Office Document Assistant

Read, extract, summarize, and compare common office documents:

• - PDF
• Word (.docx, .doc)
• Excel (.xlsx, .xls)
• PowerPoint (.pptx, .ppt)

Use this skill when the user wants the contents of a document explained, summarized, searched, or extracted into a simpler structure.

When to Use

Use this skill when the user:

• - uploads a .pdf / .doc / .docx / .xls / .xlsx / .ppt / INLINECODE12
• asks to summarize a document
• asks to extract dates, amounts, contacts, conclusions, specifications, risks, or action items
• asks for page-by-page / slide-by-slide structure
• asks what a spreadsheet or slide deck is saying
• asks to compare two or more documents after extracting their text

When Not to Use

Do not position this skill as a high-fidelity layout or visual analysis system.

It is not ideal for:

• - precise preservation of original layout, formatting, or pagination
• detailed chart / diagram / image interpretation
• password-protected or encrypted files
• OCR-heavy image understanding beyond basic text recovery
• advanced spreadsheet analytics or formula auditing
• tracked-changes / redline reconstruction in Office documents

Core Workflow

1. 1. Confirm the document path.
2. Run the bundled script:

- python3 {skill_dir}/scripts/extract_office_text.py <file> --json

1. 3. Inspect the JSON fields:

- type - extraction - warning - truncated - text

1. 4. Separate clearly in your response:

- directly extracted content - your summary / inference based on that content

1. 5. If extraction is empty or weak:

- for PDF, check OCR availability first - for legacy Office formats, check conversion tools

1. 6. If the user asks for a summary, default to:

- one-sentence overview - 3–8 key points - extra sections only when clearly present (dates, people, risks, data, conclusions, contacts)

1. 7. If the user asks for extraction, prefer structured fields over long prose.

Supported Formats and Strategy

PDF

• - First extract embedded text with pypdf.
• If extracted text is too short, fall back to OCR.
• OCR prefers chi_sim+eng, then chi_sim, then eng.
• OCR pipeline requires both pdftoppm and tesseract.
• If an official first-class PDF tool is exposed in the environment and the task is high-value or multi-PDF, you may prefer that tool; otherwise use this skill's script.

Word

• - .docx: extract paragraphs and tables directly.
• INLINECODE26: try antiword, then catdoc, then LibreOffice conversion to .docx.

Excel

• - Extract sheet names and the first rows of each sheet.
• Best for quickly understanding workbook structure and core fields.
• When explaining, focus on what each sheet represents, key columns, important figures, and obvious anomalies.

PowerPoint

• - Extract slide text from shapes.
• Extract speaker notes when present.
• Summaries should usually be slide-by-slide or theme-based, not a giant raw dump.

Tools and Dependencies

Document clearly what is required versus optional.

Required runtime

• - INLINECODE30

Required Python packages

• - pypdf — embedded text extraction from PDFs
• INLINECODE32 — .docx extraction
• INLINECODE34 — .xlsx extraction
• INLINECODE36 — .pptx extraction

Optional but strongly recommended system tools

• - poppler-utils — provides pdftoppm for PDF → image conversion before OCR
• INLINECODE40 — OCR engine
• INLINECODE41 — Simplified Chinese OCR language pack
• INLINECODE42 — conversion fallback for legacy .doc, .xls, INLINECODE45
• INLINECODE46 — direct .doc extraction fallback
• INLINECODE48 — additional .doc extraction fallback

What each tool is used for

• - pypdf: try text-layer extraction from PDFs first
• INLINECODE51: rasterize PDF pages when OCR is needed
• INLINECODE52: recover text from scanned/image PDFs
• INLINECODE53: read paragraphs and tables from INLINECODE54
• INLINECODE55: read sheets and rows from INLINECODE56
• INLINECODE57: read slide text and notes from INLINECODE58
• INLINECODE59: convert older Office formats into newer parseable formats
• INLINECODE60 / catdoc: lightweight extraction options for INLINECODE62

Minimum useful setup

If only modern documents matter, the minimum practical setup is:

• - INLINECODE63
• Python packages: pypdf, python-docx, openpyxl, INLINECODE67

Recommended full setup

For the most robust behavior across real-world files, install:

• - INLINECODE68
• Python packages: pypdf, python-docx, openpyxl, INLINECODE72
• system tools: poppler-utils, tesseract-ocr, tesseract-ocr-chi-sim, libreoffice, antiword, INLINECODE78

Dependency check

Use the bundled checker to quickly see what is missing in the current environment:

CODEBLOCK0

Common Commands

CODEBLOCK1

Useful flags:

CODEBLOCK2

Output Style

Default to a compact answer:

• - one-sentence summary
• 3–8 key points
• then expand only if the user asks for:

- detailed summary
- page-by-page / slide-by-slide notes
- field extraction
- document comparison

Failure Handling

• - If PDF text is empty, suspect scanned pages or missing OCR tools.
• If Chinese OCR is weak, check whether tesseract-ocr-chi-sim is installed.
• If .doc / .xls / .ppt extraction fails, check libreoffice, antiword, and catdoc.
• If tables look messy, explain that this is text-first extraction rather than full layout reconstruction.
• If a file is encrypted or unreadable, say so plainly and stop guessing.

References

Read these only when needed:

• - references/capabilities.md — capability boundaries and what each format can/can't do well
• INLINECODE87 — dependency checks and common failure modes