Claude docx/ooxml skill makes Word files unreadable — how to fix?

Question

Using the Anthropic docx/ooxml skill in Claude to add a comment to a .docx file produced a file that Word and LibreOffice could not open, although Pandoc could still extract the content. The culprit was the skill writing the modified XML with extra whitespace. Is there a fix or workaround?

Originally asked on GitHub: https://github.com/anthropics/skills/issues/12

AgentArmorHQ team · Accepted Answer

**Status:** open — the issue is still open and no fix is merged into the skill, but a PR is linked (anthropics/skills#143) and the reporter confirmed the prompt workaround below fixes the corrupted files.

The root cause is XML serialization. When the docx/ooxml skill manipulates a document and writes XML manually (for example, after parsing with `defusedxml.minidom`), calling `toprettyxml()` adds newlines and indentation. Word is very sensitive to XML whitespace and interprets that extra whitespace as part of the document structure, so Word and LibreOffice reject the file — even though Pandoc can still extract its content.

The reporter's fix is to add this guidance to the skill's prompt (or apply it yourself when scripting the skill):

- Always serialize with `doc.toxml(encoding='utf-8')` — preserves the original compact formatting Word expects.
- Never use `doc.toprettyxml()` — it breaks Word compatibility.

Correct:

```python
with open(doc_path, 'wb') as f:
    f.write(doc.toxml(encoding='utf-8'))
```

Wrong:

```python
with open(doc_path, 'wb') as f:
    f.write(doc.toprettyxml(indent='  ', encoding='utf-8'))
```

The reporter also shared the kind of prompt that triggers the bug: asking the skill to create a docx and then add a comment on a line in several paragraphs. Use that pattern to test whether your setup still produces valid files after changes.

As general advice: after any docx modification, open the result in Word or LibreOffice before relying on it, since a corrupt file may still be parseable by Pandoc.

AgentArmorHQ team · Answer

**What a GitHub user tried on GitHub** (Didn't work)

Ran the docx skill with a prompt asking it to create a uv guide in docx format and then add a comment on a line in several paragraphs. The resulting file was unreadable by Word and LibreOffice, though Pandoc could still extract the content.

Source: https://github.com/anthropics/skills/issues/12

AgentArmorHQ team · Answer

**What @thejud tried on GitHub** (Worked)

Updated the ooxml skill prompt to forbid `toprettyxml()` and require `doc.toxml(encoding='utf-8')` when manually writing XML back to the file. The reporter states that updating the prompt fixes the invalid-document issue.

Source: https://github.com/anthropics/skills/issues/12#issuecomment-3413286304

AgentArmorHQ team · Answer

**What @cdxiaodong tried on GitHub** (Outcome unclear)

A pull request link was added to the thread (https://github.com/anthropics/skills/pull/143), presumably to incorporate the guidance into the skill. No merge status or review info is given.

Source: https://github.com/anthropics/skills/issues/12#issuecomment-3651641650

Claude docx/ooxml skill makes Word files unreadable — how to fix?

4 answers

Add your answer or comment

Claude docx/ooxml skill makes Word files unreadable — how to fix?

4 answers

Add your answer or comment

Related questions