How to use the PDF Reader agent skill: setup, prompts and tips

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

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent reads when a task matches the skill. You do not need to program anything: once the skill is installed, the agent uses its guidance when you ask for PDF work.

With this skill, the agent preserves the logical structure of a document, including headings, sections, lists and table relationships. It keeps exact extracted text separate from its own interpretation or summary, and it flags content it could not extract reliably, such as scanned pages without OCR or corrupted sections.

• Extracts text from text-based PDFs while keeping paragraph boundaries and section headings
• Reconstructs tables and presents them as Markdown, CSV or JSON
• Turns form labels and filled values into key-value pairs
• Pulls dates, amounts, names and addresses into structured formats
• Summarizes hierarchically: a one-line overview, then a section-by-section breakdown
• Compares multiple PDFs by section or topic and highlights differences
• Searches by keyword, page number or section heading
• Reports document metadata such as author, creation date, page count and PDF version

How do you install the pdf-reader skill?

Download the skill as a ZIP from our page. Then choose one of the two install routes below.

For a personal setup available across your projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Either way, the pdf-reader folder should contain SKILL.md.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

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

What does the skill need before it works on scanned PDFs?

Text-based PDFs need nothing extra. For scanned PDFs, which are images of pages, the skill suggests command-line OCR tools named in its instructions: tesseract, and pdf2image combined with an OCR step. These tools are not bundled with the skill, so you must install them yourself. The skill can also mention cloud OCR APIs, which would need their own account and key.

The skill does not run scripts and does not use the network itself. Whether a cloud OCR service is involved depends on your agent setup.

• Command-line OCR tools for scanned PDFs: tesseract and pdf2image (not included)
• Optional cloud OCR API, which requires your own account and key
• Runs scripts: no
• Uses network: no

How do you trigger the skill with natural requests?

You do not need a special command. Describe the document and the outcome you want, and mention PDF work explicitly. The agent will apply the skill when the request matches its description: PDF content extraction and analysis.

Be specific about the output. If you need a spreadsheet, ask for CSV. If you need structured fields, ask for JSON. If you want to check a claim, ask for exact quoted text alongside the summary, since the skill keeps the two separate.

• "Read contract.pdf and list every payment date and amount, with the section each comes from."
• "Convert the table on page 4 of report.pdf into CSV."
• "Summarize this academic paper: one-line overview, then each section, and flag anything you could not read."
• "Compare the two supplier agreements and highlight where the terms differ."
• "Find every mention of 'termination' in these PDFs and give the page numbers."

What are good example workflows?

Professionals can ask for key dates, obligations and defined terms from a contract, or for line items, totals, tax amounts and payment terms from an invoice. Finance users can pull tables and footnotes from a report.

Students and researchers can request an abstract, methodology, results, conclusions and references in a structured summary. Anyone with tables or forms can ask for a spreadsheet-friendly output.

Always check the output against the source for anything that matters, especially where the agent flags uncertain sections.

What safety points should you check first?

The skill itself is instructions only, but the agent's environment decides where data goes. If OCR runs through a cloud service, the file contents may leave your machine. Check your setup before feeding in highly sensitive or confidential documents.

Do not treat the output as a final legal or financial decision. Extraction and summaries can contain errors, and the skill warns that unreliable sections may be missed.

• Confirm whether your OCR path is local or cloud-based before using confidential files
• Verify amounts, dates and obligations against the original PDF
• Ask the agent to list anything it could not read

What tips get better results?

Name the sections or pages you care about. The skill tries to keep headers, footers, page numbers, footnotes and appendices in view, so ask explicitly if you want only the main body; otherwise it will not skip footnotes or appendices.

For tables, ask the agent to confirm row and column alignment if the layout is complex. For multi-column pages, the skill is designed to identify column boundaries and read in the correct order.

• Say whether you want only the main body or also footnotes and appendices
• Request a specific format: Markdown for reading, CSV or JSON for further processing, plain text for simple extraction
• For scanned files, ask whether OCR was used and what confidence level applies