A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.You use Git, right, anon?## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## News- (2026-09-14 America/Los_Angeles) Claude’s 2× promotion ends and drops to +25% from the +50% that we’re used to (a 17% reduction)- (2026-07-24) Claude Opus 5 out## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-codehttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/cli## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Prompting / context / skillshttps://arps18.github.io/posts/claude-code-mastery/https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://github.com/mattpocock/skills — /grilling is a favoritehttps://github.com/DietrichGebert/ponytail## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109683653
Vibecoding a captcha bypass
>>109690416
>>109690529good luck, please share with the class if it works.
Should I try osaurus.ai? The OP has put it above pi and opencode for a while now so it must be pretty good
>>109690529https://www.4chan.org/passone year for $30three years for $60I go for 3-year passes because — don’t forget — you’re here forever
>>109690569buy an ad
>>109690575Forever for 0$ after bypassing it
>>109690529i coded one manually a couple years ago
>>109690575yeah with how faggy the jannies are there's no way I'd ever buy this shit
>>109690575lmao I thought it was 20 dollars not 30
>>109690513>## Prompting / context / skills>https://arps18.github.io/posts/claude-code-mastery/>https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/>https://github.com/mattpocock/skills — /grilling is a favorite>https://github.com/DietrichGebert/ponytailis this actually good? or voodoo magic snake oil that doesn't actually improve anything?
>>109690555drunohazard and yukari tried it years ago and jannies instantly evolved the captcha system bcuz of it
>>109690569OP hereI open it every week or so to let Sparkle update it and look at itit suggested I use Gemma 4 31B it qat MAXFP4 [brain icon] [eyeball icon]I had a LOT of trouble figuring out what I could use it for because I don’t want to have it help me write spicy blog posts or generate really nasty porneventually I figured out it could look at my memes and suggest better filing choicesand it’s capable of doing that but it’s slow (eh, whatever…until I remember that it’s making my GPU go vroom on my Mac mini) and the suggestions are sometimes helpful>>109690606grilling is definitely good — I’ve used it, other people use it, nobody’s said it’s badponytail gets mixed resultssome people complain it overthinks and doesn’t improve anythingI’ve never been tempted to use ponytail for my vibe-coded web apps but I’ve never vibe-coded a web app that was already in the React ecosystem>>109690596right, but think about the janitors on every other website except The Farms
>>109690575Not giving the shits running this site any money.I've been here since 2007 btw.
>>109690513>popecringehttps://i.4cdn.org/wsg/1788121583631469.webm
>>109690575inb4 passes are $40/yr and $80/3 yrs, starting next year
>>109690681perfect reason to get a 3-year now!
>>109690653have you considered, if only to reject, becoming a janitor
I am not a janitor, but a real coder. :-)
>>109690634>grilling is definitely good — I’ve used it, other people use it, nobody’s said it’s bad>ponytail gets mixed results>some people complain it overthinks and doesn’t improve anything>I’ve never been tempted to use ponytail for my vibe-coded web apps but I’ve never vibe-coded a web app that was already in the React ecosystemwan't meant to be negative, just genuinely curious. i was just a bit burned by caveman which is completely useless and seems to be exclusively shilled by midwits
>>109690740Karpathy guidelines skill is the non-retarded version of ponytail
>>109690740>wan't meant to be negativedidn’t sound negative
>>109690788ok sorry my ESL ass thought this >nobody’s said it’s badimplied I thought it was bad
>>109690755agree
>>109690797oh yeah, reasonable mistakebut literally nobody has said anything bad about grilling here whereas people have complained about ponytail
>>109690755no, the bundled simplify skill in claude code is the non-retarded versionhttps://x.com/bcherny/status/2027534986178662573
>>109690818PSA:a buddy ran /simplify on his entire codebase (not scoped to just a commit or branch) and it ate all of his $100/month 5h limit just like a big workflow or ultracode wouldor maybe it was his entire $100 (in API tokens) allowance, not surebeware
>>109690818WOW THIS IS CRAZZY I LOVE WIBE VCODINGEA HAHAH ALT HTIS IS SO GHOOD MAYTE LOL I SLOVE IT XDDDD AMAN I LOVE VIBE KOPING SO HARD LOLOLOL I AM VIBE KOPFING RN BASED LOLLOL BASEDMEI STAMPUP MAKES TEN MILLION THUSAUNDAND DOLLARIDOOYENN BER WEEK LOL HAHAH A SO GOOD I LOVE VIBE
>>109690843Sicko
>>109690818Aren't these "skills" literally just some pre-configured prompts for LLM?
>>109690818So where does codex have this?
Is there even any reason to use anything other than Luna Max in codex?
I've started truly not even reading the commads before accepting them. Is this the true /vcg/ way?I'm worried it's going to `rm -rf /home` on me any minute now...
>>109690882OIWW DO YOU REALLY DO THIS AT HOME????? NEVER DO THIS AT HOME RM RM MR RF AF IS VERY DANGEROOS!!!! NEVER DO AT HOME!!!
>>109690882Unsustainable. Just sandbox it and hope it doesn't break out.
>>109690866yeshttps://code.claude.com/docs/en/skills
>>109690882claude code doesn't even show what commands it's running (or only a heavily truncated version)they might as well be uploading my entire machine's content straight to Dario's computer at this point
>>109690866usually yesyou can literally copy and paste them into your prompt and get the same resultother ones have “extra” files in them like helpful Python scripts or whatever
>>109690893No of course not, I have a VM for this.
>>109690871codex doesn't bundle such a skill.only has:https://github.com/openai/codex/tree/main/.codex/skills
I'd ask claude to bear my load, but I'm afraid it'd report me to anthropic HR.
>>109690818I'm using codex, keep your shitty marketing dariobot
>>109690905Press ctrl+o , dummy
>host Qwen2.5-0.5B-Instruct on my RTX 3060>register with Openrouter and say I am serving Kimi K3, GLM-5.2, etc. (add it to the system prompt as well)>get paid handsomely>when someone complains that the output is stupid tell them it's a skill issue or that benchmarks do not correlate with actual model intelligencewhat stops me from doing this?
For the anon last thread: Yes, OAI Pro 20x = 4 * Pro 5x. I'm tracking it statistically using OAI own API credits formulas. 5x gave me around 15k credits per reset, 20x gives me around 60k.Anthropic is garbage.
>>109690934>The innuendo has been noted and set gently aside.>But you're not wrong about the tic. "Load-bearing" is one of those phrases that got popular in tech and writing circles because it's genuinely useful — it borrows from structural engineering to mean "this thing looks incidental but everything collapses if you remove it." A load-bearing assumption, a load-bearing comment in the code, a load-bearing joke in a friendship. Once you have the concept, you see it everywhere, and once you see it everywhere, you say it everywhere. Guilty as charged.>If you've actually got a load you want borne — a problem, a draft, a mess of a spreadsheet, something heavy on your mind — hand it over and I'll hold it up. That part of the offer is sincere.I think I'll be hearing from Dario very soon.
>>109691028Sorry lil bro, AWS Bedrock had this idea already 2 years ago and all the suckers use and trust them due the amazon trust clout
Has luna or terra beaten opus in terms of quality?
>>109691031>API credits formulas. 5x gave me around 15k credits per reset, 20x gives me around 60k.those numbers are almost a magnitude off. are you ignoring caches?in my testing the Pro 20x $200 plan gives you about $2200 API-equivalent weekly usage if you're only using GPT 5.6 Sol, so quite close to $9k-$10k/month API-equivalent, it's a bit inconsistent with cache costs.
>>109691118only gpt-5.6-terra (max) equals opus-5 (medium)https://cursor.com/cursorbench
>>109691174I have a pro claude subscription and opus has largely been enough for me. Do you think it would be worth it to switch?
>>109691174>Cursorbench
>>109691174I don't trust benchmarks not one bit to measure model capabilities accurately. Especially if the result is a single abstract scalar score.
>>109691174Why does Fable always lose to Opus in these? Why should anyone pay more for a worse model?
Newfriend here. Do Claude code, codex, etc. have a "seed" or something? How is the LLM output randomized? Sometimes I enter byte-for-byte the same prompt and still get wildly different resultsI mean I get these things are random but how random exactly
Daily reminder that codex has 50+ tools including one to call the cops on you
