speak-human-tw

Why we included this project

Anyone who writes Traditional Chinese copy that a real person will read has probably hit the problem this project targets: the text is factually right, but it reads like AI wrote it, and you cannot quite say why. Speak-human-tw scans a finished draft for more than three dozen recurring AI writing habits, from promotional filler openings and stacked emoji to lecture-style closings, and offers concrete rewrite suggestions instead of just shuffling words around. It also steers China-facing phrasing and half-width punctuation toward Taiwan conventions, which matters when your readers expect zh-TW. You load it into Claude Code, Codex, or Cursor as an agent skill, so it fits into a proofing step in your normal workflow rather than pretending to be a content generator. The project is refreshingly clear about that line: it flags issues and proposes edits, but you keep final say over tone and wording, which is what makes it useful for newsletter writers, marketers, and editors who want AI speed without the AI sound.