Translation Skill

Translate new entries in XLF translation files by finding state="new" targets, locating source context, and proposing translations for review.

Workflow

Step 1: Extract New Translation Entries

Run the extraction script using the Bash tool:

python3 skills/translate/scripts/find_new_translations.py

This outputs JSON with all changed XLF files and their state="new" entries, including translation IDs, source/target text, line numbers, and placeholders. If the files array is empty, inform the user that no new translations were found.

Step 2: Find Source Context

For each translation ID from the script output, use the Grep tool to find its usage in the codebase:

• Search for @@{ID} with file pattern **/*.ts to find $localize strings
• Search for @@{ID} with file pattern **/*.html to find i18n attributes

Use the Read tool on matched files to get 5-10 lines of surrounding context.

Step 3: Generate & Review Translations

For each state="new" entry, present to the user:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Translation ID: {id}
File: {xlf_file_path}
Target Language: {language_name}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Source (English):
  {source_text}

Context:
  {file_path}:{line_number}
  {code_snippet}

Proposed Translation:
  {proposed_translation}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Wait for user confirmation before proceeding. Options:

• Accept the proposed translation
• Provide an alternative translation
• Skip this entry

Step 4: Apply Approved Translations

For approved translations, update the XLF file:

1. Replace the <target> element content with the approved translation
2. Change state="new" to state="ai"

Before:

<trans-unit id="myTranslationId" datatype="html">
  <source>English text</source>
  <target state="new">Old or empty translation</target>
</trans-unit>

After:

<trans-unit id="myTranslationId" datatype="html">
  <source>English text</source>
  <target state="ai">Translated text</target>
</trans-unit>

Language Detection

Identify the target language from the filename pattern:

Placeholder Handling

Critical: Preserve all XML placeholder elements exactly as they appear in the source. These include:

Example with placeholders:

<source>Hello <x id="INTERPOLATION" equiv-text="{{name}}"/>, you have <x id="INTERPOLATION_1" equiv-text="{{count}}"/> messages.</source>
<target state="ai">Hallo <x id="INTERPOLATION" equiv-text="{{name}}"/>, Sie haben <x id="INTERPOLATION_1" equiv-text="{{count}}"/> Nachrichten.</target>

Translation Guidelines

1. Match existing tone: Review other translations in the file to maintain consistent style
2. Preserve placeholders: Copy placeholder elements exactly, only translate surrounding text
3. Consider context: Use the code context to understand how the text is used
4. Domain terminology: Use standard terms for:

- Mobility/transport: commute, emissions, CO2, modal split - UI elements: buttons, labels, tooltips - Business terms: audit, score, analysis

1. Formal vs informal: German typically uses formal "Sie" form in business software

Common Translation Patterns

German (de)

• "Save" → "Speichern"
• "Cancel" → "Abbrechen"
• "Loading..." → "Wird geladen..."
• "Error" → "Fehler"
• "Success" → "Erfolgreich"

French (fr)

• "Save" → "Enregistrer"
• "Cancel" → "Annuler"
• "Loading..." → "Chargement..."
• "Error" → "Erreur"
• "Success" → "Succes"

Spanish (es)

• "Save" → "Guardar"
• "Cancel" → "Cancelar"
• "Loading..." → "Cargando..."
• "Error" → "Error"
• "Success" → "Exito"

Hungarian (hu)

• "Save" → "Mentes"
• "Cancel" → "Megse"
• "Loading..." → "Betoltes..."
• "Error" → "Hiba"
• "Success" → "Sikeres"

Batch Processing

When multiple entries need translation:

1. Group by file for efficiency
2. Show a summary of how many entries need translation per file
3. Process entries one at a time, waiting for approval
4. After all entries are reviewed, show summary of changes made

Error Handling

• If no changed translation files found: Inform user and suggest running git status
• If no state="new" entries found: Inform user that all translations are up to date
• If context not found: Still propose translation based on source text alone, note missing context
• If file parse error: Report the error and skip the problematic file