Use PDF by Anthropic to work with PDF files through an agent: extract text and tables, combine or split pages, prepare forms, or create a document for sharing. Choose the operation from the actual input, then check extracted content and the rendered output before delivery.
State whether the PDF task concerns extraction, assembly, or form filling and identify the relevant pages or fields. Prepare a request that preserves source references and uncertainty, with saved-output checks matched to the actual operation instead of file appearance alone.
Identify the pages, table columns, and required output format.
Copy the request into a new Floatboat conversation and provide the required material.
Download FloatboatPreview and copy the request. It is not sent automatically from this page.
Reference: PDF https://github.com/anthropics/skills/blob/main/skills/pdf/SKILL.md Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding. My task: Extract the tables from this PDF report into a structured file. Check whether the pages are scanned, preserve row labels, and flag uncertain values rather than filling them in. Expected deliverable: Structured output with page references, preserved labels, and explicit uncertainty markers.
PDF is Anthropic’s document Skill for reading, creating, and modifying PDF files. It routes work to Python libraries and command-line tools according to the operation: extracting content, organizing pages, generating documents, processing scanned pages, or handling forms through separate instructions. A text-based report and a scanned document need different extraction methods, so the input matters as much as the requested output. The Skill provides workflow guidance and references; its required libraries and utilities must be available in the execution environment. Anthropic publishes this document Skill as source-available under its specific license terms. Obtain the Skill package from Anthropic’s official repository and review its license before use.
Report extraction, document assembly, and form filling require different PDF checks. Compare page references, ordering, field mappings, and recognition uncertainty so the resulting file can be inspected against the original rather than judged by appearance alone.
Extracting a PDF report starts with the target pages, columns, or passages you actually need. Read selectable text and tables directly, using OCR for scanned pages when necessary, then compare values and headings with the originals. Keep page references and mark uncertain recognition rather than presenting every extracted value as exact.
A combined or split PDF package needs a clear file list, page ranges, and intended order. Assemble only the requested material and inspect the saved result for page count, sequence, and orientation. Those checks establish that the document package matches the handoff request before it is shared.
A PDF form needs its original fields, label mapping, and approved values inspected before filling. Follow the dedicated field instructions and check that values reach the intended labels in the saved document. Reopen the output and identify unsupported fields instead of treating a visually plausible form as verified.
Approved text, tables, and page requirements can be assembled into a new PDF rather than extracted from an existing one. Choose the documented generation path and make the required tools available, then inspect rendered pages for pagination, readable tables, and missing content. Verify the saved artifact against the source material before treating the report as ready to share.
PDF work begins by identifying whether the file needs reading, assembly, or form operations. Match the source instructions to that task, inspect the saved pages and values, and retain uncertainty when scanned content or unsupported fields prevent exact verification.
Identify selectable text, scanned pages, and form fields. Specify the pages and output format that the task requires.
Ask the agent to follow the relevant source instructions with the required tools. Preserve the original and write a clearly named output.
Compare extracted or edited information with the source. Render layout-sensitive outputs and inspect page order, clipping, and form values.
Read another Skill guide and check its original source before using it.
