# DOCX

Create polished, fully-editable Word (.docx) documents end to end, research or read...

- Repository: https://github.com/cubeplexai/cubeplex (574 stars)
- Category: Documents
- License: see repository
- Last updated: 2026-11-01
- Runs scripts: yes, review before use

## Install

Claude Code: `npx skills add cubeplexai/cubeplex --skill docx -g -a claude-code -y`
Codex: `npx skills add cubeplexai/cubeplex --skill docx -g -a codex -y`
Cursor: `npx skills add cubeplexai/cubeplex --skill docx -g -a cursor -y`
Gemini CLI: `npx skills add cubeplexai/cubeplex --skill docx -g -a gemini-cli -y`

## Description

Create polished, fully-editable Word (.docx) documents end to end, research or read the source, outline the structure, pick a style set, build native editable content (cover / headings / body / lists / three-line tables / figures / quote / TOC / page numbers), then self-check before delivering. Handles English and Chinese (including 公文 / GB-T 9704 layout). Use whenever the user wants a report, paper, proposal, letter, memo, contract draft, or any Word document. Free fonts + open libraries only.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# docx, research, structure, and build self-checked Word documents

Don't jump straight to code. A good document is **content first, structure
second, formatting last**. Follow the pipeline; the depth is in the reference
docs, read the ones each step points to.

`<skill>` below = the absolute path `load_skill` returned. The builder is
`<skill>/scripts/docbuilder.py`; the checker is `<skill>/scripts/check_doc.py`;
reference docs live under `<skill>/references/`.

## Pipeline

1. **Understand the request & inputs.** Read every uploaded file in full first.
   Identify: language, audience, the document *type* (report / paper / proposal /
   letter / memo / 公文), any length/section requirements, and whether the user
   gave you a finished source, an outline, or just a topic.

2. **Choose the content mode**, where the content comes from
   (`references/content.md`):
   - **Summarize**, user gave a complete source → distill it (don't invent).
   - **Outline**, user gave structure/points → follow it, enrich each section.
   - **Research**, user gave only a topic → research the web first.

3. **Research (if needed).** Broad pass (map directions, tell the user what
   you'll dig into) → deep pass (read sources via the `browser` skill / `curl`,
   extract real data). Cite real sources; never fabricate. See `content.md`.

4. **Outline the document**, the section structure for the chosen type
   (`content.md` has skeletons): title block → (TOC) → sections (H1) with H2/H3 →
   conclusions/references/appendices as the type needs. One idea per section for
   reference docs; flowing prose for narrative ones.

5. **Pick one style set** for the whole document (`references/themes.md` catalog
   → read the chosen `references/themes/<name>.md`): `corporate` · `academic` ·
   `report` · `chinese` · `official` (公文). See `references/design.md` for
   profile→set and `references/styles.md` for the system. CJK specifics:
   `references/cjk.md`.

The rest: https://github.com/cubeplexai/cubeplex/blob/HEAD/backend/skills/preinstalled/docx/SKILL.md