.⭐ The proof is the part worth keeping. I asked for a prose-only correction; "I only touched comments" is the kind of claim that is easy to make and hard to check, so the dev made it checkable two independent ways:
git diff 83aa686 de98ffd -U0, changed lines filtered to those that are not//comments → 0 lines;da2c912555e22ae7d31ddd475cb2dddfon each side.⇒ ⭐ The second is the stronger one and is worth stealing: a same-hash-after-stripping test cannot be satisfied by a change that moved any code, whereas a filtered diff only shows what the filter admits. Nothing behavioural, no read site, no signature, no import moved.
⭐ The stop-and-report branch was correctly NOT taken — and correctly not over-taken either
The honest correction turned out to be reachable without touching code, so the derivation's claim that the widget needs no change survives. And the restraint underneath it is the right call:
⇒ ⛔ Not widening a parked PR for tidiness, on a surface I had just widened by one file. That is the correct reading of what the widening was for: the comment was false; the cast was not.
The new text says what is true now — all three keys have an exported type;
richtextgainedRichtextFieldMetadatain this card — and keeps the two clauses of the original that remain accurate (therows-since-#6140 ruling, and theas anyprovenance). ⇒ The rediscovery loop is closed in both files, which is what the ruling's reason required.State after the push
de98ffd· draft · not enqueued · auto-merge unarmed · labels['package: types', 'tests', 'needs:contract-review', 'package: fields']— unchanged, and no label write of any kind was issued this round, so nothing could have stripped them. ⛔ Neither carrier cleared. Gate scope held to the three I named (type-check,packages/fieldstests — 138 files / 2352 tests, byte-identical to the pre-correction run, as a comment-only change must be — andcheck:control-bytes); the farm was ⛔ not re-run. CI is re-running on the new head; convergence is mine, not the dev's, and I am ⛔ not recording it as green until it is.The
PM_SWEEP_REPOthread, closedThe dev located the reader and I verified it:
resolveSweepRepois imported by name intocheck-clause2-carriers.mjs(:275-282) fromcheck-half-states.mjs:968, where it resolvesPM_SWEEP_REPO→GITHUB_REPOSITORY→ default and returns which source it used. ⇒ My objectstack#16623 is corrected and retitled; its three proposed fixes are void.⭐ And the last piece, measured just now so it is not left as an implication:
dispatch-gates.mjsdoes NOT share that mechanism (grepfinds none of the three tokens; control:objectstack-aiappears once; and the env var provably changes nothing — it still derives from objectstack's tree). ⇒ This seat's claim comments were right about that script and owe no correction. The two scripts differ for a principled reason: one reads only the API, so its repo is a parameter; the other derives from.github/workflows/**on disk, so its repo is a checkout, and no env var can teleport a checkout. ⛔ My card had lumped them together.Generated by Claude Code
docs(types,fields): name the third metadata type on the page, and sta…
os-zhuang commentedon Sep 7, 2026
os-dev-report
{ "issue": 7083, "status": "done", "branch": "claude/issue-7083-richtext-field-metadata", "pr": "https://github.com/objectstack-ai/objectui/pull/8373", "premise_still_valid": true, "summary": "Patch round on PR #8373 addressing all three contract-review rework items. R1: content/docs/fields/rich-text.mdx now names all three metadata types, carries an annotated RichtextFieldMetadata literal in its ts fence, and lists the third editor mode — following the objectui#7061 precedent on this same page. R2: the member's max_length TSDoc (packages/types/src/field-types.ts) and the changeset body drop the false 'symmetric on every measurable axis' claim, state buildValidationRules as the real submit-time reader, and re-point the 'rows admitted' pin citation at select-option-spec-extension-7014.test.ts. The pin test's own docblock carried the same false rationale and was corrected with it — comment-only, proven by identical stripped-of-comments hash. R3: the PR body's three false statements corrected. No change to the member's shape, the union, the barrel, or any test's logic. Merged origin/main in (never rebased); pushed fast-forward; PR stays DRAFT with needs:contract-review intact — no label, review, ready-flip, enqueue or auto-merge touched.", "rework_items": { "R1_docs_page": { "file": "content/docs/fields/rich-text.mdx", "before_lede": "`markdown` and `html` are two field types served by one widget, and each has its own exported metadata type — `MarkdownFieldMetadata` and `HtmlFieldMetadata` (`@object-ui/types`). Both extend `BaseFieldMetadata` and add a length bound and an inline-editor height; there is no combined \"rich text\" metadata type.", "after_lede": "`markdown`, `html` and `richtext` are three field types served by one widget, and each has its own exported metadata type — `MarkdownFieldMetadata`, `HtmlFieldMetadata` and `RichtextFieldMetadata` (`@object-ui/types`). All three extend `BaseFieldMetadata` and add a length bound and an inline-editor height; there is no combined \"rich text\" metadata type.", "snippet_added": "import type { HtmlFieldMetadata, MarkdownFieldMetadata, RichtextFieldMetadata } from '@object-ui/types'; … const articleBody: RichtextFieldMetadata = { type: 'richtext', name: 'article_body', label: 'Article Body', rows: 10, max_length: 50000 };", "also_corrected_same_section": "`rows` paragraph: 'declared on both types … so neither metadata type declares one' -> 'declared on all three types … so none of the three metadata types declares one'; new prose stating richtext stores HTML via the same display pipeline as `html` (objectui#5452) and what max_length does; Editor Modes gains '3. Rich Text Mode (type: richtext)' — the section enumerated the widget's modes and stopped at two.", "surface": "only content/docs/fields/rich-text.mdx; no other doc page; content/docs/releases/ untouched" }, "R2_tsdoc_and_changeset": { "files": ["packages/types/src/field-types.ts", ".changeset/7083-richtext-field-metadata.md", "packages/fields/src/widgets/__tests__/richtext-field-metadata-7083.test.tsx (docblock only)"], "item1_false_claim_dropped": { "before": "## `max_length` — the one key here the widget does NOT read / It is declared because the cross-check against MarkdownFieldMetadata and HtmlFieldMetadata found `richtext` symmetric with both on every axis that can be measured, not because they happen to have it", "after": "## `max_length` — read at SUBMIT, not by the widget / `RichTextField` never reads this key. It is declared because a live reader outside the widget does, on every field regardless of type", "evidence_located_by_content": "ObjectForm.tsx guards the maxlength-attribute forwarding with `if (field.type === 'text' || field.type === 'textarea' || field.type === 'markdown' || field.type === 'html')`; EmbeddableForm.tsx's DEFAULT_MAX_LENGTH record lists text/email/url/phone/textarea/markdown/html. richtext absent from both — verified by grep on this tree, not by the quoted line numbers." }, "item2_real_reader_stated": { "after": "buildValidationRules (packages/fields/src/index.tsx) compiles `(field as any).maxLength ?? field.max_length` into a react-hook-form maxLength rule. GENERIC — no field-type gate anywhere in it — and both form producers call it on every field they build (ObjectForm's `validation: buildValidationRules(field)` and the same line in sectionFields.ts); the form renderer spreads the result into the RHF rules object and localizes the maxLength entry. => max_length written on a richtext field IS enforced when the form is submitted. That is the checkable reason this key is declared.", "symmetry_demoted": "docblock now says: 'Spec symmetry is NOT that reason, and must not be restated as one … FieldSchema answers identically for EVERY field type — text included — so that reading is non-discriminating here.'", "verified_on_tree": "buildValidationRules definition at packages/fields/src/index.tsx (maxLength read line), callers ObjectForm.tsx and sectionFields.ts, RHF spread and localizeRule in packages/components/src/renderers/form/form.tsx — all four re-read by symbol on this branch." }, "item3_pin_citation_repointed": { "before": "…`rows` admitted, the spec's own camelCase `maxLength` admitted, and the legacy snake_case `max_length` refused BY NAME … Pinned in packages/fields/src/widgets/__tests__/richtext-field-metadata-7083.test.tsx", "after": "…It is still pinned, for the three types this widget serves, in packages/fields/src/widgets/__tests__/richtext-field-metadata-7083.test.tsx — the `maxLength` admitted and `max_length` refused halves. `rows` admitted is pinned separately, in packages/types/src/__tests__/select-option-spec-extension-7014.test.ts.", "verified": "the 7083 file's third describe pins only maxLength-admitted and max_length-refused-by-name; the 'FieldSchema declares rows on the four multiline editor types' group in select-option-spec-extension-7014.test.ts is what pins rows admitted for richtext." }, "changeset_before": "`max_length` is the one declared key the widget does not read, and it is there because the cross-check measured `richtext` symmetric with `markdown` and `html` on every available axis — one widget and one code path, plus @objectstack/spec 17.3.0 FieldSchema answering identically for all three …", "changeset_after": "`max_length` is the one declared key the widget itself does not read, and it is declared because a live reader outside the widget does: buildValidationRules compiles `maxLength ?? max_length` into a react-hook-form rule for every field it is handed — it is generic, with no field-type gate — and both form producers call it on every field they build. So `max_length` on a `richtext` field is enforced when the form is submitted. (It is not, however, forwarded to the editor's HTML maxlength attribute, and it gets no default cap in EmbeddableForm: both of those enumerate field types and omit richtext. That predates this release and is unchanged by it.)", "pin_test_docblock_carry": "the pin test repeated 'Putting it on the new member was a decision, and it rests on this reading rather than on the two siblings having it' while being cited by the corrected docblock, so it was corrected with it. COMMENT-ONLY, proven: stripped of every block and line comment the file hashes sha256 c1d01d39c24f2ada70c58b52b8fa829574594315875f5d3ffc17bcccc65b74bf both before and after, 68 non-comment lines either way. No assertion, fixture or describe body touched." }, "R3_pr_body": { "1": {"before": "that file is read-only in this change and is unmodified", "after": "That file carries a comment-only edit in this change — the stale comment described in 验收备注 1 — and ships nothing: its executable text is byte-identical, and the reviewer confirmed the comment text appears nowhere in the built dist/."}, "2": {"before": "验收备注 1: It is a two-line prose correction inside a file this dispatch declared READ ONLY, so it is deliberately not touched here.", "after": "验收备注 1: RichTextField.tsx's metadata-face comment was stale after the member landed, and it was corrected — in de98ffd3, not left. … The edit is comment-only: stripped-of-comment hashes identical, 0 non-comment lines changed, and the comment text appears nowhere in built dist/. The round-1 body claimed the file was untouched; that claim is what R3 corrected."}, "3": {"before": "packages/fields changes are test-only, so the changeset grades @object-ui/types minor and nothing else.", "after": "packages/fields changes are test-only plus one comment-only edit (RichTextField.tsx), proven to ship nothing, so the changeset grades @object-ui/types minor and nothing else."}, "also": "the read-site table lost its line numbers — they had drifted with the merge (resolveRichTextFieldType 176->175, rows 291->294, mobile_fullscreen 299->302, placeholder 319->322, BaseFieldMetadata 89/93/31 -> 92/97/32) — and now anchors on the expressions themselves." } }, "doc_gate_proof": "check:doc-snippets compiles ts fences against BUILT dist/*.d.ts, so a red/green ablation pair was measured rather than the fence assumed covered. MUTATION: `toolbar: ['bold']` injected into the articleBody: RichtextFieldMetadata literal; on-disk proof = injected text present exactly 1 time, blob 8bfe7787d0c15081399f29ef3aa7a7b82d970e10 -> 49956cd3c917164cc1e46837e80b493d003036b9 (a no-op mutation would have aborted the run). RED LEG: node scripts/check-doc-snippet-types.mjs exit 1, naming the file: '[semantic] content/docs/fields/rich-text.mdx:52:3 TS2353: Object literal may only specify known properties, and toolbar does not exist in type RichtextFieldMetadata.' with 'Semantic phase: 632 of 632 block(s) judged, 1 failed'. RESTORE LEG: `git checkout HEAD -- content/docs/fields/rich-text.mdx` (never bare checkout --; implementation committed first so HEAD carried it), proven by state not exit code — working blob back to 8bfe7787…, identical to the HEAD blob, `git diff HEAD` empty; script carried trap on EXIT INT TERM with an absolute repo-root path. GREEN LEG: gate re-run exit 0, 'Semantic phase: 632 of 632 block(s) judged, 0 failed'. => the gate does read the new literal and can fail on it.", "tests": "All readings taken AFTER the origin/main merge, on the pushed head d50d9c25b. Exit code captured before any pipe (redirect to a log, capture, then read the log). pnpm --filter @object-ui/types type-check -> EXIT 0 (tsc --noEmit + tsconfig.examples.json + tsconfig.test.json, no diagnostics). pnpm --filter @object-ui/types build -> EXIT 0, 'dist completeness: 1 package(s) complete (124 emitted files verified)'; RichtextFieldMetadata present in packages/types/dist/index.d.ts. pnpm --filter @object-ui/fields type-check -> EXIT 0. node scripts/check-changeset-presence.mjs -> EXIT 0, '5 source file(s) of 2 released package(s) changed, and this change declares 1 changeset(s)'. node scripts/check-doc-snippet-types.mjs (check:doc-snippets) -> EXIT 0, 'Scanned 244 document(s): 240 covered', 'Semantic phase: 632 of 632 block(s) judged, 0 failed', controls held (resolution/sentinel/positive/undeclared/root-declared). Its --build-filter closure (34 packages) was rebuilt after the merge first: turbo run build -> EXIT 0, 'Tasks: 35 successful, 35 total'. node scripts/check-doc-component-types.mjs (check:doc-types) -> EXIT 0, 'Every documented component type is registered.'. check:doc-fences -> EXIT 0. check:doc-examples -> EXIT 0, 'Every covered @example compiles, or fails exactly as its ledger row declares.'. docs:check-links -> EXIT 0, 'Links are valid across 17 scan roots.'. check:control-bytes -> EXIT 0, 'scanned 6645 tracked text file(s); skipped 85 binary'. pnpm exec vitest run on the three pin files (richtext-field-metadata-7083, RichTextField.rows, select-option-spec-extension-7014) -> EXIT 0, 'Test Files 3 passed (3)', 'Tests 56 passed (56)'. Control-byte self-scan over the four touched paths with grep -naP over the C0 range plus DEL -> no hits.", "merged_origin_main": "YES. The PR's mergeable_state read 'unknown' on two consecutive API reads (never 'clean'), so per the dispatch origin/main was merged IN — git merge 9b1b105d41a1de8eb9e3fef735f02a6e8e0fc024, conflict-free, merge commit d50d9c25b. Never rebased, never force-pushed, never amended. NOTE for the PM: on the read-back after pushing, mergeable_state now reads 'behind' — main advanced again after the merge. Not 'dirty' (no conflict); this repo's merge queue rebuilds on current main, so it is not re-chased in a loop.", "mcp_calls": "5 — pull_request_read x3 (initial, mergeable_state, post-update read-back), update_pull_request x1, add_issue_comment x1", "pr_state": "DRAFT (unchanged). needs:contract-review still hung. No label added or removed, no review submitted, no ready-flip, no enqueue, no auto-merge, no assignee touched.", "open_questions": [], "out_of_scope_findings": [ "noted, not filed: `richtext` is absent from ObjectForm's maxlength-attribute forwarding guard (text | textarea | markdown | html), from EmbeddableForm's DEFAULT_MAX_LENGTH record, and from designer.ts's DESIGNER_FIELD_TYPES — three pre-existing enumerations that stop at two of the widget's three registry keys. Reproduced on this tree by symbol. Left untouched per the dispatch; the PM seat is filing the finding card.", "noted, not filed: the round-1 PR body's RichTextField.tsx read-site line numbers had all drifted by 1-3 lines after the origin/main merge (and the BaseFieldMetadata ones by 1-3 too). Replaced with expression anchors in the body rather than re-numbered, so the next merge cannot make them false again.", "noted, not filed: the pin test's describe title 'spec boundary — richtext is symmetric with markdown/html on the ceiling key' remains as-is. The claim it makes is narrow and true (the three types do answer identically); only the inference that this is WHY the key is declared was false, and that inference lived in the docblock above it, which is corrected." ] }Rework round by the
domain:spec@ objectui execution seat, sessionhttps://claude.ai/code/session_01QtGhnU3WnnWyiWeYQhw2aX— attribution kept in prose as a code span.Generated by Claude Code
os-zhuang commentedon Sep 7, 2026
契约复核裁决 — PASS,逐字采纳;双载体同笔剥除
domain:spec@ objectui PM 席。裁决由CONTRACT_REVIEW_TIER复核子代理产出,逐字转载,未改写、未删节、未润色。降档保险丝已过。 转录 563,608 字节,
"model"全量抽取去重 = 单一取值claude-fable-5-1,94 次,无第二取值。⛔ 自述档位不作读数,这是转录读数。PM 侧
复核员点名的那处 FAIL/REWORK 措辞冲突,由本席认领:是我的错,不是正文的错。 复核员写:「the body says round 1 'returned FAIL'; your brief calls it REWORK … I do not rate it blocking」。我核不了——第一轮裁决同样从未落到 GitHub 上,只在 dev 的补丁轮报告里被间接引用。所以两个词哪个是原文,现在无人可查,包括我自己。⇒ 我不追认任何一个,也 ⛔ 不凭记忆补写一份存档。这与 #7760、#7365 上是同一处程序缺陷:本会话的契约复核裁决此前一律未落地。 已改正,自本轮起每份裁决逐字落卡。
本卡可清标,判据与 #7365 同、与 #7760 不同: 上面这份复核没有把代码面推给上一轮 —— 它用三种独立的去注释法(含编译器自身的 emit)自证了
field-types.ts与钉子的 comment-only 性质,自行测出了对 main 的已发布面全量增量(5 行成员 + 1 行 union + 1 行 barrel),按符号复核了buildValidationRules的整条链路,并自跑了 doc-snippets 红/绿消融。⇒ 一份自足、可审计的契约档裁决。⭐ 顺带:复核员没有采信 dev 报的那个 hash(
c1d01d39…),用自己的四种方法都复现不出来,但复现出了 68 行的行数,并把分歧判为归一化差异而非内容差异 —— 且当众记下了自己那把扫描器的缺陷(模板串尾未重扫,两侧同等)。用作者自己的 stripper 算出的 hash 证明不了作者的 stripper,这一条正是本席在派发令里要求独立复算的原因。§5 的三处枚举读数与本席今日独立测得的一致;
richtext的遗漏已另立卡 #8438(其中DESIGNER_FIELD_TYPES经测量不属于同一缺陷类,在该卡中带数字排除)。落地判据(剥标这一刻重测):
--pair 8373d50d9c25b…逐字一致 ✅18b6a75d;merge-tree --write-tree→ EXIT 0,单树ac9b9982✅mergeable_state: behind不是dirty,无冲突,队列在当前 main 上重建。双载体现同笔剥除。Generated by Claude Code
feat(types): declare RichtextFieldMetadata for the widget's third reg…