Skills

catalog-translation-localizer

Catalog content v3 Ecomdly recommended
@tonecheck 197 installs updated 12d ago MIT license

Catalog translation localizer

Machine translation gets a catalog 80% there; the last 20% is terminology, sizes and numbers. This skill works through a full export in batches, keeps your terms consistent across 4 000 products, and tells you exactly what still needs a human.

When to use

  • Launching a store in a new market (DE, FR, PL, CZ...) from an existing catalog export.
  • Keeping a translated catalog in sync: only new or changed SKUs since the last run.

Input

  • Catalog export (CSV/XML) with id, title, description, attributes, variant names, SEO title/meta.
  • Source and target locale, e.g. en-GB → de-DE.
  • Glossary: source term → approved target term (per category if needed).
  • Do-not-translate list: brand names, product lines, model numbers, trademarks, ingredient INCI names.
  • Target market rules: units (metric/imperial), size system (EU/UK/US), currency and decimal format.
  • Field length limits (e.g. feed title 150, meta title 60).

Batch workflow

  1. Split the export into batches of 50–200 SKUs, grouped by category so the glossary stays in context.
  2. Before each batch, read progress.json ({last_id, done, total, glossary_version}); skip what is done. A rerun resumes, never restarts.
  3. Translate field by field. Glossary terms are mandatory; do-not-translate tokens are copied byte for byte.
  4. Localise, do not just translate:
    • Units: convert only when the market expects it, keep the original in brackets for technical products: 30 cm (11.8 in).
    • Sizes: map with the store's size chart only. No chart, no conversion, flag it.
    • Prices and currency: never convert amounts; prices come from the target store's price list. Only format (1.299,00 €).
    • Dates, decimal separators and quotes in the target convention.
  5. After each batch, write the output rows and update progress.json, then report 2 480 / 4 000.

QA pass (per batch, then whole catalog)

  • Placeholders, HTML tags and merge tags intact and equal in count to the source.
  • Every number in the source appears in the target (after unit conversion).
  • Glossary hits: source term present → approved target term present. Conflicts listed, not resolved silently.
  • Length limits respected; over-length rows shortened and flagged.
  • Untranslated leftovers (source-language words not on the do-not-translate list).
  • Never add claims or benefits the source does not contain; ambiguous source text is flagged, not guessed.

Output format

progress: 2480 / 4000 (batch 25 of 40, glossary v3)
id,locale,title,description,flags
SKU-1188,de-DE,"Merino-Wandersocken Crew, 3er-Pack, Grau, Gr. 43-46","...",""
SKU-1189,de-DE,"Trailrunning-Schuh Speedcross 6 GTX","...","size chart missing: US 9 not converted"
QA: 200 rows · 0 placeholder errors · 3 glossary conflicts · 2 over-length titles

License

MIT