Read and prepare a PDF

Before you start

You’ll make: A checked extraction or new PDF saved separately from your source.

• Claude Code installed and signed in.
• A small PDF you own or have permission to process.
• Local Python/document utilities when the task requires them; let the agent check what is missing first.

1. Install the skill

Run each slash command inside Claude Code, not your terminal shell.

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

The document-skills plugin also includes Word, spreadsheet, and presentation skills. If a PDF skill is already available, try it first instead of installing a duplicate.

Open the official source below and review SKILL.md, scripts, and the license. If the skill is missing after installation, start a new session or follow the agent’s reload guidance.

2. Try a small task

Copy this into a new Claude Code conversation. Replace example filenames with your own non-sensitive files where needed.

Use the PDF skill to read sample.pdf in this project. Summarize the headings and extract the table to a separate file. Keep sample.pdf unchanged. Tell me if any page needs OCR, and check the output against the original.

3. Check the result

• The original PDF is unchanged and outputs have separate filenames.
• You compare at least one page or table with the source.
• Scanned pages and uncertain OCR text are identified rather than silently guessed.

If something doesn’t work

The extracted text is empty

The document may be scanned. Ask for an OCR assessment; language data and OCR tools may need a separate trusted installation.

A document tool is missing

Have Claude identify the exact dependency and official installation source. Approve the installation only after reviewing it.