your research paper:

llm generated agent.md files are harmful. handwritten ones on average don't have an effect on ability to complete a task

thread article:

agents keep signing off on commits wrong. we added it to the agents.md and now they don't

benchmarks notoriously don't measure the likelihood that the results would be merged by maintainers, just that they passed a test

the old style "repo map" agents.md files that every tool used to create is no longer useful, but of course things that would otherwise be repeated prompts are still good. I cut the ones at work down from hundreds of lines each to tens. without those last few lines the amount of iteration I do - both with agent and when reviewing PRs - goes up significantly

you can also buy a nice USB/Bluetooth dac instead of the inline ones that tend to be more fragile. better quality than an internal one and the flexibility of Bluetooth if you want it. generally a little bulky but if you already have wired headphones I don't think it's significant

I've worked at several places that didn't have formatters when I started. they did by the time I left. you can incrementally adopt them and if it's automated most people at worst don't care. advocate for things you want

reassignment and hoisting are the significant ones. behavior around this does just seem more intuitive than otherwise when it comes up, so I think telling especially new devs to use const arrow fn everywhere but classes is a reasonable rule

hate to break it to you but it behaves like a variable either way. function just behaves closer to a var variable. const fns are less like variables since no assignment. intellisense/devtools all show them just fine. it really is just a minor aesthetic difference on the definition

tldw: https://github.com/Lassulus/wrappers

I find it useful since I'm normally turning it off to prevent it from connecting to something right now (speakers so someone else can connect etc), and I never have to think about turning it back on

all of their examples are pre ai anyway. it's almost impossible to change a core data model thing without ai too.

all the legacy codebases I've worked in have very much been shaped by the original abstractions they decided on. as much as I wanted to change them, there wasn't really a way to do so because of the scale and backwards compatibility requirements

this is neat but also is a closed source extension. I use https://github.com/Wilfred/difftastic which doesn't do quite as much and doesn't integrate outside of git's diff, but is still significantly better than nothing

fwiw, my state's mobile id app doesn't even ask for the location permission. so maybe some, but it's not universal

I've never used it but this one seems like the most complete currently, and it'll tell you which tests fail.

proton has support for quite a few kernel level anti cheat now, although it has to be explicitly allowed by the dev. needs to be run via steam I think, but you can add non steam games if you got them elsewhere

no fat, mixes into things better

fwiw I've used it pretty extensively on screenshots of text I keep getting sent at work, so far I haven't noticed any mistakes at all. may just be the type of images though