Install them into Claude Code, Codex, or Cursor — the AI you already have — and they organize your EB-2 NIW or EB-1A self-petition, your O-1A petition kit, or theemployment-based I-485 that follows, end to end.
Everything stays in a folder on your computer. No account, no cloud, no API keys.
Free, forever. Published software you run yourself — you decide, prepare, sign, and file. The maintainers never work on individual cases.
┌──────────────────────────────────────┐ │ your agent + an OpenNIW skill │ reads, organizes, │ (niw/eb1a/o1/i485 · any major agent)│ drafts — you decide └──────┬─────────────────────┬─────────┘ reads/ │ │ opens at structured steps writes ▼ ▼ ┌────────────┐ ┌──────────────────────────┐ │ your case/ │◄──┤ local browser pages │ the HANDS — intake, │ your files │ │ (run on 127.0.0.1 only) │ benchmark, citation │ = the only │ │ + deterministic PDF fill │ picks, forms wizard │ storage │ └──────────────────────────┘ └────────────┘
Paste your Google Scholar link. Your agent fetches your record, downloads your papers, benchmarks you against 10,999 approved cases, and lays your record out under each published criterion — Dhanasar prong-by-prong (NIW) or criterion-by-criterion (EB-1A / O-1A), with what supports and what undercuts each one. It does not decide whether you qualify, and it does not tell you whether to file — you do.
Draft options for the case frame — the canonical NIW endeavor sentence, the EB-1A field + claim frame, or the O-1 petitioner structure and itinerary scope — with what each emphasises and what each would need in evidence. You choose and confirm one; only then is it fixed, and every later document quotes it verbatim, because rewording after filing is a material-change risk.
A personalized checklist plus the citation pipeline: every citing paper harvested, screened for independence, verified in full text, scored by depth of use — you pick the best ~10 in a browser page.
Support letters plus the petition letter, drafted for your independent review in the structure these filings conventionally use — a Dhanasar three-prong brief (NIW), a Kazarian two-step brief with a Final Merits section (EB-1A), or the petitioner support letter + consultation package (O-1A) — every claim bound to an exhibit.
NIW: a browser wizard verifies the 61-field answer set card by card and generates the real I-140 and companions. EB-1A / O-1A: precise field guides for the I-140 (E11) and I-129 — identity numbers are never guessed, in any category.
A mock-officer red-team pass, then the filing package in USCIS-recommended assembly order — with current fees and the correct filing address for your category, state, and premium choice.
If USCIS pushes back: upload the RFE letter and get a deadline-driven response workflow — each challenged point mapped to the evidence that answers it, an evidence-gap loop, fresh highlighted citation examples, new letters, a supplemental statement, and the assembled response package. Works even if your original petition wasn't prepared with OpenNIW.
After the petition is approved: the separate i485-adjustment skill handles the employment-based I-485 — eligibility gating, status-history assembly, the document checklist, and part-by-part form guidance for everyone in the family. It is deliberately an assembly tool, not an eligibility tool: it stops and sends you to a licensed attorney on inadmissibility questions, any criminal history, any overstay or unauthorized work, and the decision to actually use an EAD or advance parole.
npx skills add HHHHHejia/openniw
Works with Claude Code, Codex, Cursor and 70+ agents (when the installer asks, tick your agent — e.g. "Claude Code"). One command offers all four skills: niw-petition, eb1a-petition (beta), o1-petition, i485-adjustment (beta). Manual install: copy .agents/skills/niw-petition from the repo into ~/.claude/skills/.
mkdir my-case && cd my-case claude # or your agent of choice > 帮我准备 NIW 申请 / evaluate my EB-1A case / help me file my I-485
The skill opens a browser intake page for your links and files, then your agent takes it from there — stage by stage, resumable anytime. A small helper (pip install openniw) is installed automatically for the browser pages and PDF filling.
The terminal above is the supported path and needs nothing else. But if that is the part that puts you off, there is a desktop app (beta): your agent on one side, the form pages on the other, a live stage bar across the top, and the statistical evaluation one click away — one window, no terminal.
It changes nothing about how any of this works. It starts the very same claude or codex you already installed, under your own subscription, on your own files — it never sees a password, a token or an API key, and it runs no server.
You do not have to set it up yourself. On a new case the skill asks once — terminal or window? — and if you want the window, your agent checks for Node.js and installs it for you. Switching later costs nothing either: your case folder carries the whole session, so you reopen it in the window and continue where you were.
# or by hand: git clone https://github.com/HHHHHejia/openniw ~/openniw cd ~/openniw/desktop && npm install && npm start
It runs from source rather than a downloadable installer: a double-click app has to be code-signed or your OS shows "unidentified developer" warnings, and we are not paying a certificate authority to vouch for us. Source is auditable.
Nothing, beyond what you already pay: the project is MIT-licensed and the AI is your own subscription. Your actual USCIS filing fees (2026: I-140 $715 + $300 Asylum Program Fee for NIW / EB-1A self-petitions; O-1's I-129 $1,055 + an employer-size-based fee; optional premium processing $2,965) go to the government, not to us.
OpenNIW is a fully open-source, free, public-interest project (开源利益众生). It gets better through three kinds of contribution:
Issues and PRs welcome — form mappings, new browser pages, translations, anything on the issue tracker.
Filed with or without us? Use the anonymous data-point form (field, citations, timeline, RFE, result — no names, no case numbers) — every real data point sharpens the benchmark for the next person.
Prepare NIW cases professionally? Your frontier experience — what draws RFEs now, what wording holds up — can be folded into the playbooks. Credited or anonymous, your choice.
The eb1a-petition and o1-petition skills now ship in the same repo — built from USCIS primary sources, MIT-licensed open materials, and our approved-case dataset, not from firsthand filings. That's exactly why they need you: if you have actually been through an EB-1A or O-1 case — applicant or practitioner — reviewing their playbooks is the single most valuable contribution right now. And H1B (or any other category) is still waiting for its veteran: the whole framework is reusable. Reach out — let's build it together.
Before you write: we answer questions about the software — installation, bugs, how a feature works, where a number came from. We do not answer "do I qualify", "should I file", "how should I word this" or "how do I answer this RFE", and we do not review CVs, drafts or notices — for anyone, at any price (maintainer policy). For your actual case you need a licensed immigration attorney or a DOJ-accredited representative.
Contact — GitHub issues
Compare your citations and papers against 10,999 publicly posted approved cases — by field, over time, with processing-time simulation. Runs entirely in your browser; nothing you type leaves your machine.
Open the statistical evaluation →