Rewrites non-English text to remove translationese and read as native-written prose.
npx skills add https://github.com/sharosoo/dotfile --skill naturalize-sharosoo
Please help me install this Agent Skill. Skill: naturalize Source: https://github.com/sharosoo/dotfile/tree/main/.claude/skills/naturalize Command: npx skills add https://github.com/sharosoo/dotfile --skill naturalize-sharosoo
High-intent search queries and answers about installing and using this skill.
Run a separate rewrite pass that regenerates the text in Korean rather than patching awkward spans. The Skill loads a Korean checklist targeting nominalization, passive overuse, A는 B structures, and literal calques, then iterates up to three passes naming each symptom removed.
Translationese is text whose tokens are grammatically valid in the target language but whose rhythm and structure follow English. Common tells include pronoun overuse, calqued connectives, forced passives, and literal idiom renderings.
It ships dedicated reference checklists for Korean, Japanese, Chinese, Arabic, Vietnamese, and Hindi. Other languages fall back to a language-agnostic symptom list, and new languages can be added by dropping a references/<lang>.md file.
No. The Skill explicitly excludes English output, pure code, single short phrases, and proper-noun lookups. It is a polish pass for non-English prose longer than a sentence or two.
It returns a strict two-part output: the naturalized text and a change log flagging every condensed, dropped, or glossary-mapped item as meaning-affecting or not. The calling agent must compare meaning and reject the result on any discrepancy.
Adjective instructions like write naturally do not reliably reduce translationese and can worsen it. Reframing the task as a separate rewrite-as-a-native-speaker pass, with a symptom checklist and named feedback loops, measurably improves naturalness.
thktExtract domain terms and definitions from Slack conversations into a glossary.
wimpysworldStructures public-facing comments, reviews, and pull request descriptions within strict length budgets.
githubnextCorrect speech-to-text errors in gh-aw documentation and workflow files.
Canhada-LabsTranslate documents with domain expertise, quality assurance, and cultural adaptation.
