They assume they need to write tests to ensure a feature is removed!

They assume they need to implement compat whenever an API changes!

They assume they must implement something from scratch when it may already exist!

Instead of interrupting their work and issuing corrections via the chat, I figured we could pick out these implicit assumptions and show them as an autogenerated form-style UI in the lower right. That way as I tab around to different apps I can keep one eye on the stream of (potentially incorrect) assumptions my agents are making, and correct them with one click instead of having to type in correction messages.

It works with Codex in the ChatGPT desktop app or CLI in iTerm.app for now, in the future I might add Claude support or support for Ghostty / other terminals.

Comments URL: https://news.ycombinator.com/item?id=49288359

Points: 2

# Comments: 1

I recently did a complete rewrite of our browser driver codebase by deeply analyzing playwright, puppeteer, selenium, and the chromium source code, relying heavily on auto generated diagrams to decide which parts to read myself / dive deeper on. Yes they occasionally leave things out if the underlying codebase hides the complexity or has no docs, but they're great for thoroughly documented mature codebases.

I highly recommend reading the singlefile source or https://archiveweb.page/ to see how they handle closed shadow DOMs, cross-origin iframes, websockets, media urls, deduping large assets, etc.

A11y-tree alone is not enough for many sites because lots of auth stuff happens in OOPIF frames that need special handling/stitching/interactive element filtering.

There's also the issues of many captchas around auth stuff being implemented using canvas elements (that are hard to instrument for browser agents without relying on CUA). Can their on-device 3B model really handle accurate CUA driving? I guess we'll see...

I hope Insta360, DJI, and many more competitors spring up out of their ashes. Assuming fast microsd cards for writes, how much DDR RAM do they really need to buffer videos during recording? 4gb? 8gb?

- https://www.music-map.com/ - https://everynoise.com/ - https://chartmetric.com/ - https://musicroamer.com/ - http://davidmckinney.com/app

But they're all kind of generic, I would love to see something more genre-specific with additional historic context and personality.

Libraries like Playwright and even Google's own puppeteer resorted to hacks like forcing users to foreground pages manually, or pretending "last opened == foreground".

After writing brittle state sync logic and workarounds for the last 2 years I got tired of dealing around this issue, so I contributed a fix upstream into Chromium!

Comments URL: https://news.ycombinator.com/item?id=48214533

Points: 4

# Comments: 1

But I've struggled with the IO parts, it seems every system is so different, it's hard to live without LLVM-IR as the middle layer abstracting away compiler and target differences.