Convert Word to Markdown: A DOCX Workflow and QA Checklist

Use a Markdown copy when you need searchable text, a reviewable source for summarization, or a lightweight batch of documents to compare. It is a working representation—not a replacement for the original Word file.

If acceptance depends on exact pagination, visual layout, signatures, comments, or tracked changes, keep the DOCX and review it in Word or a rendered PDF. Do not treat a clean-looking Markdown file as proof that every Word feature survived.

The specific convert-word-to-md package is a GitHub Copilot Agent Skill, not a browser upload service. Read its profile before installing and use its platform-specific install command.

Confirm that each input is .docx. The pinned workflow does not handle legacy .doc; save that file as DOCX first. In a mixed folder, explicitly account for PDFs and workbooks too—the Word script intentionally skips other file types.

The script creates Markdown and image copies. Confirm the agent environment is approved for the document, and protect the generated folder with the same access and retention rules as the original.

• List the input files and formats before a batch run.
• Decide where the output should live; by default it is created beside each source file.
• Use a fresh output destination if you need to keep a prior Markdown edit.

The upstream setup requires Python 3.10 or newer, pip, and the MarkItDown DOCX dependency. Run the setup from the Skill directory and verify the import in the same Python environment that will run the converter. The requirements file sets a lower bound, not an exact version lock.

These commands prepare the converter dependency; they do not install the Agent Skill. Use the copyable install command on the Skill profile for GitHub Copilot.

Run the bundled script from the Skill directory. Replace the example paths with the actual input path. A single file gets a same-named output folder; recursive mode finds DOCX files in nested folders and keeps their relative structure under an explicitly chosen batch destination.

A document with embedded pictures produces real image files under img/ and relative links from the Markdown. Share or move the entire output folder, not only the .md file.

• The first command uses the source-adjacent default output.
• Use -o only when a separate destination is intentional; inspect that directory before running.
• For batch jobs, review each reported failure separately instead of treating partial success as a complete conversion.

Conversion is preparation, not interpretation. Check the result against the source before asking an agent to summarize, compare, or extract consequential information.

• Compare heading order, paragraph sequence, lists, and key table labels and values.
• Open every relative image link and check that each extracted picture belongs to the expected place in the document.
• Review headers, footers, footnotes, comments, tracked changes, page breaks, and complex layout in the original; this small check did not validate them.
• Confirm the output contains only the files you intend to retain or share.

Source inspection found that the pinned script deletes an existing destination directory before writing a fresh conversion. That could remove hand-edited Markdown or other files stored there. The deletion behavior was not exercised in SkillSignal's sample check, so back up prior work or choose a new destination before rerunning.

This is a source-code finding, not a claim that SkillSignal tested destructive behavior. Read the implementation and confirm the destination before you run it on real files.

On October 14, 2026, SkillSignal ran the pinned Python script with two synthetic DOCX files and MarkItDown 0.1.8. The sample check verified selected headings, list text, two table values, image order and bytes, a text-only file, and unchanged source hashes.

It did not test GitHub Copilot loading, installation, recursive folders, complex tables, pagination, comments, tracked changes, protected files, customer documents, or security. It is not a benchmark or a general accuracy guarantee. The upstream dependency accepts markitdown[docx]>=0.1.0, so a new environment may resolve a different version.