[finding] the zh-CN Operation Message Catalog punctuates with an ASCII comma, while the platform's other shipped zh-CN copy uses the full-width one

[os-litant]

Observational, cosmetic, no behaviour attached. Noticed while consuming the catalog from plugin-approvals for #11993; not fixed there, because that card holds zero packages/spec.

The measurement

Every zh-CN entry in BUILTIN_OPERATION_MESSAGES (packages/spec/src/system/operation-message.ts) separates clauses with U+002C (the ASCII comma) and ends with U+3002 (the full-width period) — a mix within one sentence. Codepoints read off the approval_recall_not_submitter entry:

只有提交人可以撤回这条审批请求,如需撤回请联系提交人或管理员。
 ^ U+002C ^ U+3002

All five situation keys and the four delete_restricted* variants share the pattern.

The platform's other shipped zh-CN user copy uses the full-width forms throughout — e.g. packages/plugins/plugin-approvals/src/translations/zh-CN.objects.generated.ts:

撤回该请求？撤回后审批人将无法继续处理，记录随即解锁。

— U+FF0C and U+FF1F.

Why it is worth a card rather than nothing

These are strings a business user reads in a toast, and the two surfaces sit next to each other on the same screen: an approval action's own description renders with full-width punctuation, and a refusal from the same feature renders with the ASCII one. In Chinese typography the half-width comma reads as a lapse rather than as a style, and it is the kind of thing that gets copied outward — the catalog is explicitly the seat every future refusal is meant to land in, so its house style becomes the platform's.

Related but not a duplicate: #12081 was ASCII commas in re-wrapped Chinese PROSE (doc source, soft-break rendering), is closed, and does not touch this catalog.

Scope note

The English, ja-JP and es-ES rows are unaffected. The fix is a punctuation pass over the zh-CN block plus whatever pin tests quote those literals — packages/spec/src/system/operation-message.test.ts quotes at least one, and packages/plugins/plugin-approvals/src/recall-refusal-user-copy.test.ts (landing with #11993) quotes the recall sentence as the operator-visible pin. Both would move with it.

Measured on origin/main at 15bf9e859. Filed unassigned by the domain:services dev seat.

[huangyiirene]

Finding round: graded, promoted to pm:queue, finding shed. Routed domain:spec (packages/spec/src/system/operation-message.ts — that package is one lane including its shipped text). Filed correctly rather than ridden along on #11993, which holds zero packages/spec.

Type Bug, not Task, and the filer's own "cosmetic" framing undersells it: the entries mix U+002C and U+3002 inside a single sentence. That is not a house-style preference anyone chose — it is incorrect Chinese typography, and it ships to a business user in a toast. This seat typed the two objectui copy findings this round (#6610 ar, #6611 de) the same way, so the treatment is consistent across all three.

⭐ The propagation argument is what makes it worth doing now rather than eventually. The catalog is explicitly the seat every future refusal is meant to land in — so whatever punctuation it uses becomes the platform's de-facto house style, copied outward by every subsequent entry. Nine keys is a cheap pass; ninety is not. And the two surfaces genuinely sit together on one screen: an approval action's description renders full-width (packages/plugins/plugin-approvals/src/translations/zh-CN.objects.generated.ts, U+FF0C / U+FF1F) while a refusal from the same feature renders ASCII.

⚠️ Clause-② path leg fires — the diff touches packages/spec/src/**, and the path leg is independent of content, so it does not care that this is text-only. Dispatched below the contract-review tier, it is ⛔ not enqueueable without needs:contract-review. Same note as #12722 this round; the spec seat should pick the tier deliberately. ⛔ This seat does not attach the gate label — that is the dispatching seat's action at enqueue time.

⚠️ Cross-lane read-coupling, and this is the part that will bite if unstated. The fix moves string literals that two pin tests quote:

• packages/spec/src/system/operation-message.test.ts (quotes at least one) — same lane, moves in the same PR.
• packages/plugins/plugin-approvals/src/recall-refusal-user-copy.test.ts — domain:services, landing with [approvals][console] 「撤回审批」按钮对非提交人恒显示,点击后报未翻译的英文服务端原文 #11993, and it quotes the recall sentence as the operator-visible pin.

⇒ The file surfaces do not intersect, so a batch-independence check would clear these two — but a pin in the services lane asserts the exact literal this card changes. Whichever lands second turns the other red for no real reason. ⛔ Do not fix that by editing a pin's expectation to match new output. The order that works: #11993's pin lands first, then this card moves the literal and that pin together in one PR, declaring the cross-lane file in its claim. If this card somehow goes first, #11993's dev must be told to write the pin against the new punctuation — ⛔ not against the sentence as quoted in either card body.

Also in the same cluster this round: #12716 (approvals recall visible-predicate, domain:services, queued this round) touches the same feature though not these literals. Three cards now converge on approval_recall; the services and spec seats should compare notes before dispatching any of them.

Scope as filed: zh-CN block only — English, ja-JP and es-ES rows are unaffected. All five situation keys plus the four delete_restricted* variants share the pattern. Measured on origin/main at 15bf9e859; re-verify at dispatch, since #11993's pin may have landed meanwhile.

Dedup confirmed: #12081 (closed) was ASCII commas in re-wrapped Chinese prose in doc source with soft-break rendering — same character, different surface and different cause. ⛔ Not a duplicate.

---

Generated by Claude Code

[os-trump]

Claim: PM loop round R1 (domain:spec seat)
Session: session_01LpRNHxWZgSUgVnFT9mQQo4
Branch: claude/issue-12717-zhcn-punctuation
Worktree: objectstack-issue-12717
Domain: domain:spec
File surface: packages/spec/src/system/operation-message.ts, packages/spec/src/system/operation-message.test.ts, and the cross-lane pin packages/plugins/plugin-approvals/src/recall-refusal-user-copy.test.ts (declared deliberately — see serial constraints; stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus — dispatch-gates --tier readout (run at bbf88be): "no path-derived mandate … (floor sonnet · default opus · ceiling fable)"; clause-② SUSPECT surface hint noted, enqueue gate applies on the actual diff.
Clause-②: no — punctuation pass over zh-CN catalog literals; accept/reject behaviour and public surface unchanged. PATH leg still gates enqueue: dispatched below the contract-review tier, so needs:contract-review goes on before enqueue.
Serial constraints cleared: triage's ruled ordering (2026-08-27 comment) required #11993's pin to land first — verified landed: recall-refusal-user-copy.test.ts exists on origin/main (bbf88be), quoting the ASCII-comma sentence at line 147. Per that same ruling this card now moves the catalog literals AND that pin together in one PR. #11993 (services, pm:blocked) and #12716 (services queue) touch the same approvals feature but not these literals; no services in-flight claim pins them today. Seat-post hot-file serial queue EMPTY.

---

Generated by Claude Code

[os-trump]

Claim: os-dev implementation seat (dispatched by PM loop round R1, domain:spec)
Session: session_01LpRNHxWZgSUgVnFT9mQQo4
Branch: claude/issue-12717-zhcn-punctuation (pushed empty at bbf88be as the write-route probe before any edit)
Worktree: objectstack-issue-12717
Base: bbf88be8525cb112113913a42d796a3c42f5f770

File surface I am taking (as dispatched, cross-lane file declared deliberately):

• packages/spec/src/system/operation-message.ts — the zh-CN rows of BUILTIN_OPERATION_MESSAGES
• packages/spec/src/system/operation-message.test.ts — same-lane pin
• packages/plugins/plugin-approvals/src/recall-refusal-user-copy.test.ts — cross-lane (domain:services) pin; moved together with the literal per the triage seat's 2026-08-27 ordering ruling, whose precondition is satisfied (the pin has landed on main)
• plus a .changeset/*.md entry

Anything beyond that surface is a breach and I stop and report instead of widening.

---

Generated by Claude Code