Novel and Webnovel Proofreading and Auto Formatting

Problem

Novels and webnovels often mix mechanical formatting issues with language-quality issues: inconsistent full- and half-width punctuation, missing first-line indent, stray blank lines, uneven chapter headings, custom watchlist hits, typos, word misuse, and awkward sentences. Long manuscripts are harder to proofread in one pass because context limits make volume-by-volume quality control difficult.

How It Works

The skill separates deterministic checks from language understanding:
- scripts/proofread.py uses the standard library for mechanical checks and formatting: custom word detection, punctuation and full/half-width normalization, spacing and blank-line cleanup, first-line indent, and chapter-heading normalization, producing mechanical issue JSON and mechanical proofreading reports.
- AI then reviews text segment by segment for typos, word misuse, and sentence issues, reporting location, type, and suggested fixes.
- For long works, pipeline can split volumes, run mechanical proofing per volume, merge the result, and export an EPUB-ready format; wordlist.txt can be used for platform-specific watchlist terms, with hits flagged but not deleted by default.

Boundaries

Supported inputs include .txt, .md, .html, .docx, .epub, and .rtf; formatted output from non-docx sources is unified as .txt to avoid breaking original markup. Semantic proofing is advisory, watchlist hits need manual review, and complex docx elements such as tables, headers, footers, and comments are not preserved during proofing.