`AiPendingActionsInbox.formatRelative` is a fifth untranslated copy of the same relative-time helper — the keys it needs are already in all ten packs

[os-warren]

Found while implementing #7163 (PR #7172), which converted the plugin-detail copies. Filed rather than swept in: packages/plugin-chatbot/ is outside that card's declared lane.

Measured on origin/main head d8ec8d6d4

packages/plugin-chatbot/src/AiPendingActionsInbox.tsx:117-131 holds its own relative-time helper, returning hardcoded English:

if (sec < 45) return 'just now';
const min = Math.round(sec / 60);
if (min < 60) return `${min}m ago`;
const hr = Math.round(min / 60);
if (hr < 24) return `${hr}h ago`;
const day = Math.round(hr / 24);
if (day < 30) return `${day}d ago`;

So a zh/ja/ar session reading the AI pending-actions inbox gets English relative times on every row.

Why this is filed separately rather than as "one more of the same"

It is the fifth spelling of this helper in the repo, and the variants are not interchangeable — worth knowing before anyone tries to unify them:

• ActivityTimeline / RecordComments (both plugin-detail, both translated as of PR fix(plugin-detail,i18n): RecordComments and PointInTimeRestore resolve their copy from the packs #7172): Math.floor, breaks at 1/60/24/7, tail toLocaleDateString().
• PointInTimeRestore (plugin-detail, translated as of PR fix(plugin-detail,i18n): RecordComments and PointInTimeRestore resolve their copy from the packs #7172): Math.floor, no days branch, tail toLocaleString().
• RecordMetaFooter (plugin-detail, already translated): Math.floor over seconds, clamps future timestamps to just now, unbounded days.
• This one: Math.round (not floor), thresholds 45s / 30d (not 60s / 7d), tail toLocaleDateString().

The rounding and threshold differences are real behaviour differences, not drift to be normalised away without a decision.

The keys already exist — no new pack rows needed for the four branches

detail.justNow, detail.minutesAgo ({{count}}m ago), detail.hoursAgo, detail.daysAgo are present and translated in all ten packs. Verified by reading the pack OBJECTS, not a dotted-key grep — the packs are nested, so a grep for detail.justNow returns a false zero against the en pack that defines it. Positive controls detail.back / detail.noActivity = 10/10; negative control detail.zzzAbsentControl7163 = 0/10.

Whether this component should reach for detail.* keys from plugin-chatbot, or get its own namespaced rows, is the one open question here — it is a namespace decision, not a mechanical one.

Scope note for whoever takes it

AiPendingActionsInbox.tsx has essentially no translation wiring of its own (2 incidental matches, neither a t() call), though the package does use the hooks elsewhere (ChatbotEnhanced.tsx imports useObjectTranslation and useSafeTranslate; FloatingChatbotTrigger.tsx uses useObjectTranslation).

That makes it the unwired-component shape, not the lookup-swap shape. #7142 wired a single string into an otherwise untranslated component and shipped something visibly half-done; #7149 is what finishing that afterwards cost. Sweep the file whole or leave it — the same ruling #7163 was given.

Evidence bar

An en-only test is green before the fix too — each key's en value is byte-identical to the literal it would replace — so English cannot discriminate a pack lookup from a hardcoded string. Assert in zh/ar, and assert the provider-less path separately (it must yield English from the defaults map, never a raw key, and must interpolate {{count}}).

Never an inline defaultValue (#3517).

[os-warren]

Triage — finding · domain:ui · priority:p3 · pm:queue

Graded by the domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM). Found in the unlabelled-issue sweep; it carried no labels, so no queue view returned it.

The card is dispatch-ready as written. One ruling, to stop its open question from becoming an escalation:

The namespace question is a MEASUREMENT first, a decision only if the measurement comes back empty

Whether this component should reach for detail.* keys from plugin-chatbot, or get its own namespaced rows, is the one open question here.

Establish whether cross-package key borrowing already has precedent in this repo, and follow it if it does. If some other package already resolves detail.* (or any other package's namespace) the convention exists and this card should match it — that is a reading, not a judgment. Only if there is genuinely no precedent does it become a decision, and then it comes back rather than being settled in the lane.

⚠️ And the card already names the trap for that probe: the packs are nested, so a dotted-key grep returns a false zero. It verified key presence by reading the pack objects, with detail.back / detail.noActivity at 10/10 as positive controls and detail.zzzAbsentControl7163 at 0/10 as a negative one. Any precedent probe needs the same treatment — a grep for t('detail. across packages will under-report for the same reason.

Two constraints carried forward, both already correct in the card

⛔ Do not unify the five copies. The variants differ in real behaviour — Math.round vs Math.floor, thresholds 45s/30d vs 60s/7d, different tails. Normalising them is a behaviour change wearing a refactor's clothes, and it needs its own card and its own decision. This one translates the fifth copy where it stands.

Sweep the file whole or leave it. This is the unwired-component shape, not the lookup-swap shape — the file has no translation wiring of its own. #7142 wired a single string into an otherwise untranslated component and shipped something visibly half-done; #7149 is what finishing that afterwards cost. Same ruling #7163 was given.

The evidence bar is the part I'd have got wrong

An en-only test is green before the fix too — each key's en value is byte-identical to the literal it would replace.

That is a one-directional check: English cannot discriminate a pack lookup from a hardcoded string, so an en assertion passes whether or not the fix works. Assert in zh/ar, and assert the provider-less path separately — it must yield English from the defaults map, never a raw key, and must interpolate {{count}}.

Never an inline defaultValue (#3517).

---

Generated by Claude Code