kreuzberg
by kreuzberg-devThe kreuzberg skill helps you install and use Kreuzberg for document extraction across 91+ formats, including PDFs, Office files, images, HTML, email, and archives. It covers Python, Node.js/TypeScript, Rust, and CLI workflows for OCR, tables, metadata, batch processing, and practical parsing guidance.
This skill scores 91/100, which means it is a strong listing candidate for directory users: it is highly triggerable, covers a broad real workflow, and provides enough operational detail for an agent to install and use it with relatively little guesswork. The repository clearly explains when to use Kreuzberg, how to install it across multiple runtimes, and where to look for deeper API/CLI/reference guidance.
- Explicit, actionable trigger: extract text, tables, metadata, and images from 91+ formats across Python, Node.js/TypeScript, Rust, and CLI.
- Strong operational coverage: installation, sync/async extraction, configuration, batch processing, OCR, error handling, and plugins are all called out in the skill description and references.
- Good progressive disclosure: multiple reference files provide language-specific APIs, CLI commands, configuration, supported formats, and advanced features.
- Some install paths are spread across many references, so first-time adopters may need to read beyond SKILL.md to choose the right runtime and feature set.
- No install command in SKILL.md itself, so users relying on the skill file alone may need to consult the references for exact setup details and feature flags.
Overview of kreuzberg skill
What kreuzberg does
The kreuzberg skill helps you use Kreuzberg to extract text, tables, metadata, images, and OCR-backed content from 91+ document formats, with native support for Python, Node.js/TypeScript, Rust, and a CLI. It is best for people who need reliable document processing code, not just a one-off prompt that guesses at parsing.
Who should install it
Install kreuzberg if your task is to turn PDFs, Office files, images, HTML, email, archives, or academic files into structured output, especially when scan quality, batch runs, or language-specific OCR matter. It is a strong fit for ingestion pipelines, document search, RAG prep, and extraction tooling.
Why it is different
The main value of the kreuzberg skill is that it is implementation-oriented: it covers install paths, extraction modes, config, batch processing, error handling, and plugins across multiple runtimes. That makes it more useful than a generic “analyze this document” prompt when you need code you can actually run.
How to Use kreuzberg skill
Install and confirm the target runtime
For a fast kreuzberg install, start from the runtime you will actually ship:
pip install kreuzberg
npm install @kreuzberg/node
cargo install kreuzberg-cli
Then read the matching API reference first: references/python-api.md, references/nodejs-api.md, or references/rust-api.md. If you are using the CLI, begin with references/cli-reference.md. The skill is most effective when you choose one runtime and one document type first instead of asking for everything at once.
Turn a rough request into a usable prompt
A good kreuzberg usage prompt names the file type, extraction goal, runtime, and constraints. For example: “Use kreuzberg in Python to extract invoice text, tables, and OCR from scanned PDFs, keep line breaks, and return JSON suitable for downstream parsing.” That is better than “extract data from PDFs” because it tells the skill whether to optimize for tables, OCR, or clean text.
Read these files first
For practical kreuzberg guide work, read in this order: SKILL.md, references/configuration.md, the runtime API file, and references/supported-formats.md. Then open references/advanced-features.md if you need plugins, OCR tuning, or batch behavior. This order surfaces the decisions that most affect adoption: install shape, supported inputs, and configuration defaults.
Use the workflow that matches your job
If you are processing one file, start with a simple extract_file or CLI extract call, then add MIME hints or config only if the output is wrong. If you are processing many files, check batch helpers and error handling early. For kreuzberg for PDF Processing, OCR settings and output format usually matter more than the base extraction call, so validate those before you scale up.
kreuzberg skill FAQ
Is kreuzberg only for PDFs?
No. PDF is a major use case, but the skill also covers Office documents, images, HTML, email, archives, and academic formats. If your workload is mixed-format ingestion, kreuzberg is a better fit than a PDF-only tool.
Do I need to know the library before using the skill?
No, but you do need to know your target runtime and output goal. The kreuzberg skill is beginner-friendly if you can describe the document type, whether OCR is needed, and whether you want plain text, markdown, JSON, or structured metadata.
When should I not use kreuzberg?
Skip kreuzberg if your task is primarily semantic summarization, not extraction, or if you only need a quick manual prompt for a single document with no code output. It also may be overkill if your pipeline does not need OCR, tables, or multi-format support.
How is it different from a normal prompt?
A normal prompt can describe the task, but kreuzberg usage is about getting the right install, API call, config, and failure handling for document extraction. That makes it better when output quality depends on runtime setup, OCR backend choice, or batch processing details.
How to Improve kreuzberg skill
Provide the input shape upfront
The best kreuzberg skill results come from prompts that specify file type, source quality, and desired output. Include details like “scanned PDF,” “digital PDF,” “invoice tables,” “email attachments,” or “preserve headings.” Those details change whether OCR, chunking, or plain extraction is the right default.
State the failure mode you want to avoid
If your first output is poor, tell the skill what went wrong: missing tables, broken line breaks, slow OCR, bad language detection, or noisy images. For kreuzberg for PDF Processing, this helps narrow whether to adjust OCR backend, configuration, or output format instead of rewriting the whole workflow.
Iterate with concrete examples
A stronger improvement loop is to paste one failing file description and one target result, such as: “This scanned invoice should produce invoice number, total, vendor, and line items in JSON.” That is more useful than asking to “make it more accurate,” because the skill can tune extraction advice to the actual schema and document type.
Start narrow, then expand
Begin with one runtime, one format, and one extraction mode. Once the base kreuzberg install and extraction path are working, add batch processing, plugins, or advanced configuration. This reduces confusion and makes it easier to verify whether the problem is installation, OCR, or downstream parsing.