Top 200 ReposEXPLORER
Browse agents, categories and pages
Back to results
Reported by sourceThe source says this happened. Top 200 Repos has not reproduced it.

Fill out repetitive web forms with Skyvern's vision-based browser agent

Skyvern drives a browser with an LLM and computer vision to complete forms, such as quote requests or contact forms, on sites it has not seen before.

The problem

How it works

Skyvern is an AGPL-3.0 browser automation project that extends Playwright with AI actions for clicking, filling, selecting, uploading and extracting. You give it a goal and the data to enter, and it reads the page visually instead of relying on fixed selectors. The README's examples include contact forms, insurance quote requests, government form filing, job applications and invoice downloads; it installs with pip and a quickstart command, or with Docker Compose.

Results and limitations

As a vendor claim, the maintainers report 64.4% accuracy on the WebBench benchmark and say Skyvern does best on write tasks such as form filling and logins. That figure is theirs and is not a test of your own forms.

Before you try it

A benchmark score in the sixties means many runs still go wrong, so review every form before it is submitted. Skyvern supports logins, 2FA and password manager integrations; keep credentials in such an integration rather than in prompts, and automate only sites whose terms allow it. The managed cloud adds CAPTCHA solving and anti-bot tooling that the open-source edition leaves out; do not use automation to get past a site's bot checks. Results depend on your setup and were not independently reproduced by Top 200 Repos.

Tools and repositories

Tools
Skyvern (pip or Docker Compose) · An LLM provider API key · Playwright browser
Prerequisites
Python 3.11 to 3.13 for the pip install, or Docker; on Windows the pip route also needs Rust and the Visual C++ build tools.

Try it yourself: Fill a form and stop before submitting

Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].

Preview
Open [Form address] and fill in the form using only this data: [Data to enter]. If a required field is not covered by the data, leave it empty and list it instead of guessing. Do not create an account, log in, upload any file other than [File allowed for upload], or attempt CAPTCHAs. When every field you can fill is done, take a screenshot, stop before the final submit button and wait for my confirmation. Report each field you filled and the value you used.

Keep the stop-before-submit step until you have checked several runs on the same form.

Sources

  1. Skyvern-AI/skyvern README (opens in a new tab)

    Project documentation · github.com · Undated · Source checked 11 Nov 2026, 22:04 UTC

    What this source supports
    • Automates browser workflows with LLMs and computer vision
    • Extends Playwright with AI-assisted click, fill, select and upload actions
    • Examples include contact forms, insurance quotes, government forms and job applications
    • Installs with pip and skyvern quickstart, or with Docker Compose
    • Claims 64.4% accuracy on WebBench, strongest on write tasks such as form filling
    • AGPL-3.0; anti-bot measures are only in the managed cloud
Report a problem

Reported by source

Let an assistant test QA tickets in a browser via Playwright MCP

Peter Eysermans connected Claude Desktop to Playwright and GitHub MCP servers so it could walk through features from tickets, report results and draft reusable tests.

madewithlove.com · 22 Jul 2025microsoft/playwright
Reported by source

Prepare a tax return with an agent and compare it to an accountant

Kyle Corbitt worked through his 2025 US taxes with Codex, which researched rules, gathered documents and wrote a custom tax engine while a paid accountant did the same return.

corbt.com · 7 Mar 2026openai/codex, microsoft/playwright
Planned, not done

Try a desktop-operating agent safely in Anthropic's computer use demo container

Anthropic's computer-use-demo runs Claude inside a Docker desktop that you watch in the browser, with a chat panel, so you can test screen, mouse and keyboard automation away from your own machine.

github.com
Planned, not done

Watch competitor pricing and product pages with changedetection.io

changedetection.io checks chosen parts of web pages on a schedule, tracks prices and restocks, and can use an LLM to summarise changes or ignore ones that do not matter.

github.commicrosoft/playwright
Reported by source

Compare products across open tabs with Claude in Chrome, then buy yourself

Claude in Chrome can read and compare what is open in your tabs and pull details into a table; Anthropic’s own red-team numbers explain why it asks before purchases and why you should finish checkout.

claude.com · 25 Aug 2025
Reported by source

Let an AI agent complete a multi-step task in a real browser

Describe a web task in plain language and let Browser Use drive a Chromium browser to finish it and return structured output.

github.combrowser-use/browser-use