How to use the Markdown to compliant Word agent skill

What the nature-paper2word skill makes the agent do

When installed, the agent converts a research manuscript from Markdown into a Word document. The output follows a declared format: typography, styled headings and captions, native Word equations, three-line tables, cross-references, page numbers that restart by section, and a bibliography in a citation style such as GB/T 7714, APA, Nature or IEEE.

The engine runs checks before and after export. Structural problems are reported with line numbers instead of silently producing a messy file. The skill outputs only Word. It does not produce PDF, HTML or LaTeX, and it is not meant for slide decks or grant forms.

Install the nature-paper2word skill

Download the ZIP of the skill from this page. Then install it in one of two ways. For Claude Code, unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for that project only. In the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.

The skill is documentation and instructions. The actual engine is a Python package, which you install separately. Install it with pip, then confirm the md-to-word command works. The command must print a version before the first export in a session.

• Requires Python with the md-to-word-docx package.
• If md-to-word --version fails, run the pip install again and check again.

Ask for it with a natural request

You do not need special syntax. Describe the outcome you want: a submission-ready .docx, a thesis layout, or a readable copy of a manuscript. Words like 論文转 Word, 学位论文排版, 期刊投稿排版 and prepare submission docx also trigger the skill. Name the target journal or school if you have one, and say which citation style you need.

If you have no detailed specification, say so. The skill then uses a built-in preset.

• Export this manuscript as the Word file our graduate school requires.
• Typeset it to this journal template, number equations per chapter, and use GB/T 7714 for references.
• Export the same manuscript once in a general format and once in the college format.
• Make the equations and tables editable in Word, not images.

Prepare the manuscript and format file

The manuscript must follow the skill's input contract. Headings use #, ## and ### without skipping levels. Paragraph styles are set with a line that starts with %% followed by the style name, for example a Table Caption or Image Caption line. Title, author, abstract and keywords are written as style lines in the body, not as YAML. YAML is used only for bibliography, csl, crossref and link-citations settings.

Citations use [@key] or @key and need a .bib file. Without a .bib file, a manuscript with citations cannot be exported. Pasted text, .txt, .md, old .docx and PDF sources must first be converted to Markdown. The skill does not edit your original file.

Formatting comes from a requirements file. Without one, the skill looks for a file named exactly format.txt next to the manuscript. If none exists, Chinese manuscripts use the zh-academic preset and other manuscripts use en-nature.

• Use {{< toc >}} for a table of contents and {{< pagebreak >}} for page breaks.
• Equations are written as $$...$$ {#eq-name} and referenced with @eq-name.
• Do not use HTML tags, horizontal rules or percentage image widths. The checker blocks them with a suggested replacement.

Export the manuscript

Run the command from the manuscript's folder. Without -r, the built-in preset or format.txt is used. With -r, you pass your own requirements file. Use md-to-word --contract to print the input contract and the paths to the manual and presets.

The pipeline runs pre-flight validation, creates the style template, builds the manuscript model, writes the Word XML, checks number integrity and then audits the output line by line against the requirements file.

• Blocking errors stop the export and must be fixed at the reported line. Do not use --no-check to skip them.
• Warnings are visible but not compliant, such as an image wider than the text block or a skipped heading level.

Safety notes: scripts and network

The skill runs scripts. The skill itself does not use the network. The one step that downloads from the internet is installing the md-to-word-docx engine from PyPI with pip. Review what you install and run the skill on manuscripts you are comfortable processing locally.

The skill's files are agents/openai.yaml, manifest.yaml, README_EN.md, README.md, SKILL.md and UPSTREAM-LICENSE.txt. Read SKILL.md first if you want to see exactly what the agent is told to do.

• Runs scripts: yes.
• Uses network: no (apart from the pip install of the engine).