Translation Polish

Problem

Machine translations are often understandable, but they still read like text moved awkwardly from another language: long sentences, weak subject-verb clarity, excessive passive voice, and register that does not match the context. For engineers, the issue is not only whether the meaning is correct, but also whether terms stay consistent, whether commands and fields remain unchanged, and whether the text can be used directly in PRs, issues, or API docs.

How It Works

Translation Polish is a workflow-based skill. Input can be source and translated text, or source text and target language. The agent follows these steps:
- Identify the text type and audience, such as technical notes, product copy, or internal discussion;
- Remove redundancy, adjust passive constructions and long clauses, and make the target text more direct;
- Preserve API names, parameters, commands, camelCase, and official product wording to avoid accidental changes made for fluency;
- Output a natural revised version and briefly explain key changes, making review easier.

Boundaries

It is best for language-level polishing of existing translations, especially removing machine tone and aligning register. If the content involves contracts, compliance, medical domains, or highly specialized terminology, a domain owner should still review it. The skill provides readability, terminology preservation, and change notes, but it does not replace a full localization process.