[Bug]: WordParser drops soft line breaks and tabs inside DOCX paragraphs

[Xscaperrr]

Prerequisites

• I have searched the existing issues and discussions, and this is not a duplicate.
• This is a bug, not a usage question. (For questions, please use Discussions instead.)

Background / Description

WordParser.parse() silently removes soft line breaks and tabs inside a
nonempty DOCX paragraph. The surrounding text is concatenated without a
separator, corrupting the content passed to RAG chunking and indexing.

A soft line break, such as one inserted with Shift+Enter in Word, stays
inside the same paragraph. It is represented by a w:br element; a tab is
represented by w:tab. The reproduction below creates both using python-docx
and confirms that they survive saving and reopening the DOCX.

Expected: preserve the internal separators:

'Name\tAlice\nDepartment\tEngineering'

Actual: the fields and lines are joined together:

'NameAliceDepartmentEngineering'

This concerns separators inside a single paragraph, rather than blank lines
between separate paragraphs.

Error Messages

Expected: 'Name\tAlice\nDepartment\tEngineering'
Actual:   'NameAliceDepartmentEngineering'
AssertionError


The assertion is from the reproduction. The parser itself returns the altered
text without raising an error.

Steps to Reproduce

From a checkout of AgentScope, install the project and the Word parser dependency:

python -m pip install -e . python-docx

Save the following as repro.py and run python repro.py. It creates a real
DOCX in memory and requires no API key or external service.

import asyncio
import io

from docx import Document
from agentscope.rag import WordParser


async def main():
    expected = "Name\tAlice\nDepartment\tEngineering"
    document = Document()
    document.add_paragraph(expected)
    buffer = io.BytesIO()
    document.save(buffer)
    data = buffer.getvalue()

    # Check that the DOCX itself retains the tabs and soft line break.
    assert Document(io.BytesIO(data)).paragraphs[0].text == expected

    sections = await WordParser(include_image=False).parse(data, "example.docx")
    assert len(sections) == 1
    actual = sections[0].content.text
    print(f"Expected: {expected!r}")
    print(f"Actual:   {actual!r}")
    assert actual == expected


asyncio.run(main())

Environment

• AgentScope: 2.0.8, editable install at 083cbd1975c7b5ed055c69b0d19c150727e7f606.
• Python: 3.11
• python-docx: 1.2.0.
• OS: Windows.
• On that checkout, the existing WordParser tests pass:
  python -m pytest tests/rag_parser_test.py -k WordParser -q → 14 passed.
  The reproduction above fails.
• The same reproduction also fails when loading _word.py from PR fix(rag): keep text inside nested Word tables #2759's
  final commit (4a16631de82fce3e3d15fa6cc059422ccc6b4107) and upstream main
  (a38821287f35e9e45ed193d9d864cb46f263c946, checked on 2026-09-25) against
  the local environment. These additional checks exercised the parser file,
  not the full test suite of those revisions.

[Xscaperrr]

/assign

[github-actions]

This issue is still being verified. Please comment again once it is confirmed, and it is yours.

[Xscaperrr]

/assign

[github-actions]

It's yours, @Xscaperrr. If there is no pull request and no word from you by 2026-10-09 the claim is released so nobody is blocked — a comment here renews it.