pdf-to-word-docx（PDF TO word DOCX 效率）

PDF to Word Converter

Purpose

• Wraps the ComPDFKitConversion Python SDK into a reusable local conversion workflow, supporting PDF / image to Word, PPT, Excel, HTML, RTF, Image, TXT, JSON, Markdown, and CSV (10 output formats in total).

Agent Skills Standard Compatibility

• This Skill uses an Anthropic Agent Skills-compatible directory structure: pdf-to-word-docx/.
• The entry point is SKILL.md; helper scripts are placed in scripts/.
• The document uses $ARGUMENTS and ${CLAUDE_SKILL_DIR} conventions for distribution and execution in Claude Code / Agent Skills-compatible environments.

Input / Output

• Input: The target format (word/excel/ppt/html/rtf/image/txt/json/markdown/csv), the PDF or image path, and the output path are passed via Skill arguments or the command line. An optional PDF password and conversion parameters may also be provided.
• Supported input file types:

- PDF files (.pdf) - Image files (.jpg/.jpeg/.png/.bmp/.tif/.tiff/.webp/.jp2/.gif/.tga)

• Output: A file in the corresponding format (.docx, .pptx, .xlsx, .html, .rtf, image, .txt, .json, .md, .csv), or a clear error message.

Prerequisites

• Supports Windows and macOS.
• The conversion SDK must be installed first:

  pip install ComPDFKitConversion

• On first run, the script automatically downloads license.xml from the ComPDF server and caches it in the scripts/ directory:

  https://download.compdf.com/skills/license/license.xml

• The script reads the <key>...</key> field from license.xml and uses that key for LibraryManager.license_verify(...) authentication — it does not pass the XML file path directly to the SDK.
• To use a custom license, place your own license.xml in the scripts/ directory; the script will use it directly without downloading.
• During SDK initialization, the resource directory is always set to the directory containing pdf-to-word-docx.py, i.e., the scripts/ directory itself.
• When --enable-ocr or --enable-ai-layout (enabled by default) is used, the Skill also requires scripts/documentai.model. If the file does not exist, the script will automatically download it from:

  https://download.compdf.com/skills/model/documentai.model

• To reuse an existing model file, you can override the default model path via an environment variable:

  export COMPDF_DOCUMENT_AI_MODEL="/path/to/documentai.model"

Workflow

1. Confirm the Python package is installed:

   python -m pip show ComPDFKitConversion

1. The script automatically downloads license.xml on first run; the scripts/ directory is used directly as the SDK resource path.
2. In Agent Skills / Claude Code environments, prefer using the Skill's built-in script path variable:

   python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx
   python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" ppt input.pdf output.pptx
   python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" excel input.pdf output.xlsx

1. For more control, append common parameters:

   python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" excel input.pdf output.xlsx --page-ranges "1-3,5" --excel-all-content --excel-worksheet-option for-page
   python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx --enable-ocr --page-layout-mode flow

1. On startup, the script ensures scripts/license.xml exists (downloading it automatically from the ComPDF server if missing), reads the <key> field for SDK authentication, and uses the scripts/ directory as the resource path.
2. If --enable-ocr or --enable-ai-layout (enabled by default) is active, the script checks whether scripts/documentai.model exists; if not, it downloads the file automatically before initializing the Document AI model.
3. Check the return code; if it is not SUCCESS, handle license, password, resource, model, or input file issues according to the error name.

documentai.model Download Optimization

• The script preferentially uses the model file pointed to by COMPDF_DOCUMENT_AI_MODEL.
• The default model path is scripts/documentai.model.
• During automatic download, the file is first written to documentai.model.part and then atomically renamed to the final file upon success, preventing partial file corruption.
• On download failure, the script retries automatically with back-off intervals of 2s / 5s / 10s.

Invoking Directly as a Skill

• In environments that support Agent Skills, the Skill can be called directly:

  /pdf-to-word-docx word input.pdf output.docx
  /pdf-to-word-docx excel input.pdf output.xlsx --excel-worksheet-option for-page

• When the Skill receives arguments, it passes them through to the script as-is:

  python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" $ARGUMENTS

• If the environment does not support direct Skill invocation, fall back to a regular command-line call.

Supported Output Formats

• word → calls CPDFConversion.start_pdf_to_word
• excel → calls CPDFConversion.start_pdf_to_excel
• ppt → calls CPDFConversion.start_pdf_to_ppt
• html → calls CPDFConversion.start_pdf_to_html
• rtf → calls CPDFConversion.start_pdf_to_rtf
• image → calls CPDFConversion.start_pdf_to_image
• txt → calls CPDFConversion.start_pdf_to_txt
• json → calls CPDFConversion.start_pdf_to_json
• markdown → calls CPDFConversion.start_pdf_to_markdown
• csv → reuses CPDFConversion.start_pdf_to_excel with table/Excel parameters to produce CSV-friendly output

Input Source Types

• The script supports PDF and image as input sources. The SDK's start_pdf_to_* interfaces natively accept image files with no pre-processing required.
• By default, the script auto-detects the input type from the file extension:

- .pdf → pdf - .png/.jpg/.jpeg/.bmp/.tif/.tiff/.gif/.webp/.tga → image

• You can also specify the source type explicitly:

  python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.png output.docx --source-type image

• image -> * and pdf -> * share the same set of CPDFConversion.start_pdf_to_* interfaces; only the input file type differs.

Smart Defaults

The script automatically adjusts certain parameters based on the input source and output format to reduce manual configuration:

When triggered, the script prints a notice to stderr, for example:

Auto-enabled OCR for image input.
Auto-set page layout mode to BOX for HTML output.

All Parameters

Positional Parameters

General Parameters

Layout Parameters

Content Retention Parameters

Output Control Parameters

OCR Parameters

Excel-Specific Parameters

JSON-Specific Parameters

TXT-Specific Parameters

HTML-Specific Parameters

Image-Specific Parameters

Parameter Default Value Rules

• Parameters that default to True (--enable-ai-layout/--contain-image/--contain-annotation/--contain-page-background-image/--auto-create-folder/--json-contain-table/--txt-table-format) use BooleanOptionalAction; pass --no-xxx to disable.
• Parameters that default to False (--enable-ocr/--formula-to-image/--transparent-text/--output-document-per-page/--excel-all-content/--excel-csv-format/--image-path-enhance) use store_true; passing the flag enables them.
• All CLI parameter defaults are fully consistent with the SDK's ConvertOptions() defaults — omitting a parameter is equivalent to using the SDK's original default value.

Recommended Command Examples

PDF to Word (default parameters, AI layout analysis enabled)

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx

PDF to Word, box layout, no images, no AI layout analysis

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx --no-enable-ai-layout --no-contain-image --page-layout-mode box

PDF to Word, retain annotations and background images, one document per page

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx --output-document-per-page

PDF to Excel, include all content and split worksheets by page

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" excel input.pdf output.xlsx --excel-all-content --excel-worksheet-option for-page

PDF to TXT, with table formatting enabled

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" txt input.pdf output.txt

PDF to HTML, multi-page with bookmarks mode

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" html input.pdf output_dir --html-option multiple-page-with-bookmark

PDF to Image, PNG format, grayscale, 2x scaling

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" image input.pdf output.png --image-type png --image-color-mode gray --image-scaling 2.0

Image to Word (OCR auto-enabled, specify Chinese language)

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.png output.docx --ocr-language chinese

Note: For image input, the script automatically enables OCR — there is no need to pass --enable-ocr manually. To specify an OCR language, --ocr-language can still be used.

PDF with OCR enabled (multiple languages)

python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" word input.pdf output.docx --enable-ocr --ocr-language chinese english japanese

Trial License and Usage Limits

• The scripts/license.xml auto-downloaded from the ComPDF server is a Trial License, allowing a maximum of 200 conversions.
• The script uses a SHA-256 fingerprint to detect whether the current License is the default trial key; no usage limit applies when using any other License.
• After each successful conversion using the trial License, the script prints the current used/remaining count to stderr, for example:

  Trial license: 5/200 conversions used, 195 remaining.

• When the trial limit is reached (200 conversions), the script refuses to convert and prompts the user to purchase a full License:

  Error: Trial license usage limit reached (200 conversions). Please purchase a license at: https://www.compdf.com/contact-sales

• When the trial License has expired (SDK authentication fails), the error message also includes a purchase link.
• After purchasing a full License, place a custom license.xml containing the new <key> in scripts/ (overwriting the auto-downloaded trial file) — no script modifications or counter file cleanup are required.

Confirmed Facts

• ComPDFKitConversion has been successfully installed on the local machine.
• The installed package provides 10 conversion methods including CPDFConversion.start_pdf_to_word/start_pdf_to_ppt/start_pdf_to_excel.
• LibraryManager provides initialize, license_verify, release, set_document_ai_model, and set_ocr_language.
• Official documentation confirms support for PDF to Word / Excel / PPT / HTML / RTF / Image / TXT / JSON / Markdown.
• The SDK's start_pdf_to_* interfaces natively accept image file input (PNG → Word has been verified successfully).
• enable_ai_layout defaults to True in the SDK; set_document_ai_model() must be called first to load the model before use, otherwise a 0xC0000005 crash will occur.
• --ocr-language supports specifying multiple languages simultaneously (e.g. --ocr-language chinese english).

Risks / Notes

• The official requirements page states Python >=3.6, while the demo page states <3.11, but PyPI currently provides a cp314 wheel in practice; treat the locally installable wheel as the source of truth, but always verify installation in a new environment first.
• If the script cannot download license.xml from the server (network issue) and no manual file exists in scripts/, or the <key> field is empty, the script cannot complete SDK authentication and cannot perform any real conversions.
• documentai.model is a large file (approximately 525 MB); there will be a noticeable download delay the first time OCR / AI layout is enabled. Because --enable-ai-layout defaults to True, the model download will be triggered on the very first run.
• If the runtime environment cannot access https://download.compdf.com/skills/model/documentai.model, place documentai.model in the scripts/ directory in advance.
• Do not directly apply the initialization patterns from ComPDF SDKs for other languages to the Python package; this Skill is based on the locally verified LibraryManager / CPDFConversion API.

Resource Navigation

• License file: License.txt
• Script: scripts/pdf-to-word-docx.py
• SDK authentication file: scripts/license.xml (auto-downloaded from https://download.compdf.com/skills/license/license.xml if missing)
• SDK authentication source: the <key> field in license.xml
• SDK resource path: scripts/
• OCR / AI layout model: scripts/documentai.model (auto-downloaded if missing)
• Purchase a full License: https://www.compdf.com/contact-sales
• Official documentation:

- https://www.compdf.com/guides/conversion-sdk/python/overview - https://www.compdf.com/guides/conversion-sdk/python/pdf-to-word - https://www.compdf.com/guides/conversion-sdk/python/pdf-to-excel - https://www.compdf.com/guides/conversion-sdk/python/pdf-to-ppt - https://www.compdf.com/guides/conversion-sdk/python/apply-license

Acceptance Checklist

• [ ] python -m pip show ComPDFKitConversion shows the installed package
• [ ] Running python "${CLAUDE_SKILL_DIR}/scripts/pdf-to-word-docx.py" --help or an equivalent local command produces normal output
• [ ] The script auto-downloads scripts/license.xml if missing, then extracts the license key from the <key> field for authentication
• [ ] The script uses the scripts/ directory as the SDK resource path
• [ ] The script recognizes all 10 target formats: word/excel/ppt/html/rtf/image/txt/json/markdown/csv
• [ ] The script accepts both PDF and image files (.png/.jpg/.jpeg/.bmp/.tif/.tiff/.gif/.webp/.tga) as input
• [ ] When --enable-ocr or --enable-ai-layout (enabled by default) is active and documentai.model is missing, the script auto-downloads the model
• [ ] When license.xml cannot be obtained (download fails and no manual file exists) or authentication fails, a clear error is output rather than a silent failure
• [ ] The 7 parameters that default to True can be disabled with --no-xxx
• [ ] --ocr-language supports specifying multiple languages simultaneously
• [ ] After a conversion using the trial License, the usage count increments
• [ ] When the trial License reaches 200 conversions, the script refuses to convert and outputs a purchase link
• [ ] When using a non-trial License, no usage limit applies
• [ ] For image input, even if --enable-ocr is not passed, the script automatically enables OCR and prints a notice to stderr
• [ ] For HTML output, even if --page-layout-mode is not passed, the script automatically uses box (box layout) and prints a notice to stderr
• [ ] For HTML output, explicitly passing --page-layout-mode flow overrides the automatic box layout behavior

Distribution Notes

• This Skill does not depend on any machine-specific absolute paths.
• When distributing to other users, the following directory structure is sufficient:

  pdf-to-word-docx/
  ├── SKILL.md
  ├── License.txt
  └── scripts/
      └── pdf-to-word-docx.py

• Users place this directory under their own skills root directory and the Skill is ready to use.
• license.xml is auto-downloaded at runtime; no need to include it in the distribution package.

Common Pitfalls

• scripts/license.xml is missing and cannot be auto-downloaded (network unavailable or server error): the script will error out before authentication. If you are in an offline environment, place license.xml manually in the scripts/ directory.
• scripts/license.xml is missing the <key> field or its value is empty: the script will error out before authentication.
• SDK resource files required by the SDK are absent from the scripts/ directory: conversion may fail after LibraryManager.initialize().
• A password-protected PDF is provided without --password: this will trigger PDF_PASSWORD_ERROR.
• OCR / AI layout is enabled but documentai.model is not present locally and the network is unavailable: the model download will fail; place the file in the scripts/ directory manually in advance.
• When the Excel output strategy is unclear, prefer passing --excel-worksheet-option explicitly to avoid unexpected result structures.
• When converting images to other formats, the script already enables OCR automatically; if the output still contains no text, check whether documentai.model is complete and whether the OCR language matches.
• Once the trial License usage limit is exhausted, a full License must be purchased to continue; purchase link: https://www.compdf.com/contact-sales.

Copyright

This Skill is built on top of the ComPDFKit Conversion SDK.

© 2014-2026 PDF Technologies, Inc., a KDAN Company. All Rights Reserved.

• SDK Name: ComPDFKitConversion
• SDK Author: PDF Technologies, Inc.
• License Type: Commercial License (Commercial / Proprietary) — non-exclusive, non-transferable, non-sublicensable, revocable
• Official Website: https://www.compdf.com
• Contact: support@compdf.com
• Terms of Service: https://www.compdf.com/terms-of-service
• Privacy Policy: https://www.compdf.com/privacy-policy

Important: Under the ComPDFKit Terms of Service, distributing the documentation, sample code, or source code of the ComPDFKit Conversion SDK to third parties is prohibited. Please ensure you have obtained a valid ComPDFKit License before using this Skill.