Skip to content
Get packs

Guide · Agent skill

How to use the Document Workflows (LandingAI ADE) agent skill

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

The document-workflows skill gives an AI coding agent ready-made patterns for chaining LandingAI ADE steps into full document pipelines: parsing many files in parallel, sorting mixed document types, preparing chunks for search databases, loading results into CSV, DataFrames or databases, and drawing boxes on the original page images. It is aimed at developers and data teams who already use LandingAI ADE and want an agent to build the pipeline code. After this guide you will know how to install the skill, how to phrase a request so the agent picks it up, and what the agent will need from your environment before it runs.

Document Workflows (LandingAI ADE)andrewyng/context-hub · MIT

Quick start

  1. 1Download the skill ZIP. Download the document-workflows ZIP from the page where you found it.
  2. 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder for one project.
  3. 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
  4. 4Set up Python and your API key. Create a .venv with the landingai-ade SDK and put your LandingAI ADE API key in a .env file.
  5. 5Ask for a pipeline in plain language. Describe the documents, the pipeline step and the output format you need.
  6. 6Let the pre-flight run. Allow the agent to parse one to three samples and inspect the output before it writes section or heading code.
  7. 7Check the output visually. For any overlay or crop, open at least one output PNG and confirm it shows what you asked for.

What is an agent skill and what does this one do?

An agent skill is a folder of instructions and reference files that an AI agent loads when a request matches the skill's description. This one is named document-workflows and its main file is SKILL.md, with supporting reference files for batch processing, database integration, RAG pipelines, schema catalog, table stitching and visualization.

The skill is a companion to a simpler skill, document-extraction, which covers basic LandingAI ADE SDK calls such as parse, extract and split. document-workflows is for composing those operations into pipelines. The skill states that the agent will run a short pre-flight check on sample documents before writing code, so heading formats are read from real output rather than guessed.

  • Batch processing: parse and extract many documents in parallel or async, with per-document error handling
  • Classify-then-extract pipelines for mixed document types
  • RAG preparation: chunking and vector database ingestion
  • Loading results into Snowflake or exporting to CSV and DataFrames
  • Visualization: bounding box overlays, chunk crops, and word-level highlights on page images

Install the skill (download ZIP, then unzip)

Download the ZIP of the skill from the page where you found it. Then unzip it into one of two locations, depending on how you want to use it.

For a personal setup available across projects, unzip it into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder. In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/
  • Project only: .claude/skills/ inside your project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Prepare your environment before the agent runs

The workflows depend on paid LandingAI ADE API calls, so you need an ADE account and API key before anything works. The skill expects a Python environment in .venv with the landingai-ade SDK installed. The API key is usually loaded from a .env file using python-dotenv.

Other libraries the skill uses include pymupdf, Pillow, python-dotenv and pandas. Some workflows also mention chromadb, Streamlit and Snowflake connectors, and Tesseract OCR is optional for scanned or handwritten pages.

  • A .venv Python environment with the landingai-ade SDK
  • A LandingAI ADE API key, usually in a .env file
  • Libraries such as pymupdf, Pillow, python-dotenv, pandas
  • Optional: chromadb, Streamlit, Snowflake connectors for specific workflows
  • Optional: Tesseract OCR for scanned or handwritten pages

Trigger the skill with natural requests

You do not need to name the skill. The agent matches your request against the skill description. Requests that mention a pipeline, batch, RAG, a database load or an overlay on a page are the ones it is written for. Be specific about the document type and the output you want.

The skill also covers the pre-flight step: the agent will parse one to three sample documents and inspect the markdown before writing section-matching code. Let it do this, because heading formats vary between documents.

  • "Parse this folder of invoice PDFs in parallel and save each result as JSON."
  • "Prepare these parsed reports for a RAG index and export chunks to CSV with page and box columns."
  • "Load extracted fields from these forms into Snowflake, one row per document."
  • "Draw bounding boxes on the original pages for each extracted field so I can check them."
  • "Find every mention of 'L2S' in the Introduction section and highlight it on the page."

Safety notes: scripts and network access

This skill runs scripts and uses the network. The generated code calls the LandingAI ADE API, so document content is sent to that service. Check which files will leave your machine before you run them against sensitive data.

The skill mentions redacting PII. Redaction accuracy depends on the code the agent writes, so review redacted output yourself. Also visually check at least one output PNG from any crop or overlay workflow, as the skill itself requires, to catch wrong-page or wrong-region errors.

  • Runs scripts and uses the network: review generated code before execution
  • Sensitive documents need human review, especially for PII redaction
  • ADE page numbers are 0-indexed, so off-by-one errors are possible in crops

Tips for better results

Give the agent one or two representative documents and a clear target output, such as a CSV schema or a database table. Save the parse result to JSON after the first run, as the skill recommends, so later development does not repeat paid parse calls.

Keep the sample set small during development. Expand to the full corpus only once the section and heading logic works on the samples.

  • Save parse results with model_dump() to JSON and reload them during development
  • Start with 1 to 3 samples, not the full corpus
  • Name the output format you want: CSV, DataFrame, Snowflake or page overlay

What people use it for

  • Parse a folder of invoices in parallel and save each result as JSON for later review
  • Build a RAG dataset from technical reports, exporting chunks with page and bounding box columns
  • Load extracted form fields into Snowflake or a CSV file, one row per document
  • Draw bounding boxes on original pages to verify each extracted value by eye
  • Find and highlight specific terms in a named section of a paper

Frequently asked questions

How do I install the document-workflows skill?

Download the skill ZIP and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can upload the ZIP under Settings → Capabilities → Skills instead.

What do I need to use document-workflows?

You need a Python environment in .venv with the landingai-ade SDK, a LandingAI ADE API key usually stored in a .env file, and libraries such as pymupdf, Pillow, python-dotenv and pandas. The workflows call paid ADE APIs, so an account is required.

Is document-workflows free?

The skill is listed under the MIT licence. LandingAI ADE API calls are paid, and the source does not give pricing, so check LandingAI's terms for current costs.

What is the difference between document-workflows and document-extraction?

document-extraction covers basic ADE SDK calls such as parse, extract and split. document-workflows composes those operations into pipelines like batch runs, RAG preparation, database loads and page overlays. The source says the basic skill is simpler and cheaper for one-off work.

Should I use this skill for a single PDF?

Probably not. The skill is designed for pipelines and multi-document workflows. For a one-off extraction from a single PDF, the simpler document-extraction skill is the better fit.

Why does the agent parse sample documents before writing code?

Heading formats differ between documents, and a wrong guess causes silent failures such as zero matched sections. The skill requires a pre-flight parse of one to three samples so the code is based on the actual ADE output.

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

Choose document-workflows if you build document ingestion pipelines on LandingAI ADE and want an agent to write batch, RAG, database or overlay code with checks built in. Skip it if you need only one extraction from one PDF, or if you do not have an ADE account, since the skill depends on paid API calls. Review any output that touches sensitive data, because the skill's redaction depends on the generated code.

Download the skill