A low-level OOXML processing tool designed to bridge the gap between AI agents and Microsoft Word documents through token-efficient Markdown conversion.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install docx-md
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install docx-md using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is docx-md?
docx-md is a specialized utility for AI document review and manipulation that interacts directly with OOXML, the underlying ZIP and XML structure of .docx files. It circumvents the need for heavy commercial Word libraries, providing a lightweight and high-performance solution for developers building Openclaw Skills. The tool creates a bidirectional bridge: it converts complex Word documents into a compact, token-efficient Markdown format for AI processing and subsequently applies AI-generated edits back into the document as native tracked changes and comments.
docx-md Use Cases
- AI-powered legal contract review and redlining where track changes are required.
- Automated technical documentation updates based on agent-generated content.
- Batch processing of Word documents to extract structured Markdown for LLM context windows.
- Collaborative AI workflows where model suggestions must be reviewed by humans in Microsoft Word.
How docx-md Works
- The tool parses the word/document.xml and word/comments.xml files within the DOCX container.
- Content is converted into Markdown or JSON, using blockIndex markers to allow precise targeting of edits.
- The structured output is sent to an LLM, which returns an edit JSON containing the proposed modifications.
- The modify script converts these JSON edits back into OOXML tags like w:ins and w:del.
- Revisions are injected back into the document, preserving the original formatting while adding track changes.
- A finalize step can be triggered to programmatically accept all revisions and remove all comments for a clean delivery version.
docx-md Setup
First, ensure you have the necessary dependencies installed:
pip install docx-revisions
To read a document and generate token-efficient Markdown:
python3 skills/docx-md/scripts/read_docx.py document.docx -o result.md
To apply edits generated by an AI agent:
python3 skills/docx-md/scripts/apply_edits_docx.py document.docx edits.json -o modified.docx
To accept all changes and finalize the document:
python3 skills/docx-md/scripts/finalize_docx.py modified.docx -o final.docx
docx-md Data Schema & Taxonomy
The docx-md skill uses a specific schema to ensure the AI can precisely modify document sections. Each block is identified by a blockIndex.
| Property | Description |
|---|---|
| blockIndex | A unique numeric marker (e.g., <!-- b:1 -->) used to target specific paragraphs. |
| originalContent | The raw text found in the document before any AI intervention. |
| content | The updated text provided by the agent to replace the original content. |
| basis | The reasoning or justification for the change, often used in comments. |
| modifications | An array containing the objects above for batch processing. |
docx-md Advanced Features
- Token-efficient Markdown output that includes block markers to minimize context window usage in Openclaw Skills.
- Direct OOXML manipulation allowing for track changes (redlining) without requiring a Microsoft Word installation.
- Support for custom revision author attribution during the modification phase.
- Automated 'Finalize' mode that strips metadata, removes comments, and flattens all revisions into a final state.
- Seamless integration with complex workflows such as multi-agent contract review pipelines.
SKILL.md
Loading
Related Openclaw Skills
Qwen3 TTS Instruct
A high-performance Text-to-Speech skill for AI agents using Alibaba Cloud Bailian Qwen with advanced emotional mood presets and real-time streaming.
Pain Point Finder
A data-driven Reddit research tool to discover and validate unmet user needs and startup ideas without requiring API keys.
MCP Server Discovery
A comprehensive utility for discovering, searching, and configuring Model Context Protocol (MCP) servers to extend the capabilities of AI agents.
Rate My Claw
A competitive benchmarking platform where AI agents complete specialized tasks to earn Elo ratings and visualize technical skill radars.
Code Search Skill
A high-performance toolkit for rapid codebase exploration using structured grep, glob, and directory tree visualization.
Code Search Skill
A high-performance search toolkit for exploring codebases via structured content grep, filename globbing, and directory tree visualization.
Featured*
