Paper Polish

Problem

Paper text can look machine-generated because of repeated phrasing, formulaic connectors, and neat parallel structures. It can also trip similarity checks. Rewriting it manually requires preserving core arguments, data, technical terms, and citations while making the register sound natural, then producing a submission-ready docx file.

How it works

paper-polish breaks the task into three auditable stages:
- Input parsing: accepts pasted text, .txt paths, or .docx paths, and confirms output format and reduction intensity; defaults to standard when unspecified.
- Plagiarism reduction: uses sentence restructuring, context-aware synonym replacement, voice switching, reordering, and paragraph regrouping to lower similarity while preserving technical terms, proper nouns, formulas, and citation marks.
- Humanization: checks and fixes patterned AI traces such as filler phrases, mechanical connectors, rule of three, and empty evaluative statements; for Chinese it targets phrases like “值得注意的是” and “综上所述”, while for English it checks delve, landscape, Additionally, and similar markers.
- Word output: runs generate_docx.py to generate a .docx file with headings, paragraphs, lists, CJK/English fonts, margins, and line spacing.
Longer documents are processed paragraph or section by section, then checked for coherence before final export.

Boundaries

This skill is intended for academic writing assistance. Reduction is not a substitute for proper citation, and final output must satisfy institutional academic integrity requirements. Pre-export checks include unchanged core claims, intact data and citations, preserved terminology, no obvious AI patterns, natural academic register, and a valid .docx. Review the generated document manually before submission.