Right size your requests. Don't think of it as it "writes 5% better code," but more like it can handle the next 5% of previously unsolvable tasks. If Composer 2.5 is already doing well for your work, you don't need to use a more expensive model! But when you have a problem that's too hard you can go up
A clean package.json is no longer evidence that nothing runs.
The mere presence of binding.gyp is enough for code to run at install time. No scripts block needed. Payloads can hide under any field name, at any depth. The sandbox around it can be escaped. And node-gyp pulls in files automatically that nothing even points to.
The latest Miasma variant used binding.gyp. We dug deeper and found it goes much further.
@aarondfrancis ask it to draw the same diagram at 4-5 different levels of "resolution" - maybe it'll produce something good amidst all that progressive disclosure
I’m launching /omnifs today.
/omnifs is a projected filesystem that turns APIs, services, and data sources into local files and directories you can read with ordinary tools.
It is a bet that the filesystem is still the most durable interface in computing.
Grateful and flattered to have @luceboxai as the first advertiser on https://t.co/2hx2V3euUv
Such a good fit for the site and a cool project from @pupposandro and company.
Apple Watch modoki recreated with pure CSS + JS
originally made in 2015 with WebGL, beautifully ported to vanilla DOM + CSS in 2025 by @clockmaker
Check the live demo here: https://t.co/YMrVdqPZh4
I consistently find AI severly lacks taste.
Whether its in API design, design in general or overall architecture.
Taking the human out of the loop just sounds like a shortcut to mediocre and sad products.
I'd never want to ship something with confidence that way
We've added a new harness!
Cursor Composer 2.5 is live in Conductor.
It's fast, precise, and cost-efficient. And when I say fast I mean _really_ fast. Excited to hear your thoughts!
@charlieholtz How do you find it compares to GPT 5.5?
Any plans to build an intelligent router (like bitrouter) that can choose the model based on thinking and effort required, and can learn to route better with time and user feedback?
@geoffreylitt > "worried about cloud LLMs seeing my data"
might try out OpenAI's Privacy Filter running locally before anything sensitive is sent out