PDF to Word Review

PDF → Word Conversion + Automatic Quality Review

Overview

Convert a PDF into an editable Word document, then automatically audit the conversion — reconciling character count, table count, image count, and table-cell count, turning "did the conversion drop anything?" from a feeling into evidence. That is the differentiator versus plain PDF converters: they only convert; this one also tells you where the conversion may have gone wrong.

The underlying engine is the standalone open-source project pdf2word-review (AGPL-3.0, GitHub: https://github.com/xiyanjun/pdf-to-word-review ). This skill is a thin wrapper: it calls the tool and interprets results, and does not vendor any engine code.

Installation

Check whether the CLI is available first:

pdf2word --help >/dev/null 2>&1 && echo "installed" || echo "need install"

If missing, install (GitHub direct install works even before the PyPI release):

# Option 1: PyPI (after official release)
pip install pdf2word-review

# Option 2: GitHub direct install (works today, recommended)
pip install git+https://github.com/xiyanjun/pdf-to-word-review.git

Only four lightweight core dependencies are pulled in automatically (pdf2docx, PyMuPDF, python-docx, lxml). OCR is optional — see "Scanned PDF OCR".

Workflow

Step 1: Confirm the input

• Verify the user's path exists and is a .pdf.
• Batch mode: multiple PDFs are supported in one pass.

Step 2: Convert (auto-routed)

pdf2word  -o  --verify --html

The CLI detects the text layer and routes automatically:

Key flags:

Step 3: Quality review (the differentiator)

--verify runs a four-way reconciliation, each item reporting PDF value → Word value delta and a status:

Page count is reference-only (Word pagination varies with layout).

Step 4: Interpret and report

Always translate the result into plain language — never dump raw JSON at the user. See "Result Interpretation".

Result Interpretation

Each item's status is one of four values:

Report template (conclusion first, then detail):

✅/⚠️ Conversion result: {passed / N items to review}

| Item | PDF | Word | Delta | Verdict |
|------|-----|------|-------|---------|
| Chars | ... | ... | ... | ✅/❌ |
| Tables | ... | ... | ... | ✅/❌ |
| Images | ... | ... | ... | ✅/❌ |
| Table cells | ... | ... | ... | ✅/❌ |

{list each loss/extra with risk note and fix suggestion}

Key points:

• Any loss or extra must be called out by name — never say "might have issues" vaguely.
• When converted appears, proactively explain it is expected behavior to avoid false alarm.
• If --html was used, give the user the HTML report path (visual diff is clearer than text).

Scanned PDF OCR

Scanned PDFs (no text layer) are auto-routed to OCR rebuild. OCR has two backends, both optional, neither installed by default:

• macOS user processing scans → install Vision.
• Non-macOS → install PaddleOCR.
• Without OCR deps, scanned conversion errors out with an install hint — relay that hint to the user honestly.
• After OCR rebuild, char/image reconciliation does not apply (source has no text layer); rely on the OCR confidence report instead — flag text blocks with confidence < 0.7 for manual review.

Notes

1. Layout polish requires macOS: banner OCR + list-item fixes need system Vision/Quartz. On non-macOS, add --no-polish.
2. AGPL-3.0 boundary: this skill is a thin wrapper — it only calls the external package and vendors no AGPL code. To change engine logic, open a PR upstream (GitHub) rather than copying code into the skill.
3. Batch: pdf2word a.pdf b.pdf c.pdf --verify --report result.json writes to each file's directory; --report yields a result array.
4. Char-count tolerance: <5% extra chars is normal (symbol substitution/formatting); >5% flags redundancy. >2% char loss raises a separate warning.

License

Thin wrapper. The underlying pdf2word-review engine is AGPL-3.0. See https://github.com/xiyanjun/pdf-to-word-review