Repository navigation
Replies: 1 comment
|
Camoufox documents an experimental remote server, although it speaks the Playwright protocol, not CDP: Remote Server. The documented launch command is: python -m camoufox serverThen connect to the WebSocket URL it prints using Playwright's Firefox client: from playwright.sync_api import sync_playwright
with sync_playwright() as p:
browser = p.firefox.connect("ws://HOST:PORT/PATH")
page = browser.new_page()
page.goto("https://example.com/")
result = page.evaluate("() => ({ userAgent: navigator.userAgent })")
print(result)
browser.close()Replace that small expression with your probe entry point. Your existing For the comparison itself, I would also separate checks that require Chromium-only APIs from checks of inconsistent exposed values. A browser not implementing a Chromium feature should not automatically receive the same result as an inconsistent implementation of it. |
Uh oh!
There was an error while loading. Please reload this page.
I scored eleven Chromium automation stacks with an open-source CLI today and published the numbers, and Camoufox is deliberately absent from that table - I want to explain why and ask whether it has to stay that way.
https://liarjs.dev/blog/stealth-stacks-measured/
My harness attaches over CDP (
npx liarjs --cdp http://127.0.0.1:PORT --page https://example.com/). Camoufox is Firefox-based and does not speak CDP, so I could not score it the same way as the Chromium stacks. Rather than produce a number by a different method and put it in the same table, I left it out and said so in the post.The question: is there a supported way to launch Camoufox that exposes an endpoint a third-party scorer can attach to, or a recommended pattern for running arbitrary probe JS in a page under the real Camoufox config? If so I will run it and publish the result whatever it says.
Context from the run that may be relevant to this project's thesis, since it is about JS-level patching:
puppeteer-extra-plugin-stealthscored below plain puppeteer (18 vs 42). It removesnavigator.webdriverand introduces four cross-surface inconsistencies: WebGL reports an Intel GPU whilenavigator.gpureportsapple; a Web Worker returns differentlanguages/hardwareConcurrency/webglRendererthan the main thread because the override only lands on the main thread; the renderer names a desktop GPU while the extension list carries mobile-only compressed texture formats; andnavigator.gpuadvertises features from the adapter Chrome actually opened.patchrighton the Chrome channel and plainplaywrightplus--disable-blink-features=AutomationControlledscored identically, with not one check differing in status.Caveats are in the post: one run per stack, passive layer only, and the machine is Apple Silicon so part of that GPU mismatch is a property of the host and may not reproduce on Intel.
Not asking to be listed anywhere - I would just rather have Camoufox measured than absent, and I cannot do it without a way in.
All reactions