fill-i539-application-form

Fill USCIS Form I-539 PDF

Purpose

Fill the USCIS Form I-539, Application to Extend/Change Nonimmigrant Status — a 660 KB, 7-page fillable PDF — with applicant data and save a completed copy. The recommended path fetches the blank PDF and writes AcroForm field values programmatically with a PDF library, because the form does not render as DOM in a browser and therefore cannot be filled with page.fill() / evaluate(). This skill produces a filled PDF file; it does not submit anything to USCIS (there is no online submission surface for this form).

When to Use

• An agent needs to pre-populate an I-539 with a client's biographic, address, status, and signature data before a human reviews/prints it.
• Batch-generating I-539 drafts from a structured applicant record.
• Any flow where you were told to "fill the I-539 PDF with Browserbase page.fill()" — read the Gotchas first: that approach cannot work and this skill is the correct replacement.

Workflow

The blank form lives at https://www.uscis.gov/sites/default/files/document/forms/i-539.pdf (HTTP 200, application/pdf, PDF 1.7, encrypted, hybrid AcroForm+XFA, 176 form fields). The reliable method is: fetch the bytes → load in PDF.js → set values via annotationStorage keyed by each field's annotation id → saveDocument().

1. Fetch the blank PDF bytes

The Browserbase Fetch API returns the PDF base64-encoded, so decode it:

browse cloud fetch \
  "https://www.uscis.gov/sites/default/files/document/forms/i-539.pdf" \
  --allow-redirects --output i-539.b64
node -e "const fs=require('fs');fs.writeFileSync('i-539.pdf',Buffer.from(fs.readFileSync('i-539.b64','utf8').trim(),'base64'))"
# Verify header is %PDF and size ~495 KB decoded

A residential proxy / --verified stealth is not required — uscis.gov serves the static asset with no anti-bot (probe: none detected).

2. Enumerate fields and build a name → annotation-id map

pdf-lib cannot open this file (it is encrypted and uses compressed object streams — PDFDocument.load throws EncryptedPDFError, and even with ignoreEncryption:true the object streams fail to parse). Use Mozilla PDF.js (pdfjs-dist), which decrypts with an empty password and reads the AcroForm:

const pdfjs = await import('pdfjs-dist/legacy/build/pdf.mjs');
const doc = await pdfjs.getDocument({ data, password: '' }).promise;
const fo  = await doc.getFieldObjects();   // { fieldName: [ { id, type, exportValues, page } ] }

annotationStorage.setValue() is keyed by the widget annotation id (e.g. "183R"), not the field name — resolve it from fo[name][0].id.

3. Set values (text, combobox, checkbox)

const idByName = {};
for (const [name, arr] of Object.entries(fo)) idByName[name] = arr[0].id;
const set = (name, value) => doc.annotationStorage.setValue(idByName[name], { value });

// text + numeric fields (numeric fields are TEXT — always pass strings)
set('form1[0].#subform[0].P1Line1a_FamilyName[0]', 'DOE');
set('form1[0].#subform[0].Pt1Line2_AlienNumber[0]', '123456789'); // A-Number, digits only
set('form1[0].#subform[1].P1_Line8_DateOfBirth[0]', '01/15/1990'); // dates MM/DD/YYYY
// combobox: pass the option's export value
set('form1[0].#subform[0].Part2_Item11_State[0]', 'TX');           // 2-letter state code
set('form1[0].#subform[1].Pt2Line2a_NewStatus[0]', 'B-2');         // requested status code
// checkbox: pass boolean true (the stored export value is written automatically)
set('form1[0].#subform[0].Part1_Item4_Unit[0]', true);             // -> "APT"
set('form1[0].#subform[1].P3_checkbox2a[1]', true);                // Extension "Yes" -> "Y"

4. Field-name map (verified; the core value of this skill)

All names are prefixed form1[0].#subform[N]. (N = the subform index shown). The internal P#/Pt#/Part# prefixes do not line up with the printed Part numbers — always resolve names from getFieldObjects(), never assume Part N = subform N. Key fields:

There is no "read English" checkbox in this revision — the applicant statement is selected via #subform[3].P4_checkbox6_Yes[0] / ..._No[0] style radio pairs (each has export value "Y"; set the intended one to true).

5. Validate mandatory fields, then save

Before saving, confirm required fields are non-empty by re-reading getFieldObjects() values (Family Name, A-Number or DOB, Part 2 type + status, signature + date). Then:

const outBytes = await doc.saveDocument();   // returns Uint8Array
fs.writeFileSync('i-539-filled.pdf', outBytes);

Re-open the saved file with getFieldObjects() and assert each value round-trips (this skill verified DOE/JANE/TX/B-2/APT/01-15-1990 etc. all persist).

Browser fallback

There is no working browser fallback for filling this PDF. For completeness, if you only need to view it, browse open <pdf-url> --remote loads it, but the form is not fillable through the DOM (see Gotchas). Do not spend turns trying selectors — enumerate browse snapshot, observe zero form refs, and switch to the PDF.js path above.

Site-Specific Gotchas

• page.fill() / evaluate() cannot fill this PDF — confirmed dead end. Chrome renders it in the PDFium plugin, not the DOM. A driven session showed browse snapshot returning only RootWebArea > scrollable, html with urlMap: about:blank and 0 textbox/checkbox/combobox nodes; the CDP trace shows one 200 application/pdf response handed straight to the plugin. An independent 6-turn autobrowse run reached the same conclusion.
• Headless viewport is solid black. After browse open + 15 s wait + scroll the screenshot is entirely black — headless PDFium does not paint this XFA form, so even a "screenshot then OCR" approach fails. To rasterize the filled result, render with PDF.js (getPage().render) or a real desktop PDF viewer, not headless Chrome.
• The PDF is encrypted. pdf-lib throws EncryptedPDFError; with ignoreEncryption:true its parser then fails on the compressed object streams (Invalid object ref, Expected instance of PDFDict). Use PDF.js with password: '', or pre-decrypt with qpdf --decrypt in.pdf out.pdf and then pdf-lib will open the decrypted copy.
• Hybrid AcroForm + XFA. getMetadata() reports IsXFAPresent:true and IsAcroFormPresent:true. saveDocument() prints benign Warning: Node not found for path: ... for a handful of XFA-only / barcode nodes (e.g. the page-6 supplement fields, PDF417BarCode1, P1_Checkbox12c) — the AcroForm values are still written and round-trip correctly. Standard viewers (Chrome, macOS Preview, Acrobat) read the AcroForm layer, so the fill shows up.
• annotationStorage.setValue is keyed by annotation id, not field name. Map name → getFieldObjects()[name][0].id first. Passing the field name as the key silently no-ops.
• Numeric fields are text fields. A-Number, USCIS Online Account #, SSN, I-94 #, ZIP, phone — all text type. Pass strings; never numbers (a JS number can drop leading zeros and throws in some libs).
• Dates are free-text MM/DD/YYYY. No date picker/validation in the field; supply the formatted string exactly.
• Checkboxes carry meaningful export values, not "On". Part1_Item4_Unit → APT/STE/FLR; Yes/No pairs → Y/N; Part 2 applicant-count P2_checkbox → A/B/C. In PDF.js set the widget to boolean true; in pdf-lib call field.check(). Only one member of each radio-style group should be set.
• Two confusingly-named status comboboxes. Pt1Line15a_NewStatus[0] is the applicant's current status (Part 1); Pt2Line2a_NewStatus[0] is the requested/target status (Part 2). Both contain "NewStatus" — do not swap them.
• Internal part prefixes are offset from printed Parts. The applicant signature is P6_Line7_*, preparer is P7_*, the biographic questions are P4_* on #subform[3]. Resolve everything by enumerating fields; never assume Part N equals subform N or prefix P N.
• The Browserbase Fetch API returns base64. browse cloud fetch --output writes a base64 string (starts JVBERi0x = "%PDF-1"), not raw bytes — decode before loading (a raw-bytes load will report the file as ~660 KB vs the true ~495 KB).
• Form revision drift. USCIS periodically reissues the I-539 with a new edition date and can rename/re-index fields. Re-run the enumeration step against the freshly fetched file each run rather than hardcoding ids across editions.

Expected Output

{
  "success": true,
  "method": "pdfjs-annotationStorage",
  "source_pdf": "https://www.uscis.gov/sites/default/files/document/forms/i-539.pdf",
  "output_pdf": "i-539-filled.pdf",
  "fields_set": 27,
  "fields_verified_roundtrip": 27,
  "missing_fields": [],
  "save_warnings": ["Node not found for path: form1[0].#subform[6].P8_Line4_A_PageNumber[0]"],
  "notes": "AcroForm values persisted and re-read correctly; XFA 'Node not found' warnings are benign."
}

Browser-attempt outcome shape (why the naive path is rejected):

{
  "success": false,
  "method": "browser-page-fill",
  "form_fields_visible_in_dom": 0,
  "error_reasoning": "Chrome renders the I-539 in the PDFium plugin. browse snapshot returned RootWebArea + scrollable html only (urlMap about:blank), zero form refs; viewport is solid black. page.fill()/evaluate() have no DOM targets. Use the PDF.js annotationStorage method instead."
}