PDF Form Filler

Fill flat PDFs (no AcroForm fields) by transplanting answers from a prior filled copy, drawing text at exact PDF coordinates, and verifying every page by rendering it and looking at the pixels. Requirement: you (the agent) must be able to view images. There is no non-VLM fallback; without vision, do not attempt this workflow, you will ship misaligned text and never know.

Core insight

A form is a measured coordinate range, never a visual impression. The whole discipline is: get exact geometry from the PDF's own data, convert coordinate conventions correctly, and verify placement against rendered pixels because every library lies at least once.

The one workflow that works

1. Mine a prior filled copy before writing any code

Search the user's machine (Downloads, Documents, email attachments) for a previously filled version of the same form. Recurring forms (yearly enrollment, renewals) are almost always the identical template year over year. A prior copy provides:

• Every answer (and the positions of its answers, which become your coordinates)
• The user's signature as vector ink (see signature section)
• Baseline offsets that a human already got right (steal them)

Diff the prior template against the new blank: extract all text lines with coordinates from both and compare position sets. If blanks align, transplant answers wholesale and update only what changed. Ask the user a batched list of just the changed fields (insurance plan, phone numbers, dates, names); never guess these.