Tag a PDF with AI
Add headings, reading order and accessible form names to a text-based PDF. Compare the rendered pages and inspect the structure of a separate output file.
To tag a PDF with AI, start with a file whose appearance is already right. A volunteer form might have a clear heading and a neatly labelled email box, yet a reader moving between fields may hear nothing more useful than “edit”. The pdf-tagger skill helps your agent connect what is on the page with what the document tells assistive software.
The distinction matters in tables too. Bold text might be a section heading on one page and a row label on the next. The workflow uses a configuration specific to the document. When an item has no classification, the agent can revisit the actual page and decide what it means.
What it does
- Keep a heading with its paragraph in the reading sequence.
- Give a form control a meaningful name that survives navigation between fields.
- Distinguish a table heading from an ordinary data cell.
- Mark decorative rules so they do not interrupt the text.
- Compare the rendered original and output while reviewing their internal structure separately.
Install and use
- Install the skill and its Python dependencies in your agent environment. The optional independent PDF/UA check uses veraPDF.
- Provide a text-based PDF and identify the intended order of any columns, tables or fields.
- Review the labelled output, inspect the reading sequence and check that the visible page still matches the original.
FAQ
Source
Made by XR Navigation. Original skill: pdf-tagger, published on GitHub under the MIT licence.
View on GitHub