feat(i18n): convert product name catalog batch 1

[ShotaroKataoka]

Problem / Motivation

#1964 added the supported {{productName}} interpolation seam, but the existing i18n catalogs still hardcode Kiro Crew. Converting all 181 eligible keys in one PR produces a diff over the 1 MB review-pipeline limit, while splitting by locale would violate catalogParity placeholder parity.

Why it matters

Until the catalog conversion finishes, a downstream edition can call setProductName() but most visible copy still renders the stock name. Converting reviewable batches closes that gap without asking maintainers or AI review lanes to audit a 1.4 MB mechanical diff, and without leaving any intermediate commit where one locale carries a placeholder that another drops.

What changed (motivation → approach → change)

This is batch 1 of 3 catalog PRs. It converts exactly 60 product-name keys across every catalog at once:

• Replaces the literal Kiro Crew with {{productName}} in the same 60 logical keys across en.json, en.manual.json, and all 11 shipped translations.
• Regenerates en-XA.json from the merged English catalog; it is not edited by hand.
• Keeps every apps.<id>.manifest.* value unchanged because [manifest-sync] requires byte-identical Python app.json prose.
• Normalizes Korean particles whose form depends on the rendered product name to both forms (이(가), 을(를), 은(는), 와(과), (으)로). An edition may supply a name with or without a final consonant, so a single fixed form is no longer grammatically valid.
• Deliberately skips 12 eligible keys that carry inherited untranslated values. Merely replacing the product name in those values makes the diff-scoped [changed-passthrough] gate correctly fail; the final batch will translate and convert them together rather than mixing translation cleanup into this mechanical chunk.

For every non-Korean shipped value in this diff, resolving productName to the stock default reproduces the previous value exactly. The Korean differences are only the required variable-dependent particle forms.

Tests

• I18N_BASE_REF=upstream/main npm run i18n:check — all 18 checks pass, including placeholder parity, changed-value QA, changed-passthrough, pseudolocale freshness, DNT integrity, and manifest sync.
• npx vitest run src/i18n/ — 604 passed, 1 expected failure.
• npm run build — passes.
• BRAND_BASE_REF=upstream/main python3 scripts/check_brand_name.py and ./scripts/scrub-lint.sh --no-history — pass.
• Full frontend coverage run: 22,205 tests passed; two timing assertions in the unrelated CliPanelCoverage.test.tsx theme-sync suite failed only in the 38-minute whole-suite run. That file passes 41/41 in isolation, and this PR changes catalog JSON only. Electron tests pass.

Manual verification

N/A — the structural behavior is covered more strongly by the catalog-wide parity and changed-value gates: all locales carry the same placeholder set, non-Korean stock rendering is exactly preserved, and Korean variable-dependent grammar is enforced by its style gate.

Related Issues

• Follow-up to feat(i18n): interpolate the product name via defaultVariables #1964
• Progress toward feat(i18n): overridable productName via interpolation.defaultVariables for edition builds #1957; the final catalog batch will close it

Checklist

• Single commit with a Conventional Commits title (feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)
• Existing tests pass and new tests added for new functionality (no new behavior in this data-only batch; feat(i18n): interpolate the product name via defaultVariables #1964 added the contract tests)
• Self-review completed; code follows project style guidelines
• Documentation updated (not applicable to this batch; feat(i18n): interpolate the product name via defaultVariables #1964 documented the seam and migration)
• No secrets, credentials, or internal references in the diff

[github-actions]

GPT 5.6 Review (fork) — ✅ no blocking findings

Reviewed 088c1bb19f985ca6099ef12d4cb66072e504f6c8 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 088c1bb

[github-actions]

Design Review (Fable 5, fork) — ✅ PASS

Design-level review of 088c1bb19f985ca6099ef12d4cb66072e504f6c8 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

The design checks out: the interpolation seam exists in the base (website/src/i18n/index.ts setProductName/defaultVariables), the batch axis (same keys across all locales) is the only one compatible with the placeholder-parity gate, the manifest-key exclusion matches the documented exception, and the Korean both-forms particle treatment is the standard correct solution for a variable noun. Description ↔ diff line up (60 logical keys; en split 34+26 across generated and manual catalogs).

Design-Verdict: PASS

Mechanical catalog conversion riding a seam built for exactly this, batched on the only axis the parity gates permit; no design-level concerns.

[DESIGN-REVIEWED] 088c1bb

[github-actions]

UX Review (Fable 5, fork) — 🟡 CONCERNS

UX-level review of 088c1bb19f985ca6099ef12d4cb66072e504f6c8 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Reviewed the full patch (14 locale files, 60 keys), the i18n interpolation seam in website/src/i18n/index.ts, and the PR description.

UX-Verdict: CONCERNS

Rendering is byte-identical for stock users everywhere except Korean, where correct particles were downgraded to hedged dual forms across ~30 visible strings.

Watch

• Korean grammar regression in the stock build. Every converted ko string trades a previously correct particle for the parenthesized dual form — e.g. "Kiro Crew가 열려 있고" → "{{productName}}이(가) 열려 있고", which renders as "Kiro Crew이(가)…" for all Korean users. These include high-frequency chrome (gateway disconnected banner, all four approval-mode tooltips/descriptions), so the awkward form is seen constantly and permanently — friction, not failure, but a visible quality drop where quality existed. Smallest fix: an i18next interpolation formatter that picks the josa from the resolved name's final jongseong ({{productName, josa(이/가)}}), keeping "Kiro Crew가" in stock and staying correct for editions.
• Mixed product names on one surface for editions mid-migration. In the approval-mode popover, yolo_tooltip now interpolates while the adjacent skipped key yolo_expiration_until_shutdown ("Stays on until Kiro Crew restarts…") stays literal, so an edition renders two product names in one popover until batch 3 lands. Transitional and edition-only; worth confirming batch 3 lands before any edition ships from an intermediate commit.

[UX-REVIEWED] 088c1bb

[github-actions]

First Principles Review (Fable 5, fork) — ✅ PASS

Premise-level review of 088c1bb19f985ca6099ef12d4cb66072e504f6c8 via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All evidence checked. The patch is a pure data conversion: every one of its 780 added lines carries {{productName}}, every one of its 780 removed lines carries the literal (720 Kiro Crew, 60 accented pseudolocale equivalents), 780 = 60 keys × 13 catalogs exactly as declared. The seam it feeds is already live at base (website/src/i18n/index.ts:187 defaultVariables: { productName }), and website/docs/i18n-catalog.md:97-100 explicitly plans this batched conversion. The Korean dual-particle forms follow a convention already used 171 times in the base ko.json.

First-Principles-Verdict: PASS

Completes a documented, already-seamed migration in declared mechanical batches; every difference is either the conversion itself or a deferral the description names.

What this change ships

Intent: make existing catalog copy actually render an edition's overridden product name — a FIX of the gap #1964 left (seam live, catalogs still literal).

1. 60 strings in all 13 locales now render the configurable product name — justified (mandated by website/docs/i18n-catalog.md §product-name, transitional note plans exactly this).
2. Korean copy shows dual particles (이(가), 을(를), 은(는), (으)로) around the name — declared; matches 171 existing dual-particle uses in base ko.json (grepped), required once the name is variable.
3. Pseudolocale en-XA regenerated with matching placeholders — justified, generated file.
4. 12 eligible keys left literal until the final batch — declared deferral, keeps the [changed-passthrough] gate honest.
5. apps.<id>.manifest.* strings keep the literal — justified ([manifest-sync] byte-identity invariant).

No new key, flag, code path, or public surface is introduced; the only consumer mechanism (interpolation.defaultVariables, index.ts:187) predates this PR. Verified no rider: zero added lines lack {{productName}}, zero removed lines lack the literal.

[FIRST-PRINCIPLES-REVIEWED] 088c1bb

[github-actions]

Opus 4.8 Review (fork) — ✅ no blocking findings

Reviewed 088c1bb19f985ca6099ef12d4cb66072e504f6c8 via the fork AI-review pipeline; updated in place on each push.

Review details

I've verified the key risks directly from the code:

• JSON structure & interpolation wiring: The {{productName}} variable pre-exists in index.ts (wired as interpolation.defaultVariables, defaulting to "Kiro Crew"), so rendered English is byte-identical to the old literals and component tests asserting "Kiro Crew" still pass.
• Manifest exception: In en.json the apps.*.manifest.* values (e.g. line 1150) appear only as unchanged context lines — they retain the literal, so the byte-for-byte check-app-manifest-sync.mjs gate is not broken.
• Placeholder parity (catalogParity.test.ts): the substitution is a brand-aware codemod, not a naive replace — it substituted the en-XA pseudolocale's transliterated form (Ķìŕø Çŕèẁ → {{productName}}) too, and every key I sampled (star_kirocrew_on_github, feature_off, updating_kirocrew, the agentImportFlow block) is substituted consistently across en, en-XA, and the translated catalogs, while a deliberately-skipped key (kiro_crew_uses_kiro_cli_as_its_agent_engine_comp) is left literal everywhere.

The discovery pass produced no candidates, and I could not ground any defect myself to the (a)/(b)/(c) bar at 80+ confidence.

No findings.

[OPUS-REVIEWED] 088c1bb