Skip to content
Get packs

Guide · Agent skill

How to use the Document Extraction (ADE) agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The document-extraction skill teaches an AI agent to use LandingAI's Agentic Document Extraction (ADE) service to read PDFs, images, spreadsheets and presentations, return them as structured Markdown, pull out named fields, and sort mixed batches by document type. It is aimed at developers and analysts who work with many documents. After this guide you will know what the skill does, how to install it, how to phrase requests so the agent triggers it, and what it needs to run.

Document Extraction (ADE)andrewyng/context-hub · MIT

Quick start

  1. 1Download and unzip. Download the ZIP from this page and unzip the document-extraction folder into ~/.claude/skills/ or a project's .claude/skills/ folder.
  2. 2Get an API key. Create a LandingAI ADE API key at va.landing.ai and store it as VISION_AGENT_API_KEY in a .env file or an environment variable.
  3. 3Prepare Python. Use a local venv, uv or poetry project and install landingai-ade and python-dotenv into it.
  4. 4Ask for a document task. Describe the documents and the fields or output you want in plain language.
  5. 5Review the output. Check the Markdown, extracted fields and grounding, and confirm the values against the source document.

What the skill makes the agent do

Agent skills are folders of instructions and scripts that an agent loads when a task matches them. This skill covers three jobs. It can parse documents into structured Markdown with layout understanding. It can extract specific data, such as invoice fields or table data, using a schema. It can also classify and separate multi-document batches by type, for example invoices versus receipts or statements versus forms.

The skill also covers visual grounding, meaning bounding boxes and page numbers that show where extracted content appears in a document. Per the skill's description, it can help the agent write code that runs over a set of documents, so the agent does not have to load every file into its own context window.

  • Parse: PDFs, images, spreadsheets and presentations into Markdown
  • Extract: named fields from a document using a JSON schema or Pydantic model
  • Split: classify and separate batches by document type
  • Ground: bounding boxes and page numbers for extracted content

Install the skill from the ZIP file

Download the document-extraction ZIP from this page and unzip it. To make the skill available to all your projects, place the unzipped folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.

In the Claude apps you can instead upload the skill under Settings, then Capabilities, then Skills.

  • Personal install: ~/.claude/skills/document-extraction
  • Project install: .claude/skills/document-extraction inside your project

Set up Python packages and your API key

The skill runs Python scripts, so it needs a Python environment. The skill prefers a local venv or a uv or poetry project and says not to install packages globally without your approval. It installs landingai-ade and python-dotenv into that environment.

You need a LandingAI ADE API key, stored as VISION_AGENT_API_KEY. You can get one from va.landing.ai at the API key settings page. The skill looks for a .env file in the skill folder and in parent directories, or you can export the variable yourself. The .env file is gitignored. If you use the EU endpoint, the skill notes you should set environment to eu when creating the client.

  • Required packages: landingai-ade, python-dotenv
  • Required variable: VISION_AGENT_API_KEY
  • Network access to LandingAI's API endpoint is needed
VISION_AGENT_API_KEY=your_actual_api_key_here

Trigger the skill with natural requests

You do not need to name the skill. Ask for a task that involves document content, and the agent will load it when the request matches. Requests that mention parsing, extracting fields, sorting a batch by type, or showing where data sits on the page are the most likely to trigger it.

Name the document type and the fields you want. Specific requests work better than general ones.

  • Convert this folder of scanned PDFs into Markdown files I can search.
  • Pull the invoice number, invoice date, vendor name and total amount from each invoice in this folder into a CSV.
  • Separate this mixed batch into invoices, receipts and bank statements.
  • Show me the bounding box and page number for each total amount so I can highlight it in review.

Parse and extract with a script

The skill's basic parse example loads your key, creates a client, sends a PDF with the dpt-2-latest model, and saves the Markdown. Each chunk in the response has a type and grounding page. For extraction, the skill defines a schema, converts it to JSON Schema, and runs extract-latest against the Markdown from the parse step.

Parsing is the first step. Extraction reads the parsed Markdown, so keep the parse output if you plan to run several extractions.

  • Choose dpt-2-latest for complex documents with logos, signatures or ID cards.
  • Choose dpt-2-mini for simple, digitally-native documents, which the skill describes as faster and cheaper.
from landingai_ade import LandingAIADE from pathlib import Path client = LandingAIADE() response = client.parse(document=Path("document.pdf"), model="dpt-2-latest") print(response.markdown[:500])

Handle file limits and large jobs

The skill description says large documents can be processed asynchronously, up to 1GB or 1000 pages. The file format reference lists 20+ formats. The Playground accepts PDFs up to 100 pages and spreadsheets up to 10 MB, while the API has no spreadsheet size limit. Check the rate limits in LandingAI's documentation for API PDF limits.

If a run fails, the skill includes a troubleshooting reference you can ask the agent to consult.

  • PDF, images, DOC/DOCX/ODT, PPT/PPTX/ODP and CSV/XLSX are covered
  • Office files are converted to PDF before parsing

Safety, privacy and scripts

This skill runs scripts and uses the network. Document content is sent to LandingAI's hosted SaaS service. Do not use it for private or confidential documents unless your organisation has approved that data handling. The skill also depends on a remote API and an API key, so it will not work offline or in an air-gapped environment.

Review the scripts the agent writes before it runs them on sensitive files, and keep your API key out of version control.

  • Content leaves your machine for LandingAI's hosted service
  • Requires network access and an API key
  • Runs Python scripts written by the agent

What people use it for

  • Turning a folder of scanned supplier PDFs into searchable Markdown files
  • Pulling invoice numbers, dates and totals from many invoices into one CSV
  • Separating a mixed batch of invoices, receipts and statements before further processing
  • Showing reviewers the bounding box and page number of each extracted value

Frequently asked questions

How do I install the document extraction skill?

Download the document-extraction ZIP from this page and unzip it into ~/.claude/skills/ for all projects or into a project's .claude/skills/ folder for one project. You can also upload it in the Claude apps under Settings, then Capabilities, then Skills.

Is the document extraction skill free?

The skill is released under the MIT licence. The source does not state pricing for the LandingAI ADE service, so check va.landing.ai before you rely on it for production work.

What do I need to run the ADE skill?

You need a Python environment, the landingai-ade and python-dotenv packages, a LandingAI ADE API key stored as VISION_AGENT_API_KEY, and network access to LandingAI's API endpoint.

Can I use this skill offline or with confidential documents?

No for offline use, since it depends on a remote API. For confidential documents, note that content is sent to LandingAI's hosted SaaS service, so get approval for that data handling first.

Why does the agent say it cannot find my API key?

The skill checks the environment variable VISION_AGENT_API_KEY first, then looks for a .env file in the current directory, parent directories and the skill folder. Make sure the .env file contains VISION_AGENT_API_KEY=your key, or export the variable in your shell.

Do I need this skill for a quick summary of one short file?

Probably not. The skill's own assessment says the agent can read a short file directly, so this setup is best for batches, field extraction or visual grounding.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a developer or analyst who processes many invoices, forms or statements, or needs bounding boxes for document review, and you can send the documents to LandingAI's hosted service. Pick something else if the documents must stay private, you need offline processing, or you only want a quick summary of a short file.

Download the skill