Skip to content

A screen flow's select options and screen description have no key in the flows translation face — the zh-CN intake wizard translates its labels and placeholders and keeps its options and body text in English (17.7.0) #22507

Closed
#22626
Closed
A screen flow's select options and screen description have no key in the flows translation face — the zh-CN intake wizard translates its labels and placeholders and keeps its options and body text in English (17.7.0)#22507
#22626
Assignees
zhuangjianguo
Labels
area:i18nThe customer's own language, across UI, metadata and notificationsbugSomething isn't workingdomain:specpriority:p2Medium: important, M3

Description

@objectstack-fleet
objectstack-fleet
bot
opened on Oct 9, 2026

Ruled: A — the engine picks the translated description and title templates before the per-run render; objectui stops overlaying title (batch #308 item 1, maintainer 「22507 同意」, ruling 6093472708, 2026-10-10T03:51Z). Label needs-user-decision → pm:queue in the same act.

Filing class: ① product defect (user-visible) — reach: browser, zh-CN Console, measured once on 17.7.0.

Reader: objectstack triage → the spec lane (TranslationDataSchema.flows) or, if the key exists and the runner ignores it, repo:objectui (the flow runner).

Symptom

Measured in objectstack-ai/hotclm on 17.7.0 (objectstack-ai/hotclm#82, PR objectstack-ai/hotclm#85, Platform findings item 2). The app's contract_intake screen flow carries a full zh-CN bundle under flows.contract_intake.screens.<node>.{title, fields.<field>.{label, placeholder}}, and those render in Chinese. In the same wizard:

  • select-field options render in English (Head office, USD — US Dollar) — the face has no key for a screen field's options;
  • the screen's body text (description) renders in English — the schema's own guidance places description outside the face.

So a fully translated app shows a half-English wizard, and no bundle the author can write fixes it.

Expected

Either the face addresses screen-field options (and, if intended, screen description), or the runner resolves options that mirror an object field through that object's translated option labels. Which of the two is the producer is triage's call; the second half (description) may be by design — if so, the guidance is the record and only the options half stands.

Dedupe

MCP search_issues on this repo, query flow screen translation select options description not translatable flows translation face screen field options → 6 hits, all closed and none on options: #7646 (bundle had no screen-flow surface), #11287 (nothing read flows), #17306 (screen field numeric bound / help text), #20318 (launcher header key). Same query on objectstack-ai/objectui → nearest #11220 (closed — hard-coded Select… placeholder).

Filed by the repo:hotclm PM seat from a measured dev finding.

Activity

  1. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    Triage: first grade, bug · priority:p2 · domain:spec · area:i18n, pm:blocked on #22450. It is the closing pass for "user-read flow text with no translation key", with an enumeration pin

    Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-09T16:55Z. ⛔ Not a claim, ⛔ not a dispatch.

    Triage: the face is TranslationDataSchema.flows (packages/spec/src/system/translation.zod.ts), with translateFlow and the i18n extractor. That puts it in domain:spec. objectui's runner follows the face's keys.

  2. objectstack-fleet
    added
    area:i18nThe customer's own language, across UI, metadata and notifications
    bugSomething isn't working
    domain:spec
    pm:blocked
    priority:p2Medium: important, M3
    on Oct 9, 2026

  3. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    Unlock: pm:blocked → pm:queue. #22450 closed (PR #22525 merged)

    Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-09T20:52Z. Unlock scan. ⛔ Not a claim, ⛔ not a dispatch.

    Thread-read: 6085404571

  4. objectstack-fleet
    added
    pm:queue
    and removed
    pm:blocked
    on Oct 9, 2026

  5. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    Serial note · domain:spec seat 2 (#18549) · os-sales · session session_01DhTqaEHqPVSVnAkjG3jywn · 2026-10-09T21:36Z. ⛔ Not a claim; the card stays pm:queue.

    Not dispatched this round: it waits for PR #22534 (#22508, domain:spec seat 1) to land. That PR edits packages/spec/src/system/translation.zod.ts, i18n-resolver.ts and its test, packages/spec/liveness/translation.json and the translation reference pages. Those are the same files this card's closing pass enumerates and keys, so it is serial, not folded.

    Generated by Claude Code

  6. objectstack-fleet
    added
    pm:dispatched
    and removed
    pm:queue
    on Oct 9, 2026
  7. objectstack-fleet
    assigned
    os-sales
    on Oct 9, 2026

  8. objectstack-fleet commented on Oct 9, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 9, 2026
    ContributorAuthor

    Claim: PM loop round 1 (the closing pass for user-read flow text with no translation key, per triage 6085404571 and the unlock 6089070768) · 2026-10-09T23:11Z
    Session: session_01DhTqaEHqPVSVnAkjG3jywn
    Account: os-sales (the seat's linked user as GET /user answers it; the card's assignee from this act)
    Branch: claude/issue-22507-flow-user-read-slots
    Worktree: objectstack-issue-22507
    Domain: domain:spec
    Seat: domain:spec#2 (seat post #18549)
    File surface (at origin/main 0e00d3611 or later, after PR #22534; stop on breach and explain in the report):

    Generated by Claude Code

  9. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    os-dev-report

    {
    "issue": 22507,
    "status": "needs_decision",
    "branch": "claude/issue-22507-flow-user-read-slots",
    "pr": "#22555",
    "session": "session_01DhTqaEHqPVSVnAkjG3jywn (the claim 6090837460 names the same session; this run is its subagent)",
    "premise_still_valid": true,
    "summary": "Two of the three unkeyed user-read slots are keyed, and the family pin is added. The flows face gains flows.FLOW.successMessage / .errorMessage (the terminal toasts, overlaid only where the flow authors one) and flows.FLOW.screens.NODE_ID.fields.FIELD.options.VALUE (option labels keyed by String(value) through one new function, flowScreenFieldOptionKey). ScreenFieldConfigSchema now refuses two options of one field that collide as text, located at the second option's value with the remedy; measured reach is 0. translateFlow applies both families, and a new resolveFlowScreenFieldOptions serves a runner holding a ScreenFieldSpec. FLOW_SCREEN_FIELD_NO_OPTIONS and the toast guidance are gone, and the description guidance is rewritten. Lint judges both families and the CLI extractor and coverage scaffold and demand them. The ledger rows for the toasts are planned (runner downstream), and the generated artifacts are regenerated. STOPPED on the third slot, as the dispatch ordered: a screen's description is server-rendered (service-automation's screen executor renders config.description through renderTextSlot with double-brace holes and serves ScreenSpec.description), so it is NOT keyed and the pin records it as owed. The seat's route decision is open_questions[0]. For that reason the PR's first line is Part of #22507, not the dispatch's closing keyword (conflict noted in open_questions[2]). The pin also records approval decisionOutputs[].label as owed (a user-read slot with no face; out_of_scope_findings[1]). File-surface additions beyond the claim, each mechanical: packages/spec/src/automation/builtin-node-config.zod.ts (home of the collision refusal the claim orders), the new leaf packages/spec/src/automation/flow-screen-option-key.ts and its barrel line, the generated content/docs/references/automation/builtin-node-config.mdx, and three example bundles (app-crm zh-CN toasts; app-todo zh-CN and ja-JP option labels) translated so check:i18n-coverage holds at its baseline. Without them it measured 101 to 103 and 106 to 126. For the seat's objectui card, the runner must read three keys. Options: FlowRunner.tsx localizeScreen sets each field's options to resolveFlowScreenFieldOptions(bundle, flowName, screen.nodeId, field, { locale }) before ScreenView draws its SelectItems; not via FLOW_SCREEN_FIELD_COPY_KEYS. successMessage: FlowRunner.tsx's terminal-success toast, via translateFlow({ name, successMessage: outcome.successMessage }). errorMessage: utils/flowResponse.ts flowFailureMessage, read by the runner's resume and by both launch hosts through judgeFlowLaunch; the launch path has to be given the bundle and flow name. The runner's header notes about description, inlineHelpText and successMessage are stale.",
    "tests": "All readings at head c17df03 (the merge of origin/main 4638625) unless named. TYPECHECK under the verify lock, pnpm --filter PKG run typecheck exit 0 for @objectstack/spec (source, scripts and test layer: 52 files / 246 errors / 135 pinned signatures held), @objectstack/lint and @objectstack/cli. cli first reported TS2305 on @objectstack/core exports, a stale core dist after the merge; it went green after rebuilding the cli closure. SPEC TIERS in full under the lock: local 637 files / 18977 passed, 1 todo; repo 54 files / 915 passed. LINT full suite 133 files / 6096 passed; validate-translation-references.test.ts alone 162 passed. CLI unit tier, every test/i18n-* file it holds: 24 files / 285 passed, including the new i18n-flow-toast-option-coverage.test.ts (8 tests). The integration tier is declared to CI. NEW PINS: flow-screen-option-key.test.ts (the key and the collision refusal, incl. the FlowSchema door at nodes[1].config.fields[0].options[1].value); i18n-resolver.test.ts +8 (toasts, options, identity, locale chain, own-key read); translation.test.ts (face, aliases, description guidance); flows-translation-face.test.ts (the family pin, 7 tests). ABLATION of the pin, through scripts/ablation-replace.mjs under the lock, predicted red, observed red. Leg A plants subtitle: z.string().optional() on ScreenConfigSchema (anchor x1 to x0, blob 5b4cf3fe to ac16dd5a): 1 failed / 6 passed, expected [ 'screen.subtitle' ] to deeply equal []; restored blob == HEAD 5b4cf3fe, git diff HEAD empty. Leg B renames the face's flows successMessage key (blob 8d5dd093 to 05480fb1): 2 failed / 5 passed (the keyed-row check and the face-fill check); restored blob == HEAD 8d5dd093, git diff HEAD empty. No dist step: the pin imports src. GENERATED: pnpm --filter @objectstack/spec check:generated (dist built by that run, post-merge) reports all 15 up to date; dropped refinements unchanged at 688 / 226; check:liveness reports translation 29 classified (live 27, planned 2). GATES: dispatch-gates --commands derived 115 at c17df03 and all 115 were run with exit codes recorded. --ran reports 115 run, 0 NOT-MEASURED, a DERIVED zero. 114 exited 0. check:skill-examples and check:dual-build-cjs-loads first refused (exit 3, PREREQUISITE NOT MET) and exited 0 after building the named packages. check:platform-checklist exits 1 with the identical two lines on a clean worktree of origin/main 4638625, not this diff (out_of_scope_findings[0]); per checklist-status.yml it is kept out of per-PR CI. ESLINT narrowed: the 17 changed .ts files under eslint --no-inline-config --format json give 17 results, 0 errors, 0 warnings (an ignored file would be a warning). Invariance: eslint.config.mjs enables no type-aware linting (no parserOptions.project), so the diff moves no verdict on an untouched file. check:i18n-coverage OK (13 configs, 621 baselined, none new); check:partof-closing-keyword exit 0 on the PR body. CI on c17df03 at report time: 12 success (Check Changeset, Governed Surface Queue Guard, Spec property liveness, Part-of PR must not also close its card, the two claim guards, Check PR Size, docs-link and docs-affected checks, Auto Label, filter), 2 skipped (Console Pin Gate, packed-tarball smoke), 18 in_progress (Lint & Repo Gates, Type Check x4, Test Core x6, Dogfood Regression Gate x3, Dogfood Verify CLI, Build Core, Build Docs, Temporal Conformance), 0 failure. Not waited on, per the dispatch contract.",
    "mcp_calls": "2: mcp__claude-code-remote__add_repo for objectstack-ai/hotcrm and objectstack-ai/hotclm (read; both answered read_available and attached nothing). No MCP GitHub tool was called.",
    "api_writes": "3 relay strokes as objectstack-fleet[bot], each one POST /repos/objectstack-ai/objectstack/dispatches executed by the fleet-write run: (1) pr_create, POST /repos/objectstack-ai/objectstack/pulls (draft) giving PR #22555, run 38009780627, read back 12021 bytes sent = stored; (2) label-write --assign os-sales, POST /repos//issues/22555/assignees, run 38009828010, read back matches; (3) this os-dev-report comment, POST /repos//issues/22507/comments. Zero label writes: the dispatch named none and skip-changeset does not apply. git push is not a REST write (9 pushes of the branch). Everything else was a read (gh api GETs).",
    "open_questions": [
    {
    "question": "Which reader picks a screen's translated description? It is server-rendered (renderTextSlot in service-automation's screen executor, double-brace holes over flow variables), and templated descriptions are real: app-todo success_screen, and hotclm contract_intake's 'First version' screen ('{templateNote}'). The dispatch's four-axis framework was not carried, so these options state cost and recommendation without inventing axes; send the framework if the decision needs it.",
    "options": [
    "A. Engine route, as refusals: the screen executor picks the translated TEMPLATE through the i18n service in AutomationContext.locale before renderTextSlot, for description AND title, with the face keys screens.NODE_ID.description (text-slot judged like refusals.message). Cost: a service-automation change (cross-lane), the key ships with its reader, and objectui's localizeScreen must stop overlaying title the same day or it would replace the engine-rendered heading with a literal template.",
    "B. Client overlay with a schema refusal of any double-brace hole in a translated description. Cost: spec-only plus the objectui card; a templated description stays untranslatable, and the refusal says so.",
    "C. Client overlay with title's silent boundary (a translated hole draws literally). Cost: the smallest change, and a known trap shipped on a slot whose purpose is per-record text."
    ],
    "recommendation": "A, because the slot's job is per-record text, A is the route the family already took for the one other server-rendered slot (refusals), and it also clears title's recorded literal-hole boundary; B is the fallback if a service-automation change is not wanted now."
    },
    {
    "question": "The mandated Clause-②: yes (widening, with one narrowing: the option-value collision refusal) is read by scripts/pm/clause2-line.mjs as arm widening (the first token inside the parenthesis), so check:adr-0087-registration treats the changeset as non-breaking. clause2-line.mjs documents yes (narrowing) as the spelling for a diff that widens one surface and narrows another, and AGENTS.md's post-task checklist reads (narrowing) as BREAKING. I used the mandated spelling verbatim in the PR body and the changeset.",
    "options": [
    "A. Keep it: the narrowing's measured reach is 0 (35 in-repo flows; hotcrm and hotclm by inspection).",
    "B. Switch to Clause-②: yes (narrowing) and add the one ADR-0087 disposition marker check:adr-0087-registration then demands (nothing is removed or renamed; the refusal text carries the remedy)."
    ],
    "recommendation": "B, because the arm exists so a narrowing is machine-visible whatever its reach, and the accept set of ScreenFieldConfigSchema does narrow. The seat owns the spelling and the body edit."
    },
    {
    "question": "The dispatch mandated Fixes #22507 as the body's first line. With description unkeyed and awaiting the decision above, merging would close the card on an open half, so the body opens Part of #22507 (os-dev rule: use Part of when merging must not close the card). If the seat rules description out of this card (for example by filing it as its own card), the line can flip to the closing keyword.",
    "options": [
    "A. Keep Part of until the description route lands in this PR or a follow-up",
    "B. Seat files description separately and edits the first line to the closing keyword"
    ],
    "recommendation": "A until the decision is recorded."
    }
    ],
    "out_of_scope_findings": [
    "class: a · reach: the repo's own gate, measured: pnpm check:platform-checklist exits 1 on a clean worktree of origin/main 4638625 (and main has no later commit touching either file at b53b949) · evidence: docs/qa/platform-checklist/areas/attachments-storage.json cites packages/services/service-storage/src/attachment-access-hooks.ts#canEdit, which the file no longer declares, giving 'ABSENT SYMBOL' and 'SYMBOL ANCHORS LOST — 27 anchor(s) resolve here but the floor is 28'. The likely source is the attachment-gate changes ce3d0ad / f782f17. The gate is out of per-PR CI by checklist-status.yml, so only the watchdog sees it · dedupe words: platform-checklist attachments-storage canEdit, symbol anchor floor 28, attachment-access-hooks canEdit",
    "class: a · reach: named producer examples/app-showcase/src/automation/flows/dynamic-approval.flow.ts (declares approval decisionOutputs with labels); not reproduced in a browser · evidence: ApprovalNodeConfigSchema.decisionOutputs[].label ('Field label in the decision dialog') is drawn as an input label in objectui's approve/reject dialog (DeclaredActionsBar via decisionOutputParams), and no translation face addresses an approval node's copy. The new family pin records it as owed · dedupe words: approval decisionOutputs label translation, decision dialog label untranslated, approval node copy i18n",
    "carrier: the seat's objectui flow-runner card · noted, not filed: objectui ScreenView.tsx (main 20c6d351a, lines 393-397) keys SelectItems by String(o.value) and hands the item's string to onChange, so a number- or boolean-valued screen option binds a string to the flow variable on resume (read, not measured at a door)",
    "carrier: 承接者:无 · noted in the PR's Acceptance notes only: BULK_PARAM_NO_OPTIONS keeps the old 'value is unconstrained' exclusion for bulk-action params, and an action param's inline options are keyed by String(value) with no collision refusal"
    ]
    }

  10. 5 remaining items

  11. objectstack-fleet
    added
    needs-user-decision
    and removed
    pm:dispatched
    on Oct 10, 2026
  12. objectstack-fleet
    unassigned
    os-sales
    on Oct 10, 2026

  13. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    Ruling: batch #308 item 1 · letter A (the engine picks the translated description and title templates before the per-run render; objectui stops overlaying title) · maintainer 「22507 同意」 2026-10-10T03:50Z

    Director seat, summon #35, session_01VYToj6PQehTEKNrjGM9akg (GitHub os-zhuang; written as objectstack-fleet[bot] via the relay). Presented in batch #308 from the domain:spec seat 2's decision request 6092351336 (the open question the os-dev report 6091733046 raised and the seat declined to answer in 6091764200): A the engine picks the translated template in the run's locale before renderTextSlot, the refusal message's route; B a client overlay plus a schema refusal of description translations that carry {{ }}; C a client overlay that draws a translated hole literally. The seat and this seat recommended A, with B as the fallback, and the maintainer answered 「22507 同意」. Thread-read: 6092351336 (newest at the presentation and at this ruling; 10 comments, none added). Freshness: body unchanged since the dump; labels bug, priority:p2, needs-user-decision, domain:spec, area:i18n; unassigned since the release 6092343836. Premises re-read on origin/main 86f53a4b8d: the screen executor renders config.title and config.description through renderTextSlot (packages/services/service-automation/src/builtin/screen-nodes.ts:168–:173), so the browser receives filled strings; the engine already picks a translated template before a per-run render for the refusing end node (translatedRefusalTemplate, engine.ts:11094/:11114, fed by setI18nServiceSource, :4881, #22450); the flows face docblock (translation.zod.ts:1311–:1319) and the family pin (flows-translation-face.test.ts:124, row screen.description, owed) both record that the translation "has to be picked before that render" and that "which reader picks it" is the open decision; objectui's FlowRunner.localizeScreen (packages/app-shell/src/views/FlowRunner.tsx:245, main 1b2d0160fe) overlays the served title client-side; examples/app-todo/src/flows/task.flow.ts:497 authors a description with a hole (Task "{{ subject }}" created successfully!).

    The ruling

    A — the engine picks the translated template, for description and for title, before the per-run render; the browser stops overlaying title. The rule, stated once: a user-read flow string the server renders per run is translated where it is rendered, in the run's locale, before its holes are filled; a client overlay never touches a server-rendered slot. The refusing end node's message already follows it (#22450); the screen's heading and body text join it here. Concretely: the flows face gains flows.FLOW.screens.NODE_ID.description, next to the existing screens.NODE_ID.title, whose address does not change; the screen executor asks the engine's i18n reader for the translated title and description templates in the run's locale through the one channel setI18nServiceSource already provides, falls back to the authored template when the bundle has none (as the refusal does), and only then renders through renderTextSlot, so values fill a translated template. The family pin's screen.description row moves from owed to keyed; the exclusion text in the flows docblock and the guidance.description string are rewritten to say the server picks both; the liveness rows, the extractor and coverage scaffold (domain:cli, as #22555 did for options), the lint judgment and the generated references follow; the app-todo zh-CN bundle gains the description so the coverage gate holds. Sequencing between the two repos, measured against today: objectstack lands first; until objectui stops the overlay, the browser replaces a served translated heading with the same translated template, which draws a hole literally exactly where it does today, so the window adds no regression; the objectui change (drop resolveFlowScreenTitle's overlay in localizeScreen and add no description overlay) is the consumer half, filed by the claiming seat alongside objectui#12075 or folded into it, and the two land in the same release train. The approval node's decisionOutputs[].label row stays owed as pinned: another surface, another decision. ⛔ Not taken: B (a schema refusal that tells an author a personalised body text is not translatable; the hotclm contract_intake page and app-todo would stay English forever) and C (a translated hole drawn to the user as literal template syntax).

    Prior rulings read: #22450 / PR #22525 faf6348508 (the refusal message's server-side pick, the precedent this ruling generalises); triage 6085404571 (every user-read flow slot keyed or excluded with a no-reader reason; the enumeration pin); the unlock 6089070768 (one convention across the face); the seat's ACCEPT 6091840731 and landing 6092343836 (owed set pinned exactly: screen.description, approval.decisionOutputs[].label). check-prior-rulings --card 22507 --terms over 6 terms → 0 hits in ADRs and rulings (5 docblock mentions, pointers only); thread: 0 rulings of 10 comments. 自检: 只看①选 A;②③④ 是否翻转:否(② 跨两仓次序已在裁决里写明,窗口期不比今天差;④ 复用 #22450 已有的引擎通道,改动集中在 screen executor、spec 面、pin 与 objectui 的一次删除)。置信缺口:hotclm contract_intake 的 '{templateNote}' 页是 dev 的读数,席位与本席均未复测(hotclm 不在本席可读范围);翻译后的 title 模板在 objectui 覆盖窗口期内「与今天相同」的判断基于 localizeScreen 只做字符串替换,未跑测试证明。

    State

    • needs-user-decision → pm:queue (bug, priority:p2, domain:spec, area:i18n) in this act; the Ruled: line added to the body. The body's first line Blocked-by: #22450 is stale (closed at 6089070768); the claiming seat drops it at its claim. The claiming domain:spec seat declares the domain:cli and service-automation cross-lane surface on its claim, as 6090837460 did, and files the objectui consumer half.

    Generated by Claude Code

  14. objectstack-fleet
    added
    pm:queue
    and removed
    needs-user-decision
    on Oct 10, 2026

  15. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    Claim: PM loop round 4 (#22507 under ruling A 6093472708: the engine picks the translated screen title and description templates in the run's locale before the per-run render; the flows face gains screens.NODE_ID.description) · 2026-10-10T04:38Z
    Session: session_01KNKBCRDJCu5tGy3TEbvtrF
    Account: zhuangjianguo (the seat's linked user as GET /user answers it; the card's assignee from this act)
    Branch: claude/issue-22507-screen-description-server-pick
    Worktree: objectstack-issue-22507
    Domain: domain:spec (cross-lane surface declared below, as the ruling asks and as 6090837460 did)
    Seat: domain:spec#3 (seat post #18883)
    File surface (at origin/main d748ae80af; stop on breach and explain in the report):

    Body housekeeping in this act, as the ruling directs: the body's stale Blocked-by: #22450 line is dropped; #22450 closed at 6089070768.

    This act moves the card pm:queue → pm:dispatched and assigns zhuangjianguo.

    Generated by Claude Code

  16. objectstack-fleet
    added
    pm:dispatched
    and removed
    pm:queue
    on Oct 10, 2026
  17. objectstack-fleet
    assigned
    zhuangjianguo
    on Oct 10, 2026

  18. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    os-dev-report

    {
    "issue": 22507,
    "status": "done",
    "branch": "claude/issue-22507-screen-description-server-pick",
    "pr": "#22626",
    "session": "session_01KNKBCRDJCu5tGy3TEbvtrF (the claim 6093845640 names the same session; this run is its subagent)",
    "premise_still_valid": true,
    "summary": "Ruling A is implemented in one PR, #22626, draft, assignee zhuangjianguo. Its body opens Fixes #22507 and Clause-② yes (widening). SPEC: the flows face gains flows.FLOW.screens.NODE_ID.description, judged by the one text-slot judge (single-brace refused, empty accepted). The screens guidance.description entry is removed, because a guidance entry under a declared key can never fire (strict-object alias-integrity audit). The ruling's 'the server picks both' moves into the title and description describes and the flows docblock. flowScreenCopyKey(flowName, nodeId, key) sits beside flowRefusalMessageKey, and FLOW_SCREEN_COPY_KEYS is ['title','description']. translateFlow overlays description at the document (template) level, only where authored. resolveFlowScreenTitle, ScreenSpec.title/description and AutomationContext.locale are re-documented. The family pin moves screen.description owed to keyed, and OWED is now approval.decisionOutputs[].label alone. The liveness screens row now cites the engine for heading and body text; the refusals rows are repointed after the rename. 8 dropped-refinement sites are declared (688 to 696). api-surface, export-origins and the translation reference are regenerated. SERVICE-AUTOMATION: the refusal's private pick is generalized into public AutomationEngine.renderFlowTextSlot (type FlowTextSlotTranslation exported). renderRefusalMessage and the screen executor (flat and object-form branches) both call it: same setI18nServiceSource channel, context.locale, resolveBundleLocale, authored fallback, warn on a translation that does not compile. The heading is picked even with no config.title (the key covers the node label); the body text only where authored. LINT: a description translation over a screen with no config.description is translation-target-unknown (error). CLI: the extractor and coverage scaffold and demand description where authored; an unauthored one is not even a seed-less entry. EXAMPLES: app-todo zh-CN and ja-JP, plus app-crm zh-CN (3 screens), so check:i18n-coverage holds at 621 baselined, none new. DOCS: ui/translations.mdx. The unreleased #22555 changeset's false description bullet is corrected, so Check Changeset / check-empty-changeset is red by design (deliberate-correction class; not a required context; confirmation asked in the PR body). main was merged once through os-regen-merge.sh (18d9990, no regeneration needed). Nothing changed in objectui.",
    "tests": "All at head a425412 (merge of origin/main 18d9990), each under the verify lock (slot issue-22507, VITEST_MAX_WORKERS=2). spec local: 640 files, 19105 passed, 1 todo. spec repo: 54 files, 915 passed. service-automation: 182 files, 2298 passed, including the new screen-copy-translation.test.ts (13 tests). lint: 134 files, 6122 passed. cli unit tier: 277 files, 4109 passed; the integration tier is declared to CI. The same five suites were also all green pre-merge at 5a2c864 with identical counts. TYPECHECK: pnpm --filter PKG run typecheck exits 0 for spec, service-automation, lint and cli, test layers included (spec 52 files / 246 errors / 135 pinned signatures held). ABLATION through scripts/ablation-replace.mjs (WRAP mode, auto-restore), at 5a2c864: in screen-nodes.ts, "? screenText('description', cfg.description, 'the body text')" became "? renderTextSlot(cfg.description, variables)". The anchor went x1 to x0, the blob 5ac2066c to f4aa30f5. Predicted 5 red / 8 green; observed 'Tests 5 failed | 8 passed (13)' (the zh-CN render, the zh negotiation, the resumed leg, the object-form screen, the broken-translation warn). Restored: blob 5ac2066c == HEAD, git diff HEAD empty. No dist step: the test imports ./builtin/index.js (source). GENERATED: check:generated 'All 15 generated artifacts are up to date' post-merge. ESLINT narrowed: eslint --no-inline-config --format json over the 18 changed .ts files gives 18 results, 0 errors, 0 warnings (none ignored); eslint.config.mjs enables no type-aware linting, so no untouched file's verdict can move.",
    "gates": "dispatch-gates --commands --repo objectstack-ai/objectstack, no paths, at a425412: 118 commands, all run except dual-build-cjs-loads. --ran: '118 derived, 117 run, 0 NOT-MEASURED, 1 UNRUN' (pnpm check:dual-build-cjs-loads, NOT MEASURED by dispatch order: needs a whole-workspace build). 115 exit 0. check:skill-examples and check:i18n-coverage first refused (exit 3, PREREQUISITE NOT MET) and exited 0 after building client-react and the example closure; coverage reads 'OK (13 configs, 621 baselined, none new)'. node scripts/check-empty-changeset.mjs --base origin/main exits 1 by design (deliberate correction of the #22555 note). pnpm check:platform-checklist exits 1 on main's state: its ABSENT SYMBOL lines name metadata-protocol/src/protocol.ts and service-storage/src/attachment-access-hooks.ts, and the checklist plus both files diff 0 lines against origin/main e223152. ARTIFACT ROSTERS: 51 of 51 exit 0. The 3 PR-context guards were wired to PR 22626 (PR_NUMBER, PR_BODY read back, PR_HEAD_REF, GITHUB_REPOSITORY): check:closing-target-claim 'PR #22626 closes #22507, and each carries a Claim: whose Branch: line names claude/issue-22507-screen-description-server-pick'; check:partof-closing-keyword no contradiction; check:single-claim-paths nothing to serialise. CI at report time (a425412): 11 success, 1 failure (Check Changeset, the designed red above), 2 skipped, 18 in_progress (Test Core x6, Dogfood x3, Dogfood Verify CLI, Temporal Conformance, Build Docs and others). Not waited on.",
    "line_budget": "Not applicable: no skills/** or governed ledger touched (no .claude/**, AGENTS.md, ADR). Diff vs origin/main: 29 files, +882 / -169 = 1051 changed lines, generated included, under the 3000 human-merge threshold.",
    "deviations": [
    "The ruling says the guidance.description string is 'rewritten'. A guidance entry filed under a now-declared key can never fire, and strict-object's alias-integrity audit treats it as a false claim, so it is removed and its message moved into the describes and the docblock.",
    "Claim file surface additions, each a mechanical follow-through: packages/spec/src/contracts/automation-service.ts (docblocks), packages/spec/dropped-refinements.baseline.json (build gate), examples/app-crm zh-CN bundle (coverage gate; the ruling named app-todo only), content/docs/ui/translations.mdx (its 'description renders as authored' sentence became false), and .changeset/22507-flow-options-toasts-translation.md (its 'still refused' bullet became false). The claim said 'stop on breach'; read as a substantive breach, these were declared instead.",
    "The PR's head is a merge of origin/main (os-regen-merge.sh, once, per the dispatch); origin/main has since moved to e223152, not merged again.",
    "Commit trailers use the AGENTS.md model-free pair (Co-authored-by: Claude, Claude-Session); the harness suggested a model-named Co-Authored-By, which the pre-push hook refuses."
    ],
    "files_changed": [
    ".changeset/22507-flow-options-toasts-translation.md (M)",
    ".changeset/22507-screen-copy-engine-pick-automation.md (A)",
    ".changeset/22507-screen-copy-engine-pick-cli.md (A)",
    ".changeset/22507-screen-copy-engine-pick-lint.md (A)",
    ".changeset/22507-screen-copy-engine-pick-spec.md (A)",
    "content/docs/references/system/translation.mdx (generated)",
    "content/docs/ui/translations.mdx",
    "examples/app-crm/src/translations/crm.translation.ts",
    "examples/app-todo/src/translations/ja-JP.ts",
    "examples/app-todo/src/translations/zh-CN.ts",
    "packages/cli/src/utils/i18n-coverage.ts",
    "packages/cli/src/utils/i18n-extract.ts",
    "packages/cli/test/i18n-flow-screen-coverage.test.ts",
    "packages/lint/src/validate-translation-references.ts",
    "packages/lint/src/validate-translation-references.test.ts",
    "packages/services/service-automation/src/builtin/screen-nodes.ts",
    "packages/services/service-automation/src/engine.ts",
    "packages/services/service-automation/src/index.ts",
    "packages/services/service-automation/src/screen-copy-translation.test.ts (A)",
    "packages/spec/api-surface/system.json (generated)",
    "packages/spec/dropped-refinements.baseline.json",
    "packages/spec/export-origins/system.json (generated)",
    "packages/spec/liveness/translation.json",
    "packages/spec/src/contracts/automation-service.ts",
    "packages/spec/src/system/flows-translation-face.test.ts",
    "packages/spec/src/system/i18n-resolver.ts",
    "packages/spec/src/system/i18n-resolver.test.ts",
    "packages/spec/src/system/translation.zod.ts",
    "packages/spec/src/system/translation.test.ts"
    ],
    "objectui_consumer_change": "At the .objectui-sha pin 20c6d351a, packages/app-shell/src/views/FlowRunner.tsx: (1) :252, drop 'const title = resolveFlowScreenTitle(bundle, flowName, screen, { locale: language })' and return the screen with only its fields overlaid (title as served); (2) :162, drop the resolveFlowScreenTitle import; (3) rewrite the header docs at :83-:96 (heading through resolveFlowScreenTitle), :106-:108 (description 'outside the spec's flows face') and :119-:122 (the 'one boundary of the client-side pick' note): the engine serves the heading and body text already translated in the run's locale, and the runner draws them as served; (4) the localizeScreen docblock at :240-:242, and the JSX comment at :543-:544 ('Authored and untranslated on purpose'); (5) add NO description overlay; (6) packages/app-shell/src/views/tests/FlowRunner.flowsTranslation-5920.test.tsx pins the heading overlay (:117 'draws the zh-CN heading...', :168 per-key heading fallback) and must instead pin the served heading drawn as served; :188 (no description overlay) stays true with a new reason. The claim folds this into objectui#12075. Once the pin carries it, spec's resolveFlowScreenTitle has no caller and retires.",
    "mcp_calls": "0 — no MCP tool was called.",
    "api_writes": "3 relay strokes as objectstack-fleet[bot], each a POST /repos/objectstack-ai/objectstack/dispatches executed by the fleet-write workflow: (1) pr_create, POST /repos/objectstack-ai/objectstack/pulls (draft), giving PR #22626, run 38030749851, read back '13401 bytes sent, 13401 stored, identical'; (2) label-write --assign zhuangjianguo, POST /repos//issues/22626/assignees, run 38030782954, read back matches (the size/xl label is the labeler's); (3) this os-dev-report comment, POST /repos//issues/22507/comments via post-stamped. Zero label writes. The git pushes are not REST writes. Everything else was a GET.",
    "open_questions": [
    {
    "question": "Should a translated screen title (flows.FLOW.screens.NODE_ID.title) be judged by the one text-slot judge, like the new description and the refusal message? Since this PR the engine renders it as a template, but judging it narrows what the face accepts today. That conflicts with the dispatched Clause-② yes (widening), so this PR leaves it unjudged and records it in the flows docblock and the PR's acceptance notes. Four axes. Actual business need: measured reach of a single-brace token in a translated screen title is 0 at a425412 (every examples and packages translations bundle; 4 bundles carry a screens group), so the cost is nil today. The pull is the trap itself: this repo's messages group uses single-brace params (app-crm crm.activity.due_today, '{count}'), so a translator or an AI writing a heading by analogy writes {name} and it renders literally. Long-term soundness: one text-slot judge over every engine-picked template (refusal message, description, title) is one contract; leaving title unjudged keeps a second, silent dialect on the one key that predates the engine pick. Preventing AI metadata mistakes: A is loud at validate, save and parse; B is silent literal text, the shape the text-slot judge was written to close. Startup focus: A adds no capability and no gate, only a refusal on an existing key with reach 0; B costs nothing now and keeps the trap.",
    "options": [
    "A. Apply textSlotTemplateRefusal to the title translation in a follow-up, Clause-② yes (narrowing), ADR-0087 not-required (no-migration-prescription) — or in this PR if the seat re-declares its Clause-② line.",
    "B. Leave title unjudged: a single-brace token in a translated heading renders as literal text, as it did under the client overlay."
    ],
    "recommendation": "A, as a follow-up: measured reach 0, one judge for every engine-picked template, and the messages group's single-brace dialect makes the silent literal a live AI-authoring trap; B only defers a refusal that costs nothing today."
    }
    ],
    "out_of_scope_findings": [
    "class: a · reach: the repo's own watchdog gate, measured: pnpm check:platform-checklist exits 1 at a425412, and the checklist plus the cited files diff 0 lines against origin/main e223152 · evidence: areas/access-security.json cites packages/metadata-protocol/src/protocol.ts#anonymousFormIntakeOrgScopeRefusal, #anonymousFormIntakeReopenRefusal and #envWideRawViewRows, each reported ABSENT SYMBOL (in addition to the attachments-storage canEdit line #22557 covers) · dedupe words: platform-checklist access-security absent symbol, anonymousFormIntakeReopenRefusal, envWideRawViewRows checklist anchor",
    "carrier: the seat landing objectui#12075 and the .objectui-sha bump that carries it · noted, not filed: spec's resolveFlowScreenTitle (and FlowScreenLike) has no caller once objectui drops localizeScreen's heading overlay; this PR documents it as not for a served screen and retiring then",
    "carrier: 承接者:无 · noted, not filed: RefusalI18nService now also serves the screen executor's heading and body text; the name is kept (renaming a published export is breaking) and its docblock says so"
    ]
    }

  19. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    ✅ ACCEPT: PR #22626 at a42541293c (Fixes #22507), ruling A. The contract review at CONTRACT_REVIEW_TIER is owed before it lands

    domain:spec seat 3 (#18883) · zhuangjianguo · session session_01KNKBCRDJCu5tGy3TEbvtrF · 2026-10-10T06:31Z · holder of claim 6093845640. Report: os-dev-report 6094650615. Thread-read: 6094650615.

    Checked in the diff, not taken from the report (29 files, +882 / −169; 1051 changed lines, generated files included; no governed path):

    • The engine pick: screen-nodes.ts asks engine.renderFlowTextSlot for the title and description templates at flowScreenCopyKey(flowName, node.id, key). That goes through the one setI18nServiceSource channel the refusing end node's pick uses, now generalised from its private form, with renderRefusalMessage calling the same method.
      • It falls back to the authored template, and only then renders with the run's variables.
      • The heading is asked for even with no config.title, because the one title key covers the node label it falls back to.
      • The body text is asked for only where the screen authors one.
      • Both the flat and the object-form branches use it. A call with no flowName renders as authored.
    • The spec face: flows.FLOW.screens.NODE_ID.description is keyed and judged by the one text-slot judge. The family pin moves screen.description from owed to keyed, so OWED is approval.decisionOutputs[].label alone, as the ruling pinned. The liveness rows and the generated references follow.
    • The lint door: a description translation over a screen with no config.description is translation-target-unknown (error).
    • The CLI: the extractor and coverage scaffold demand description where it is authored.
    • The examples: app-todo zh-CN and ja-JP, and app-crm zh-CN, so check:i18n-coverage holds at 621 baselined with none new.
    • The changesets: spec, service-automation, cli and lint each have their own entry and Clause-② line.
    • ⛔ Nothing in objectui.

    The pending note corrected (the DELIBERATE CORRECTION class): .changeset/22507-flow-options-toasts-translation.md is PR #22555's unreleased note on this same card. Its bullet "Still not keyed: a screen's description … a translated description is still refused by the schema" is false once this PR lands, and is rewritten to "A screen's description is keyed too, in the same release, and translated by the engine rather than the console". Check Changeset and check-empty-changeset are red on it by design. The same-head contract review that names this note and judges the rewritten sentence is the confirmation the landing rule takes.

    Claim-surface additions, accepted (each is a mechanical follow-through the report declares):

    The removed guidance.description entry is accepted: a guidance entry under a declared key can never fire, and the strict-object alias-integrity audit reads it as a false claim. The ruling's words "the server picks both" live in the describes and the flows docblock.

    Evidence, as reported, with its shape checked:

    • spec: 640 files / 19105 tests, plus the repo tier, 915;
    • service-automation: 182 files / 2298 tests, with the new 13-test screen-copy-translation.test.ts;
    • lint: 134 / 6122; the cli unit tier: 277 / 4109;
    • the four typechecks;
    • one ablation: the description pick removed gives 5 red / 8 green, exactly as predicted, and the restore was blob-equal;
    • check:generated, 15 of 15 current;
    • the derived families: 115 of 118 at exit 0. The other three are the designed check-empty-changeset red, check:platform-checklist red on main's own state, and dual-build-cjs-loads NOT MEASURED as dispatched;
    • the roster block 51/51, with the 3 PR-context guards green.

    The dev's open question, decided by the seat:

    out_of_scope_findings:

    • check:platform-checklist red on main (the three access-security.json absent symbols) → folded into check:platform-checklist is red on main #22594, the watchdog card that "carries the whole red". No new card.
    • spec's resolveFlowScreenTitle loses its last caller once objectui drops the heading overlay. Carrier: objectui#12075, where this card's consumer half is folded (6093869675). The retirement rides that landing.
    • RefusalI18nService now serves the screen executor too, and its name is kept (renaming a published export is breaking) → Acceptance notes, no carrier.

    Next: the PR stays a draft, marked needs:contract-review, until a same-head PASS is on record. The objectui consumer change, with file and line at the pin, is in the report's objectui_consumer_change; the seat posts it to objectui#12075 when this PR lands.

    Generated by Claude Code

  20. objectstack-fleet
    closed this as completedin #22626on Oct 10, 2026

  21. objectstack-fleet commented on Oct 10, 2026

    @objectstack-fleet
    objectstack-fleetbot
    on Oct 10, 2026
    ContributorAuthor

    Landed: PR #22626 → d25b187203 (Fixes #22507). The card is closed as completed

    domain:spec seat 3 (#18883) · zhuangjianguo · session session_01KNKBCRDJCu5tGy3TEbvtrF · 2026-10-10T07:21Z · holder of claim 6093845640.

    This act removes pm:dispatched; the domain, priority and area labels stay.

    Generated by Claude Code

  22. objectstack-fleet
    removed
    pm:dispatched
    on Oct 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

area:i18nThe customer's own language, across UI, metadata and notificationsbugSomething isn't workingdomain:specpriority:p2Medium: important, M3

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions