Publisher DOCX Skill

Document Conversion & Editing Toolkit

This skill provides bidirectional conversion between DOCX/PDF and Markdown, plus DOCX editing capabilities:

• DOCX/PDF → Markdown: Convert documents to clean, reader-friendly Markdown
• Markdown → DOCX: Create properly-styled Word documents from Markdown
• DOCX Editing: Modify existing DOCX files (find/replace, append, insert)

[!IMPORTANT] Deep-interview intake comes first. When this skill is used, do not begin conversion, editing, linting, or template selection until you first ask the user one concise, deep-interview-style ambiguity-reduction question. The question must clarify: (1) the intended document purpose/audience, (2) desired output files (.docx, .pdf, Markdown, lint report, edited DOCX), (3) styling target or preset (elsevier, acs, ieee, nature, korean-generic, ko-executive-report-*, custom), (4) source file location and where outputs should be saved, and (5) hard constraints such as language/locale, journal or executive-report tone, deadline, font/color preferences, and what must stay out of scope.

Suggested first question:

"Before I touch the files, what is the document's purpose/audience, which output files do you want, which style or preset should it follow, where are the source/output paths, and are there any hard constraints on language, fonts, colors, publisher/report tone, deadline, or out-of-scope changes?"

If the user has already provided all of that information explicitly, restate the resolved brief in one sentence and proceed. If the user explicitly says to skip clarification, proceed but note the assumed preset/output choices before running commands.

[!IMPORTANT] If the user asks for an Elsevier / ACS / IEEE / Nature / korean-generic / ko-executive-report manuscript or report, do not start with the basic Markdown→DOCX path. Start with:

python scripts/convert_md_to_docx.py draft.md out.docx --publisher <profile>
python scripts/slop_lint.py draft.md --publisher <profile> --report report.md

Use the plain convert_md_to_docx.py input.md output.docx command only when no publisher-specific workflow is requested.

[!CAUTION] > Do not write your manuscript files into this skill folder. Your input .md and output .docx/.pdf files belong in your working directory. Run scripts using absolute paths to this skill folder, e.g.:

python /path/to/publisher-docx-skill/scripts/convert_md_to_docx.py ./input.md ./output.docx

The templates/, references/, and scripts/ subdirectories inside this skill are intentionally extensible and are maintained by the skill author. Running scripts/generate_docx_templates.py regenerates templates/docx/*.docx from templates/registry.yaml; this is the supported way to update publisher styles.

When to Use

• Need production-grade Markdown from DOCX or PDF for docs, LLM context, or web publishing.
• Want to create DOCX documents from Markdown with proper styling.
• Need to edit existing DOCX files programmatically (batch replacements, content insertion).
• Documents include tables, images, footnotes, math, or tracked changes.
• Preparing a publisher-styled manuscript from Markdown? Prefer convert_md_to_docx.py --publisher ... over the plain legacy Markdown→DOCX path.
• Want a non-destructive manuscript quality check before submission? Run scripts/slop_lint.py after the publisher DOCX workflow.

Recommended workflow for submissions

If the user asks for a journal-ready manuscript, use this order:

1. python scripts/convert_md_to_docx.py draft.md out.docx --publisher <profile>
2. python scripts/slop_lint.py draft.md --publisher <profile> --report report.md
3. Optional TeX-enabled path: python scripts/convert_md_to_pdf.py draft.md out.pdf --publisher <profile>

Use the plain convert_md_to_docx.py input.md output.docx flow only when no publisher-specific styling is requested.

Prerequisites

• Core: python3
• DOCX → MD: pandoc
• PDF → MD: pymupdf (pip install pymupdf)
• MD → DOCX: python-docx, markdown, beautifulsoup4 (pip install python-docx markdown beautifulsoup4)

DOCX → Markdown Workflow

1. Convert with pandoc (stable tables, extracted media):

pandoc "input.docx" \
  -f docx -t gfm \
  --wrap=none \
  --extract-media="media" \
  --reference-links \
  --markdown-headings=atx \
  --output "draft.md"

• Tracked changes: add --track-changes=all (or accept/reject first).
• If table lines wrap, add --columns=200 and re-run.

2. Clean the Markdown:

python scripts/clean_markdown.py draft.md output.md

3. Review with the checklist: references/markdown_quality_checklist.md

4. (Optional) Inline images for self-contained MD:

python scripts/embed_images.py output.md output.inline.md --root media

PDF → Markdown Workflow

pip install pymupdf  # if not installed
python scripts/convert_pdf.py input.pdf output.md

This handles multi-column layouts, paragraph merging, and header/footer removal.

Journal Manuscript Workflow (NEW)

Generate a publisher-styled manuscript directly from a Markdown draft using the --publisher flag. The skill ships a data-driven style registry at templates/registry.yaml and a generated DOCX template at templates/docx/<publisher>.docx. Supported publisher profiles are elsevier, acs, ieee, nature, korean-generic, ko-executive-report-core, ko-executive-report-navy, and ko-executive-report-forest.

python scripts/convert_md_to_docx.py input.md output.docx --publisher elsevier
python scripts/convert_md_to_docx.py input.md output.docx --publisher acs
python scripts/convert_md_to_docx.py input.md output.docx --publisher ieee
python scripts/convert_md_to_docx.py input.md output.docx --publisher nature
python scripts/convert_md_to_docx.py input.md output.docx --publisher korean-generic
python scripts/convert_md_to_docx.py input.md output.docx --publisher ko-executive-report-core
python scripts/convert_md_to_docx.py input.md output.docx --publisher ko-executive-report-navy
python scripts/convert_md_to_docx.py input.md output.docx --publisher ko-executive-report-forest

YAML frontmatter contract

Put a YAML block at the top of your Markdown. The converter uses it to build the title/author/affiliation/abstract/keyword block per the publisher's design.title_block rules.

---
title: "Catalytic CO2 reduction on Cu/ZnO: a mechanistic study"
authors:
  - { name: "Woojin Go", affiliation: 1, corresponding: true }
  - { name: "Jane Doe",  affiliation: [1, 2] }
affiliations:
  1: "Department of Chemical Engineering, University X"
  2: "Advanced Materials Institute, University Y"
abstract: "We report ..."
keywords: [CO2 reduction, Cu/ZnO, DFT, operando XPS]
corresponding_email: "wj@example.ac.kr"
---

Markdown → DOCX Workflow (Basic / non-publisher)

Use this path only when the user does not need publisher-specific styling.

python scripts/convert_md_to_docx.py input.md output.docx

Features:

• Headings (H1-H6 → Word heading styles)
• Bold, italic, inline code formatting
• Bullet and numbered lists
• Tables with proper styling
• Code blocks (monospace font)
• Links (as hyperlinks)
• Local images

With template:

python scripts/convert_md_to_docx.py input.md output.docx --template template.docx

What --publisher does automatically

When --publisher is set, the following anti-slop filters run on your text before and after conversion:

• Micro-typography: straight "..." → curly "...", 10-15 mg → 10–15 mg (en-dash + non-breaking space), word - word → word — word, ... → …, Fig. 1 / Table 1 / Eq. 1 → with non-breaking space.
• Heading auto-numbering: Heading 1 → 1., Heading 2 → 1.1., etc.
• Booktabs tables: python-docx's Table Grid (vertical rules) is replaced with top/header-bottom/bottom horizontal rules only.
• Title block injection from YAML frontmatter (superscript affiliations, corresponding-author marker, abstract label, keywords line).
• CJK font pairing: every run gets w:eastAsia=Pretendard so Hangul in the source renders as Pretendard rather than the default system CJK font.
• Banned-font / banned-style stripping: Calibri, Cambria, Aptos, Table Grid, Intense Quote, List Paragraph, Subtle Emphasis are removed if they leak in.
• Fenced code blocks and inline backtick code are preserved byte-for-byte.

Locale auto-detection

python scripts/convert_md_to_docx.py input.md output.docx --publisher elsevier --locale auto

--locale defaults to auto and detects Korean via Hangul codepoint ratio (>5% → ko, 1–5% → mixed, else en). The Korean-specific lint rules are scheduled for M6 of the journal enhancement work. The korean-generic profile is available now for mixed KR/EN manuscripts that need explicit Pretendard-first typography.

Regenerating templates

After editing templates/registry.yaml, rebuild the generated .docx files:

python scripts/generate_docx_templates.py                   # all publishers
python scripts/generate_docx_templates.py --publisher elsevier
python scripts/generate_docx_templates.py --publisher acs
python scripts/generate_docx_templates.py --publisher ieee
python scripts/generate_docx_templates.py --publisher nature
python scripts/generate_docx_templates.py --publisher korean-generic
python scripts/generate_docx_templates.py --publisher ko-executive-report-core
python scripts/generate_docx_templates.py --publisher ko-executive-report-navy
python scripts/generate_docx_templates.py --publisher ko-executive-report-forest

Adding a new publisher is a data-only change: copy an existing block in templates/registry.yaml, edit the values per the publisher's author guide, and rerun the generator.

LaTeX / PDF path

Generate a submission-style PDF via pandoc + XeLaTeX:

python scripts/convert_md_to_pdf.py input.md output.pdf --publisher elsevier
python scripts/convert_md_to_pdf.py input.md output.pdf --publisher korean-generic

This path uses publisher-specific templates under templates/latex/<publisher>/ and checks for the required TeX Live class before compilation. If xelatex, kpsewhich, or the required class is missing, the script exits non-zero with a clear install hint instead of failing halfway through compilation.

Anti-slop lint

Run a non-destructive prose lint pass before submission:

python scripts/slop_lint.py input.md --publisher elsevier --report report.md
python scripts/slop_lint.py input.md --publisher nature --report report.md --fix-whitespace

Current rule coverage includes banned AI-tell phrases, hedging pileups, stock openers/closers, citation-density heuristics, section-order checks, heading-case checks, qualifier/adjective-stack warnings, and a safe whitespace/typography autofix mode that writes a .fixed.md sibling.

Publisher Registry

The publisher registry lives in templates/registry.yaml.

• Edit the YAML to add or refine a publisher profile.
• Regenerate .docx templates with scripts/generate_docx_templates.py.
• The DOCX, lint, and optional PDF paths all consume the same registry.
• This keeps publisher support data-driven rather than hard-coded.

Edit Existing DOCX (NEW)

Modify existing DOCX files with common operations:

Find & Replace

python scripts/edit_docx.py input.docx --replace "OLD_TEXT" "NEW_TEXT" -o output.docx

Add --case-insensitive for case-insensitive matching.

Append Markdown Content

python scripts/edit_docx.py input.docx --append-md extra_content.md -o output.docx

Insert at Placeholder

python scripts/edit_docx.py input.docx --insert-at "{{MARKER}}" "Replacement text" -o output.docx

Extract Text

python scripts/edit_docx.py input.docx --extract-text

Show Document Structure

python scripts/edit_docx.py input.docx --show-structure

Quality Checklist

• Headings: correct hierarchy, single blank line before, none empty.
• Tables: blank line before/after; merged cells rewritten if needed.
• Lists: consistent indentation; avoid hard wraps mid-bullet.
• Images: relative to media/; add alt text; widths removed.
• Math/footnotes: render and resolve; definitions present.
• Spacing: no trailing spaces; ≤2 consecutive blank lines.

References