Skip to content
Beta Multi-model support (OpenAI, Gemini, and more) launching soon.
RBA RBA Skills

localization-pack-generator

Installable Ships in copy-localization-generator — or copy the raw skill below.
Open Claude.ai

Paste into Claude Code, Claude.ai, the desktop app, or the browser extension — ⌘/Ctrl+V.

ExperimentalNo security review

Localization Pack Generator

Convert English source copy into a translation-ready pack for a target locale. The output is structured for direct import into the target CMS or TMS (Phrase, Lokalise, Crowdin), with cultural adaptation notes, placeholder integrity checks, and character expansion budget warnings so translators know what constraints apply before they start.

Usage

/localization-pack-generator <source file or directory>
  --locale <locale-code>
  [--format json|xliff12|xliff20|sitecore|optimizely|umbraco]
  [--expand-budget]
  [--notes]

Examples:

  • /localization-pack-generator src/locales/en.json --locale fr-CA --format json
  • /localization-pack-generator content/en/ --locale de --format xliff12 --notes
  • /localization-pack-generator src/locales/en.json --locale ja --expand-budget --format json

Supported locale codes (common)

Code Language / Region
fr-CA French — Canada
fr-FR French — France
de German — Germany
es-ES Spanish — Spain
es-MX Spanish — Mexico
ja Japanese
zh-CN Chinese Simplified
zh-TW Chinese Traditional
pt-BR Portuguese — Brazil
nl Dutch
it Italian
ko Korean
ar Arabic (RTL — flag layout impact)
he Hebrew (RTL — flag layout impact)
pl Polish
sv Swedish

Phase 1 — Source file intake

Read the source file(s). Detect format automatically:

  • JSON — flat { "key": "value" } or nested i18n objects (next-intl, i18next, react-intl)
  • XLIFF 1.2 / 2.0 — TMS export format
  • Markdown — standalone copy files from brand-voice-copy-generator output
  • CSV / TSV — spreadsheet exports

Extract:

  • All translatable string keys and their English source values
  • Any {{placeholder}}, {variable}, %s, %(name)s patterns — these must survive translation intact
  • Any ICU MessageFormat pluralisation patterns — flag these for translator context
  • Any HTML tags embedded in strings — flag; translators must preserve them

Count strings and produce an inventory summary before generating the pack.

Phase 2 — Character expansion budget

Different languages expand English source copy at different rates. When --expand-budget is set (or when targeting Japanese, German, Finnish, or Arabic), calculate expansion budgets for every string that has a known character limit.

Target locale Typical expansion over English Notes
German (de) +30–35% Compound nouns; long verbs
French (fr-*) +15–25% Gendered articles; formal register
Spanish (es-*) +15–25% Inflected verbs
Italian (it) +10–25% Similar to Spanish
Dutch (nl) +20–30% Long compounds
Polish (pl) +20–30% Inflected endings
Japanese (ja) −10–20% (chars) Kanji compression; but line breaks differ
Chinese (zh-*) −20–30% (chars) Very compact; line length looks short
Arabic (ar) +20–30% RTL layout; also right-aligns numerals
Hebrew (he) −5–10% RTL; shorter words

For each string with a CMS character limit, compute:

budget = limit - (english_length × expansion_factor)

Flag strings where the expansion budget leaves fewer than 10 characters of headroom — these need translator instructions to be concise.

Phase 3 — Cultural adaptation notes

When --notes is set, produce per-locale cultural notes in the <note> element (XLIFF) or a _notes sidecar JSON. These are guidance for human translators or review of AI translation.

Common adaptation topics by locale

fr-CA vs fr-FR:

  • Vocabulary differences: "courriel" (CA) vs "email" (FR); "magasiner" (CA) vs "faire du shopping" (FR)
  • Tone: Québec French skews more informal in marketing copy; France French skews formal
  • Currency: Canadian dollars vs euros; ensure {currency} placeholders are not hardcoded

de:

  • Formal "Sie" vs informal "du" — enterprise software defaults to "Sie"; consumer apps often use "du"
  • Compound nouns: translators may need to hyphenate for short CMS fields
  • CTA style: German CTAs are often longer; flag short-CTA fields (< 20 chars)

ja:

  • Honorific register (keigo) — enterprise and financial content uses formal keigo
  • Katakana for foreign tech terms (e.g. "サービス" for "service")
  • No spaces between words — character count is reliable but line-break logic differs

ar / he (RTL locales):

  • Flag all layout-impacting strings (navigation labels, CTA buttons, form labels)
  • Note that numbers and dates may need locale-aware formatting (not just translation)
  • OG images and hero images with embedded English text need locale variants

es-MX vs es-ES:

  • Pronoun: MX uses "tú" (informal) and "ustedes" (plural formal); ES uses "vosotros" (plural informal)
  • Vocabulary: "computadora" (MX) vs "ordenador" (ES); "carro" (MX) vs "coche" (ES)

Phase 4 — Pack generation

Format: json (default)

Flat key-value JSON. Values are machine-translated placeholders unless --translate flag is passed (requires external TMS integration — not in scope here; flag this clearly).

{
  "_meta": {
    "sourceLocale": "en",
    "targetLocale": "fr-CA",
    "generatedAt": "[ISO-8601]",
    "stringCount": 42,
    "translationStatus": "source-copy-ready-for-translation"
  },
  "auth.login.email.label": "Email address",
  "auth.login.email.placeholder": "name@company.com",
  "auth.login.submit.label": "Sign in"
}

When --notes is set, produce a companion [locale].notes.json:

{
  "auth.login.submit.label": {
    "characterLimit": 20,
    "expansionBudget": 7,
    "note": "CTA label — keep short; imperative verb required. de: use 'Anmelden'. fr-CA: 'Se connecter' (13 chars, PASS)."
  }
}

Format: xliff12

XLIFF 1.2 — compatible with Sitecore Translation Provider (SDL Trados, Lionbridge), Phrase, and most enterprise TMS platforms.

<?xml version="1.0" encoding="UTF-8"?>
<xliff version="1.2" xmlns="urn:oasis:names:tc:xliff:document:1.2">
  <file source-language="en" target-language="fr-CA"
        datatype="plaintext" original="en.json"
        date="[ISO-8601]">
    <header>
      <note>Character limits and expansion budgets in trans-unit notes.</note>
    </header>
    <body>
      <trans-unit id="auth.login.email.label" approved="no">
        <source>Email address</source>
        <target state="new">Email address</target>
        <note>Form field label. Limit: 40 chars. Expansion budget (fr-CA +20%): 33 chars.</note>
      </trans-unit>
    </body>
  </file>
</xliff>

Format: xliff20

XLIFF 2.0 — for modern TMS platforms (Lokalise, Crowdin). Same structure adapted to the 2.0 schema.

Format: sitecore

Produces an XML sitecore package structure compatible with the Sitecore Translation Provider and Language Fallback item export. Mirrors the Sitecore item field path structure.

<languages>
  <language Name="fr-CA">
    <item id="{...}" key="auth-login-email-label">
      <field id="{...}" name="Value">
        <content>Adresse courriel</content>
      </field>
    </item>
  </language>
</languages>

Format: optimizely

Produces a fr-CA.xml language file in the Optimizely LangFolder format:

<?xml version="1.0" encoding="utf-8"?>
<languages>
  <language name="French (Canada)" id="fr-CA">
    <auth>
      <login>
        <emaillabel>Adresse courriel</emaillabel>
      </login>
    </auth>
  </language>
</languages>

Format: umbraco

Produces a JSON export structured for Umbraco Dictionary items + Language Variants:

{
  "locale": "fr-CA",
  "dictionaryItems": [
    {
      "key": "Auth.Login.Email.Label",
      "value": "Adresse courriel",
      "sourceValue": "Email address",
      "note": "Form field label."
    }
  ]
}

Phase 5 — QA pre-flight

Before delivering the pack, run these checks:

  • All source strings are present in the pack (no missing keys)
  • All placeholder patterns ({{var}}, {var}, %s) are preserved in target strings
  • ICU plural patterns are flagged for translator attention
  • HTML tags within strings are preserved and uncorrupted
  • RTL locales (ar, he) are flagged for layout impact review
  • Strings with expansion headroom < 10 chars are flagged
  • No _notes keys have been accidentally included as translatable strings
  • XLIFF <target state> attributes are set to "new" (not "translated" — TMS sets this)

Reference

See reference.md for:

  • Full locale code list with expansion rates
  • Placeholder pattern reference (ICU, printf, Handlebars, i18next)
  • CMS language configuration notes (Sitecore Language Fallback, Optimizely Language Branches, Umbraco Language Variants)
  • TMS integration patterns (Phrase webhooks, Lokalise CLI, Crowdin API)
Technical details
Plugin
copy-localization-generator
Author
Dakota Smith
Skill name
localization-pack-generator
Maturity
experimental
Security review
No security review
Owners
  • design
  • strategy
Argument hint
<source file or dir> --locale <locale-code> [--format json|xliff12|xliff20|sitecore|optimizely|umbraco] [--expand-budget] [--notes]
Allowed tools
  • Read
  • Write
  • Edit
  • Bash(find *)
  • Bash(ls *)
  • Bash(cat *)
  • Bash(grep *)
Source
plugins/copy-localization-generator/skills/localization-pack-generator/SKILL.md

Edit this skill on GitHub — opens GitHub's web editor. No git knowledge needed.