The plugin ecosystem of DeepSeek Harness (DSH) supports extending a wide range of capabilities. In office scenarios, directly converting AI-generated content into a well-formatted Word document is a common requirement. The dsh-word-docs plugin provides this capability.

This is an office Word document plugin for DeepSeek Harness. It is maintained by Ei-Ayw. The core feature allows DSH to output a formal document with a single request.

Core Capabilities

Generation, Editing, Reading, and Statistics

The plugin supports generating, editing, reading, and computing statistics for documents, as well as converting documents to PDF. The underlying implementation uses only the Python standard library (zipfile + xml.etree) and has no third-party dependencies.

Chinese Typography Optimization

It automatically handles Chinese fonts (SimSun/eastAsia) and supports the layout of formal official documents, including first-line indentation and Chinese-style heading numbering.

Standard Plugin Interface

As a standard DSH plugin, after installation it registers the word_docs tool and the word-docs skill.

JSON Specification Definition

It supports defining documents through a JSON specification, including headings, body text (bold/italic/code), nested lists, tables (supporting merged cells and column width settings), images, and page breaks.

Markdown Conversion

It supports converting Markdown to Word, compatible with ATX headings, pipe tables, code blocks, blockquotes, links, and other formats.

Document Editing

It supports editing existing documents, including replacing text (preserving formatting), appending content, deleting paragraphs, bulk-filling templates (such as {placeholder}), and applying formatting.

Installation

dsh plugin --profile web add git+https://github.com/Ei-Ayw/dsh-word-docs.git

Typical Usage

Convert Markdown to Word via the command line:

python3 scripts/worddocs.py md2docx --input demo/sample.md --output demo/meeting.docx --page-number --first-line-indent --number-headings chinese

Generate a document via a JSON specification:

{
  "page": { "footer": { "pageNumber": true } },
  "headingNumbering": "chinese",
  "content": [
    { "type": "heading", "level": 0, "text": "2026 年第 35 周工作周报", "align": "center" },
    { "type": "paragraph", "text": "完成 **API 网关** 限流改造,压测 QPS 提升 40%" },
    { "type": "table", "colWidths": [3.5, 3, 3, 3],
      "headers": ["指标", "本周", "上周", "环比"],
      "rows": [ ["需求交付数", "5", "4", "+25%"],
                [{ "text": "合并列", "colSpan": 2 }, "3", "-57%"] ] }
  ]
}

Notes

  • Static numbering: Heading numbers are static numbers written in at generation time, not Word automatic numbering fields.
  • Feature scope: Images support png/jpg/gif; headers and footers are set at generation time; footnotes/table of contents fields/watermarks are not supported.
  • Local installation: When installing locally, pay attention to peer dependency resolution and run ln -s ~/.dsh/profiles/node_modules/@deepseek-ai node_modules/@deepseek-ai.

Ecosystem Background

The DSH philosophy is “everything is a plugin.” The community catalog is an independent site and has no official affiliation with DeepSeek / Huanfang.

The plugin lowers the barrier to document generation through its standard interface and pure Python implementation. For more details, see the catalog page and the GitHub repository.