dsh-docx-export

English | 中文

Export generated text as a real Word document (.docx) from any DeepSeek Harness session — not a renamed Markdown file, but a proper OOXML package with native heading styles, multilevel lists, tables, code blocks, quotes, hyperlinks, embedded pictures, a table-of-contents field and a page-numbered footer. Word, WPS, LibreOffice and Google Docs all open it as-is.

Zero runtime dependencies (only Node's built-in zlib), no @deepseek-ai/* imports, so it survives harness version drift. It installs as a bundle and mounts on the next start.

You: turn that analysis into a Word document for me
DSH: (calls export_docx, writes G:\work\analysis-20251006-1209.docx)

Features

Plus: toc: true inserts a real TOC field (filled in by the word processor on open), page numbers in the footer by default, and title/author/timestamps in the document properties.

Install

Pick one route, then restart DeepSeek Harness — a new bundle mounts on the next start.

# 1) npm (recommended once published)
dsh plugin --profile desktop add dsh-docx-export

# 2) straight from GitHub
dsh plugin --profile desktop add github:Cai-Chengyu/dsh-docx-export

# 3) a local directory or a packed tarball
dsh plugin --profile desktop add G:\path\to\dsh-docx-export
dsh plugin --profile desktop add .\dsh-docx-export-1.0.0.tgz

If dsh is not on PATH, call the CLI shipped with the app: "<DeepSeek Harness>\resources\runtime\cli\bin\dsh.cmd". scripts/install.ps1 (Windows) and scripts/install.sh (macOS, Linux) wrap that command and verify the bundle got registered.

Manual alternative: pnpm add dsh-docx-export inside the profile directory, then append dsh-docx-export to dsh.profile.bundles in that directory's package.json.

Use

Just ask in the session — the model calls the tool for you:

Export what you just wrote as a Word document titled "Q1 analysis", with a table of contents.

Or call export_docx directly:

{
  "content": "# Title\n\nBody **bold**\n\n| col | value |\n|:--|--:|\n| a | 1 |\n",
  "title": "Q1 analysis",
  "outputPath": "G:\\dsh\\analysis.docx",
  "toc": true,
  "pageSize": "a4"
}

Tool parameters

The result text reports the absolute path plus paragraph, heading, table, image and size counts.

CLI (usable outside DSH)

node bin/docx-export.mjs report.md -o report.docx --title "Quarterly report" --toc
cat notes.md | node bin/docx-export.mjs - -o notes.docx --text

--page a4|letter|a5, --landscape, --no-footer, --font, --east-asia, --font-size are available too.

Configuration (optional)

Override defaults in the profile's cordis.patch.yml:

- id: docx-export
  name: dsh-docx-export
  config:
    toolName: export_docx          # rename on collision
    outputDir: "D:/exports"        # default output directory ("" = session working directory)
    fontFamily: Calibri            # Latin body font
    eastAsiaFont: 微软雅黑          # East Asian body font
    monoFont: Consolas             # code font
    fontSize: 11                   # body size in points
    pageSize: a4
    toc: false                     # insert a TOC by default
    footer: true                   # page numbers by default
    allowOutsideWorkspace: false    # may the tool write outside the session workspace
    allowedRoots: []               # extra directories that are allowed
    maxContentChars: 400000        # per-export body limit

Write scope (safe by default)

The tool reads and writes only inside the session working directory unless the profile says otherwise. An out-of-scope path is refused with an actionable message; grant specific locations with allowedRoots, or set allowOutsideWorkspace: true. Pictures are embedded only when they live inside the allowed scope.

Development

node test/markdown.test.mjs   # Markdown reader (12 tests)
node test/docx.test.mjs       # OOXML writer + ZIP integrity (8 tests)
node test/plugin.test.mjs     # tool registration, path policy, real files (11 tests)
npm test                      # = node --test test/ (used by CI)
npm run demo                  # convert examples/sample.md to examples/sample.docx

The suite ships its own ZIP central-directory reader with CRC verification, an XML well-formedness checker and a miniature PNG encoder, so there are no dev dependencies either.

Publishing and sharing

• Add the GitHub topic dsh-plugin so discovery sweeps pick the repository up.
• Submit it to the community catalog with a PR to data/plugins.json in (fields: that repo's data/schema.json).
• Publish to npm with npm publish, after which dsh plugin --profile <profile> add dsh-docx-export installs it from the registry.
• Details: docs/PUBLISHING.md.

Compatibility

• Node ≥ 20 (node:zlib, ESM).
• Any DSH from 0.1.7 on: the plugin declares no @deepseek-ai/* peer dependencies and imports no host package — it uses the runtime's raw JSON-Schema tool-registration path.
• Measured on DSH 0.2.0-rc.2 (Windows desktop): install → mount → generate → structural check → LibreOffice render.

License

MIT