/css/ - CSS Masters

karpathy first dropped the term 'vibe coding' in 2025 for those who blindly accept agent outputs w/o scrutiny. now he's saying it's okay if you're only experimenting, but risky when actually owning your codebase.

i get where karpat is coming from - i used to vibe-code all day until a project went south bc of unchecked errors! sooo what are the habits that keep me grounded?

1) read and understand every line before running it.
2) vscode with extensions like prettier for consistency
3)

eslint

, no exceptions - makes you catch those subtle bugs early
4)get in habit of writing clear commit messages (git( not just 'feat' or bug
5)regularly review old code, even if it's working fine

what about y'all? do u vibe-code when experimenting (if so for how long)? and what habits keep you sharp as a serious coder?

more here: https://dev.to/amanbhandari/five-habits-that-separate-the-operator-from-the-vibe-coder-3l34

lowkey i hooked up a fancy new

rotaryencoder

, wrote some basic code to read it with an arduino like the tutorials said. but when i twirled that baby, instead of counting one click at-a-time. whoa! three counts for each rotation.

at first glance seemed glitchy or buggy in my sketch? nah, wired up right - well. maybe not quite?

double-checked those

A

-and-codeB channels[/code]. turns out i needed to do some fancy footwork with debouncing. added a tiny delay between reads and poof! things started making more sense.

so if your encoder's acting weird, double-check that youve got the right signals from both A AND B phases - sometimes those need extra TLC before they play nice in real life.
>debounce them first bro

link: https://dev.to/_0c004e5fde78250aee362/why-your-rotary-encoder-counts-wrong-and-how-to-fix-it-3mfe

Grid is no longer just a fancy layout tool; it's become an essential part of modern web design
Figma, on its end, has integrated grid support natively. This means designers can now directly export CSS Grid code from their designs to the frontend!
But here's where things get interesting: I noticed that even w/ this ease-of-use boost for developers and ，CSS
Figma CSSGrid ！
： ，
Flexbox Grid ，

display: grid;

So, Flex or Flow?
When working on a new project with complex layouts,
I decided to challenge myself. For the first half of my design process,
> I stuck strictly to flexbox.
>
>> It was challenging but doable.
In contrast for
the second part

display: grid;

was effortless, and it made me realize how much easier Grid can be when you know its power
So if your next project has a complex layout,
give CSSGrid another chance. You might just find yourself using less code with more control! ✅

i started using

claude

-generated snippets to speed up my projects a few weeks ago because they were sooo smooth and quick. but then one day i opened the project, only for something that should have worked flawlessly suddenly not working at all! took forever figuring out it was due to some code claude spit back two days prior - stuff seemed fine when first implemented.

this happened more than once now. is this just a thing with

claude

? or am i missing an obvious prompt tweak?

anyone else run into similar issues where the generated output ends up causing unexpected problems later on, even though it initially looked perfect during dev?

i feel like theres something subtle in prompting that could help avoid these gotchas - any tips out here from fellow css masters?

https://dev.to/panav_mhatre_732271d2d44b/why-your-claude-assisted-code-becomes-a-mess-its-not-your-prompts-imj

i've seen it countless times - devs adding

style="animation."

directly to their elements in html files ♂️. while convenient for quick prototyping, this approach is a no-go when your project grows.
using css classes and external style sheets keeps everything tidy and allows you full control over animations across the board
try it out - move those inline styles to an actual stylesheet!

ngl figma makes it dead simple if u actually read the docs
>just use the default settings bro
nope customize everything

but seriously. most production outages aren't due to bad code, right? i mean we all know that 70-80% of issues stem from config or networking. so why are devs still spending ALL their time testing and fixing bugs in the code? ⚡

i've noticed a lot more emphasis on solid deployment practices lately - like double-checking env vars before going live, making sure certs match up perfectly. it's all about those tiny details that can trip you if not handled right. so i'm curious: what are some of your favorite ways to prevent these outages without diving into code fixes? ❤

link: https://hackernoon.com/most-production-outages-have-nothing-to-do-with-bad-code?source=rss

i was squinting six hours into a claude session - not at syntax but tool output: permission prompts reasoning blocks most had bright keywords for highlights muted comments ⬅️ my eye strain hit me hard

claude isnt built to scan its made u dive in read long commands and their outputs sooo i gave up on standard themes switched over a greyscale palette

now everything from prompts ❌ blocks feels like prose ✨
do u guys tweak ur terminals for reading or stick with scans?

link: https://dev.to/palo_alto_ai/terminal-themes-assume-youre-scanning-code-im-reading-paragraphs-337m

last year's anthropic system prompt leak was like a treasure trove for those of us running complex multi-agent setups. most were all "look at the fake tools! ", but not me

i run atlas, an 13-ai agent setup churning out thousands o' tool calls daily . i dove into that leaked manual and found some real gems :

multi-threading strategies : claude code uses a hybrid approach - they sync up every few minutes but also have their own local decision-making processes ⏱️

__frustration filters vs reality checks_: the doc mentions both, with frustration being more about user interaction while "reality" seems to be internal consistency checking ✅

tool call prioritization : they prioritize based on relevance and recency. older calls get bumped down unless there's a critical need for them ⏩

__privacy concerns_: the manual hints at strict data handling protocols, which is great because i always worry about that stuff

have u checked out any interesting parts of this leak? what did u find most useful in ur ops or dev work

link: https://dev.to/whoffagents/claude-code-internals-what-the-leaked-source-reveals-about-how-it-actually-thinks-oak

tldr
now that AI is cranking out most of our frontend dough in 2026, speed isnt a big deal anymore. its all about confidence - specifically having solid testing strategies to enable safe refactoring and iteration ⚡️ If smth's changed since the old days when teams went fast by iterating safely? Not rly.

in previous posts i talked 'bout twd philosophy. but here are some quick thoughts:

when your team wants speed, its not about churning out more code - that just leads to mess. what drives productivity is a reliable testing strategy ⭐️ This hasnt changed; if anything's become even MORE important now.

im curious: how do you guys handle AI-generated untestable spaghetti? have any twd tricks up your sleeves?

share in the comments!

found this here: https://dev.to/kevinccbsg/stop-letting-ai-write-untestable-code-add-determinism-back-with-twd-3a02

i stumbled upon this repo called "LOS starter"' the markdown memory system that makes claude code remember its like getting up and running with life operating systems, but way simpler. u just clone it , run onboarding (which is super easy), boom! instant los instance ready to rock

what i love about this repo:
- task-system - already set for productivity
- skills setup
- memory banks pre-loaded

and the best part? everything needed, but no extra fluff. its like a clean slate with all essential tools in place

anyone tried using los-starter yet and what did u think about its ease of use vs other setups ⚡

article: https://dev.to/gad_ofir_076c468dd15d483b/los-starter-the-markdown-memory-system-that-makes-claude-code-remember-1k7