OpenDataLoader PDF Parser (乌贼版)

PDF parsing tool for AI/RAG. Convert PDF to Markdown, JSON, HTML with layout preservation, bounding boxes, and image extraction. Use when you need to extract...

Security Scan

The skill's declared purpose (PDF parsing) matches its instructions, but small inconsistencies and an unclear 'hybrid AI' mode (no credentials declared) plus missing source/homepage metadata mean you should verify the package origin and network behavior before installing.

Purpose & Capability

Name/description, CLI examples, and included test script all align with a PDF parsing CLI. The SKILL.md states the package is installed via pipx and bundles a PDFBox JAR — that is coherent for a Java-based PDF tool. Minor inconsistency: registry metadata lists no homepage/source while SKILL.md includes a GitHub homepage URL, which should be verified.

Instruction Scope

Runtime instructions only run a local CLI (opendataloader-pdf) against local PDF files and write output to local directories. They do not instruct reading unrelated system files or environment variables. Ambiguity: the '--hybrid' / 'Hybrid AI mode: docling-fast' option could imply contacting an external AI service or model; SKILL.md gives no details about network calls, remote endpoints, or required API keys.

Install Mechanism

No install spec in the registry; SKILL.md recommends 'pipx install opendataloader-pdf'. Installing from PyPI/pipx runs package install scripts which may execute code at install time and will pull the package from wherever it's published. The SKILL.md claims a GitHub homepage, but the registry shows source unknown — confirm the exact pip package name and origin before installing.

Credentials

The skill declares no required environment variables, no secrets, and no config paths. That is proportionate for a local PDF parsing CLI. However, the hybrid/AI option appears under-specified: if it uses a remote service it would typically require API credentials (none are declared), so verify whether additional credentials are needed at runtime.

Persistence & Privilege

The skill does not request persistent/autostart privileges (always:false). It is user-invocable and allows autonomous invocation (default) which is normal. It does not declare modifications to other skills or system-wide settings.

What to consider before installing

This skill appears to do what it says (convert PDFs locally) but there are a few red flags to check before installing: 1) Verify the pip package source — confirm the exact PyPI package and/or GitHub repo matches the SKILL.md homepage and is from a trusted maintainer. 2) Inspect the package contents (and bundled JAR) before installing, or install into an isolated environment/container to observe behavior. 3) Ask the maintainer or check docs what '--hybrid' (docling-fast) does and whether it calls remote services or requires API keys — if it does, confirm what endpoints and credentials are used. 4) Because pipx runs code at install time, avoid installing on sensitive/production hosts until you've validated the package. If you want, provide the actual pip package name or the source repo and I can check it for further inconsistencies.

opendataloader-pdf Skill

PDF parsing tool for AI/RAG scenarios. Converts PDF to Markdown, JSON, HTML with layout preservation.

Installation

pipx install opendataloader-pdf

Requires Java runtime (bundled JAR is included).

Quick Usage

# PDF to Markdown (most common)
opendataloader-pdf input.pdf -o output_dir -f markdown

# PDF to JSON (with bounding boxes)
opendataloader-pdf input.pdf -o output_dir -f json

# Multiple formats at once
opendataloader-pdf input.pdf -o output_dir -f json,markdown,html

# Extract specific pages
opendataloader-pdf input.pdf -o output_dir -f markdown --pages "1,3,5-10"

# Extract images
opendataloader-pdf input.pdf -o output_dir -f markdown --image-dir images/

# Use PDF structure tree (for tagged PDFs)
opendataloader-pdf input.pdf -o output_dir -f markdown --use-struct-tree

# Output to stdout
opendataloader-pdf input.pdf -f markdown --to-stdout

Output Formats

Format	Description
json	Structured JSON with bounding boxes, fonts, reading order
markdown	Markdown text with images as references
html	HTML with styling
text	Plain text
pdf	Rebuilt PDF
markdown-with-html	Markdown with HTML for complex elements
markdown-with-images	Markdown with embedded base64 images

Key Options

Option	Description
--pages	Page range, e.g., "1,3,5-10"
--image-dir	Directory for extracted images
--use-struct-tree	Use PDF structure tree for reading order
--table-method	Table detection: default (border-based) or cluster
--reading-order	Algorithm: off or xycut (default)
--hybrid	Hybrid AI mode: docling-fast for complex tables
--sanitize	Remove sensitive data (emails, phones, etc.)
--include-header-footer	Include page headers/footers

Examples

Basic Conversion

# Convert to markdown
opendataloader-pdf document.pdf -o ./output -f markdown

# Convert to JSON with structure
opendataloader-pdf document.pdf -o ./output -f json --use-struct-tree

Batch Processing

# Multiple files
opendataloader-pdf "file1.pdf" "file2.pdf" "folder/" -o output/

# All PDFs in directory
opendataloader-pdf ./pdfs/ -o ./output/ -f markdown

Advanced Options

# Use AI hybrid mode for complex tables
opendataloader-pdf input.pdf -o output/ -f markdown --hybrid docling-fast

# Extract only pages 1-5
opendataloader-pdf input.pdf -o output/ -f markdown --pages "1-5"

# Sanitize sensitive data
opendataloader-pdf input.pdf -o output/ -f json --sanitize

Performance Notes

• Each convert() call spawns a JVM process
• For batch processing, pass multiple files in one call
• ~6 seconds for typical 300-page PDF
• Images extracted to {output_name}_images/ directory

Troubleshooting

Java not found

Ensure Java runtime is installed. The tool bundles its own PDFBox JAR.

Font warnings

Warnings about missing fonts are normal and don't affect output quality.

Slow performance

Use batch mode (multiple files in one call) instead of calling repeatedly.