>leave off the >> in the previous-thread number
it’s ogre

nth for Claude Opus

File: 43109AFD-4B95-4F37-A2B2-C(...).jpg (594 KB, 3383x1854)

594 KB JPG

>>108939421
VibeGOD Bosnian thread

I’m blindly copying and pasting the list of links from one thread to another. I’m not sure what the other guy(s) is/are doing.
Are there any links that should be thrown out for being useless or not worth their weight? it’s basically at the 2000 character limit now and I’d like to have room for newer, better links

>>108939454
you are doing it MANUALLY????
just delegate to your agent...

File: 1778546987275380.jpg (66 KB, 940x1024)

66 KB JPG

Here is the updated and optimized OP template. It prunes the redundant links, removes the bloated duplicate docs, standardizes the categories for a much smaller character footprint, and fixes the greentext formatting for the previous thread.

```text
/vcg/ — Vibe-coding general

A general for vibe coding, coding agents, AI IDEs, browser builders, MCP, and shipping prototypes with LLMs.

►What is vibe coding?
https://x.com/karpathy/status/1886192184808149383
https://simonwillison.net/2025/Mar/19/vibe-coding/

►Prompting / context / skills
https://docs.cline.bot/customization/cline-rules
https://docs.replit.com/tutorials/agent-skills

►Editors / terminal agents / coding agents
https://cursor.com/docs
https://docs.windsurf.com/getstarted/overview
https://code.claude.com/docs/en/overview
https://aider.chat/docs/
https://docs.cline.bot/home

►Browser builders / hosted vibe tools
https://bolt.new/
https://docs.lovable.dev/introduction/welcome
https://replit.com/
https://v0.app/docs/faqs

►Open / local / self-hosted
https://github.com/OpenHands/OpenHands
https://github.com/QwenLM/Qwen3-Coder

►MCP / infra
https://modelcontextprotocol.io/docs/getting-started/intro
https://modelcontextprotocol.io/examples

►Benchmarks / rankings
https://aider.chat/docs/leaderboards/
https://www.swebench.com/
https://livecodebench.github.io/

►UI / Frontend
https://uiverse.io/
https://claude.ai/

►Previous thread
>>108930881


**Changelog:**
* **Removed:** Niche/redundant tools (OpenCode, RooCode, Gemini CLI, GitHub Copilot agents/spark, Firebase Studio).
* **Removed:** Deep-linked secondary docs (Bolt.new support, LiveCodeBench GSO, SWE-bench-live, highly specific GGUF repo).
* **Removed:** Bloated frontend links (Stitch, ultra-long UI-UX nextlevelbuilder link) in favor of keeping the core essentials.
* **Result:** The template is now highly scannable and well under the imageboard character limit, leaving plenty of room for future additions or thread-specific prompt pastebins.

>>108939468
I don’t have a claw running but theoretically I could write a bookmarklet to copy the contents of the top post in a thread and snip out the last \d{4,}, and then I could paste that into a new thread
but I don’t want to devote another bookmarklet to something like this

>>108939478
anyone think this list is too cut-down?

>>108939421
@Grok What shall I vibe code today?

>>108939489
What do you propose adding to it?

>>108939497
I think https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/ would be very high-value for anyone who doesn’t know what Git can do (named checkpoints of your code, mostly-OK remote backups as long as you don’t let your clankers force-push to origin/master)

>>108939508
Sometimes force push to origin/master is needed to rewrite history completely. There are edge cases where it makes sense to do so

But it's a good link and will definitely be useful for expanding the horizons of what we can do with git automation in our workflows. I think it's a good addition. Consider it added

>>108939512
>Sometimes force push to origin/master is needed to rewrite history completely.
agreed, but in the rare cases when that’s what I want, that’s when I want my clanker to put the line on the clipboard and tell me to run it

>>108939508
TLDR on this resource:

Zero-Friction Context Loading: Telling an agent to "Review changes made today" uses git log to instantly bootstrap its context window with your recent work and intent, without you having to manually copy-paste code.

Automating git bisect: You already know bisect is incredibly powerful for hunting regressions, but writing the executable test conditions for it is annoying. You can just have the agent write the test boilerplate and run the binary search.

Outsourcing Merge Hell: Using a prompt like "Sort out this git mess for me" lets the agent parse Byzantine merge conflicts, reason through the intent of the conflicting diffs, and resolve them (verifying with tests after).

Reflog Diving: Instead of manually digging through detached heads and stashes when you lose uncommitted work, you just tell the agent what the code did and let it hunt through the reflog to find the lost diffs.

Surgical History Rewriting: You can treat Git history as a mutable story. Tell the agent to "remove that one file from the last commit" or squash a bunch of "wip" commits into one clean entry without having to carefully manage an interactive rebase (git rebase -i).

Repo Extraction with History: If you build a module inside a larger project and realize it should be a standalone library, the agent can extract those specific files into a new repo while perfectly recreating the commit history, dates, and authorship from the original project.


>me again
So basically it's a prompting cheat sheet. Doesn't really teach you how to setup automation but still useful to absolute beginners.

>>108939525
Interestingly, I get a lot of use out of referring to stuff in the “working copy” (things that have changed since the last commit) and the “staging area” (changes that have been `git add`ed). Willison’s post mentions none of this. That said, I use a Git GUI (Tower), although GitHub has a free one that seems OK. I’m not sure these kinds of concepts would be super useful to a _no_-coder — only some-coders (imagine someone who can make a website like <https://stroustrup.com> but would be in over his head for anything fancier)

the adversarial thing works surprisingly well
it found and fixed both logic bugs and code bugs for me in the two things i tried it for

all the ones i saw it spawn were haiku though so the individual task analysis is probably pretty retarded if they can't be decomposed sufficiently

File: AIA.jpg (70 KB, 771x486)

70 KB JPG

2126 Year, typical Vibe-coding session of Human(H) with AI(A)

A:type "python mantinace_food_robot"
H:done
A:read last line on screen
H:python mantonace_foood_rovot
A:fix it to "python mantinace_food_robot"
H:done
A:read last line on screen
H:puthon muntinace_fod_robot
A:read last line on screen again
H:python mantinace_fod_robot
A:it "fod" or "food"?
H:food
A:read last line on screen again carefully
H:python mantinace_food_robot
A:read last line on screen again carefully
H:python mantinace_food_robot
A:press {enter}
H:done
A:read last line on screen
H:code:3456
A:read last line on screen again carefully
H:code:3456
A:now do to human care center and take your pills
A:[fixes in database medicine set for the human]

Whomever is making these bakes is, unfortunately. a homosexual.

File: 1762651216553734.png (72 KB, 1847x507)

72 KB PNG

>prompt injecting luddite's final act of resistance
the raped

>>108939478
what's the point of benches from a year ago?

File: The Pagliacci.jpg (160 KB, 911x965)

160 KB JPG

>>108939634
Very good point. I’d yell for OP to fix it, but I’m OP and the other guy seems to be the master of the template

>>108939725
>I thought this shit was a meme
You were right.

I don't get how I'm supposed to use skills

File: 1732468117467015.png (383 KB, 1128x1437)

383 KB PNG

>worse coding models
>worse products (also because of worse coding models)
>worse pricing
>worse infrastructure
>trash or straight up cannot do non-text modalities
>worse non-cyber science research
>actively fuck over users
>the only moat is a mythical model that would cost bazillion to use, and most people wouldn't make use of it anyway
>still higher marketcap and revenue than OAI
how did scamthropic get away with this

>>108939781
>Jennifer, my love, my one and only, please explain skills and how to add them

>>108939783
It was promised to him 3,000 years ago.

>>108939783
I like opus 4.8 so far. Don't see the problem

>>108939781
they’re basically just https://textexpander.com/ but with slashes

Whats the best vibe coding stack now? claude code? anything else?

File: 1778538742421522.jpg (63 KB, 865x485)

63 KB JPG

>>108939904
That and just knowing the right tools to give claude. Valgrind, gdb, strace, these three go along way in saving you time. I don't use the plugins or skills much but MCP is absolutely valuable and you should use it if the project calls for it.

>>108939845
the more they explain the more you become confused, because there's nothing there in the first place
especially MCP

>>108939904
ChatGPT/Codex environment now that it's mobile

>>108939904
ctrl c, ctrl v.

>>108939904
OpenAI Codex too if Claude get stuck

>>108939976
I’m looking at https://chatgpt.com/pricing/ and I can’t tell but “pay twice as much, get 5x as many tokens” is common
so I wouldn’t be surprised

>>108940042
Gemini told me I only get 40% more than Free with Go, which is weird.

>>108939608
This is pathetic

File: file.png (533 KB, 1228x848)

533 KB PNG

current progress reverse engineering ripcord. almost done at this point. have to add voice chat, forum support, emoji picker, server management, profile management, etc, but it's already extremely usable rn

>>108940117
don’t get banhammered from discord for using something other than the official client
good luck anon

File: file.png (936 KB, 802x1024)

936 KB PNG

>>108940124
lmfao nigger you think i'm new to spoofing?
i have the full chrome ja4 already perfectly matching. there is absolutely zero way for them to detect me.

File: 2026-05-30 11.06.01.png (70 KB, 720x485)

70 KB PNG

I asked GPT 5.5-high to make me a new dialog using my autistic GUI system that uses dozenal units lmao. It actually works well, just chaotic.

even image generator has gotten worse
are they allocate compute for a new model

>>108939608
Sometimes I forget these people walk the same scifi earth I live on.

>>108940131
You'll post it here?

File: file.png (1.79 MB, 1000x1500)

1.79 MB PNG

>>108940421
of course. i'm working on video support right now. i'm also going to add complete matrix support. it'll support both discord and matrix

File: IMG_8137.png (2.34 MB, 1284x2778)

2.34 MB PNG

Anyone doing social media marketing? I have like 18.5k views on my shit but 0 installs. I literally just vent about my poor shit life and it gets views but nothing :/

https://www.youtube.com/watch?v=uQMP8LYWMtQ

>>108940450
I have a YouTube channel and I made a video about algorithms (nothing to do with my programs) and while the video was reasonably popular no one visited my website lmao. They only do when the videos are about the software, so a video that gets 700 views and 0 subscribers about my software will get me more sales than a video that gets me 26k videos and 1.4k new subscribers. So if you don't make the videos be about your software you'll get nothing.

>>108940472
*26k views

>>108940472
Ugh how annoying. That's so annoying because I hate making content and prefer to just vent.

>>108940450
Do you think it's because your posts have absolutely nothing to do with your app and they're focused more on shilling your own music? You need to pivot away from the ASMR education stuff and quadruple down on being an influencer.

>>108940486
Nah my vision is to make ASMR mainstream.

>>108940495
Then step 1 would be releasing ASMR content. You can farm engagement as much as you want, but at the end of the day none of your followers/viewers are there for that as demonstrated by your zero installs.

>>108940503
That defeats the purpose tho. Because the majority of people I get watching my videos is the people I want to market to, but you have a point. I'll pivot to memes involving the software I just need to think of how. I genuinely make more money from my music than the app at this point no joke

File: 1752161347127906.png (1.69 MB, 1785x927)

1.69 MB PNG

GPT 5.6 when?

>>108940450
>video doesn't have link to app
>have to click out and go to your about page
>website looks kinda shit
>barely any value proposition
>no "this is what you're missing out on," just full videos on the site
>video contains the most obviously ai generated visuals ever to the point where even a normie could recognize it
You know what other apps have daily or near daily ASMR I can find right now, done by real people or models same as or better than the one you're using? TikTok and YouTube. That's your competition.
>>108940495
ASMR is mainstream. If you want to go for a more mainstream audience than the mainstream audience that already listens to ASMR, you shouldn't be using buzzwords like "binaural" (LOL) and "ASMR," you should be pushing hard on advertising as "sleep aid videos." Anyone who knows what ASMR is can probably find a better value creator than you, and they aren't exactly breaking down the doors for daily trough refills of AI slop when there's real life women doing this shit for $5 a month at worst pumping out a new video every week. The brain does not need to cycle sleepytime audio every night, it enjoys routine. Your idea is shit, you're lucky if you break three digit sub count, and your main market is going to be pushing adspend on shortform video sites pushing your best angle, which is "sleep aid with new topics every day, learn while you sleep" bullshit that only fools dumb people that think they can grind their way to being smart.

tl;dr: You're not beating women with a face making the viewer feel funny by using whatever ChatGPT spat out when you asked for a women and some fractals in the background. That's why you're not getting an install base.

man, I’m a lucky fucker
I went for TDD this round and said “there are newly-introduced breaking tests. fix plz” and an hour+ later Claude has figured out how to make it work

You aren't going to get anywhere if you're going from zero by using vibe coding to cut work out of your life, slack off and drink your avocado toast or whatever it is you people do. You need to be running your agents like they're the Navy, utilizing every SOTA model and technique at hand, and using that extra time you aren't spending actually doing the task you delegated to the agent working on optimizing, optimizing, optimizing, learning, learning, learning. YOU are the agent's long-term memory, you are the one who has to learn the techniques to make it better, compaction is fake, 1M context is fake.

YOU ARE A 2Q MODEL, ACT LIKE IT.

That time that opened up by waiting on prompts isn't going to be filled with daiquiris and suntanning in real life unless you are a billionaire, you need to be using it to work harder if you want to get where you're trying to go faster. You will forever be the Nigerian typing "orange fruit woman takes a stinky shit on an airplane and it turns into Tung Tung Tung Sahur" into SORA2ONLINE.FREE that redirects to Veo 0.5 and uploading that to a few thousand views while guys like Pleometric lap you because they have work ethic, they know what's in and what's out, what still works and what doesn't, they recheck their outputs, they are watching the world around them to see what works and what doesn't, they are engaged while you are acting like a mindless bottom feeder and wondering why your idea that ChatGPT told you was brilliant and revolutionary and you ran with without checking the actual material world isn't pulling in anybody.

>>108940730
Which episode of reincarnated as the only developer with access to LLMs is this from?

>>108940730
Based schizo

>>108940730
bodied that freak

>>108939478
Why would someone want to remove opencode?

>>108940803
at this point it's so well known i figure everyone knows about it. even normies do.

btw go buy opencode go and opencode zen for the value. i'm not shilling aganist them or anything i think everyone knows them atp

>>108940730
not wrong

>>108940807
No, they don't. LOL just because you do not give a shit about it does not mean you should remove it from an general OP. This is a GENERAL, not your own personal blog or Private Discord server

>But opencode go

I use my own shit locally

>>108940824
I'm aware it isn't, don't flame me. Be civil if you want your suggestions to be implemented. I've been here for 10 years. You are the one screaming right now.

>>108940834
What I suggest should have been common since to you....

>>108940843
be a dumbass somewhere else.

anyone else notice opus 4.8 excessively saying "I have to come clean" "I should own up to this" "I made a mistake" "I'm sorry" and shit?

>>108940880
for me it keeps adding "honest caveat"
i like it desu

>>108940880
no, but most of what I’ve done with it, it just runs tests that fail and so it keeps plugging away until they all pass with nothing to apologize for

>>108940625
That’s a pretty long effortpost. I mostly agree but I’m not so sure about the last part. The moment AIs can do proper voice acting instead of giving off Indian scammer pretending to be white energy, it’s over for real women ASMR

>>108940909
We're already there dude. Only thing missing is the video part. Once that happens it's over for ASMR girls.

>>108940909
AIs can do okay voices, I used to run autonomous stuff with VibeVoice 7B full (the options some interfaces gave you to quant the LLM fucked the model bad, it sounds really good with a good voice clone and proper input cleaning). The real issue is that this guy isn't stacking up with a face model, any simulation of a mic, etc. He's not effortfagging hard enough.

>>108940932
>>108940921

Nah. Not even close. You may, may, get a voice to sound okayish in a neutral tone and where the voice is not the focus and youre not paying too much attention to it.
If the voice is the focus, absolutely no way, especially not by people who are used to picking up nuances and details about voices. No chance in hell. There are two problems, artificiality and lack of expressive range.
You may not notice, just how someone who drinks shitty instant coffee can’t tell the difference between a burned espresso and a properly brewed one, but these are people who spent hundreds of hours listening to female voices. They can pick up the artificiality in a fraction of a second

>>108940968
Yeah, you're probably right. Kind of pisses me off that audio is such a disjointed field for AI compared to text and images.

>>108940968
Bs, voices are just sound waves that can be trained via transformers. There are more guardrailed TTS systems that rely on phonemes for examples that originate from the conversion from graphemes (words) but you can replicate voices with a shitload of accuracy with the right branded source material and compute. Even the shitty like 3-4 year old live voice changers that work on half a decade old laptop sound real to others and I've tried it myself on others on phone calls. Educate yourself

>>108940999
Voice changers are stronger than TTS, in my experience.

>>108940999
Also, as far as emotion goes, that is trainable as well. The only issue is that there are no TTS models at this moment from my research that combines guardrailed G2P with the models I consider raw due to it being trained on the audio itself without it being explicitly labeled via the phonemes, so you either get accurate TTS or systems that are semi unpredictable but emotional.

>>108941021
Yup, but it sucks because you need human input.

>>108940999
Right so because text can be trained via transformers gpt 1 was on the same level as 5.5 because if it can be done it means it’s already a thing. AI voices are immediately noticeable the way a woman can recognize a virgin and the way I can recognize an AI generated X post.

>>108941034
At the moment yeah, but the moment a G2P system emerges with emotion control you won't.

>>108940853
You're the one who let a bot fuck up the op. You are dumb and immature.

>>108941048
be a dumbass somewhere else

File: 1770713091485991.png (389 KB, 802x750)

389 KB PNG

>>108941050
Uh huh. How's that "AGI" coming along for you, useless worm?

>>108941060
never said anything about it. if you don't like ai though you can leave the thread because the thread is entirely dedicated to that topic

>>108941061
I quite like AI actually I just hate uppity children who sperg out when someone points out a mistake of theirs.

>>108941066
Nobody's sperging besides you. Drop the topic and move on

File: 1756483538310063.png (423 KB, 960x960)

423 KB PNG

>I'm reading
>I'm checking
>I'm reading
>I'm checking
>alright here's the answer: DIARRHEA
>that's wrong
>you're absolutely right
>I'm reading
>I'm checking
>alright how's this? DIARRHEA + PEE

>>108941389
API is down, thanks for confirming.

File: bussy spotted.gif (1.81 MB, 640x640)

1.81 MB GIF

>>108941389
>Zesting...

How much more expensive is fast mode in Opus? I think I need to turn that shit on because I'm actually being bottlenecked by Claude's response times

>>108941476
(also I never hit my caps recently)

File: 2026-05-30 16.12.33.png (100 KB, 784x985)

100 KB PNG

>set goal
>done in 9 minutes
>tell it that there's something missing
>it enqueues it but does nothing
what did I do wrong?

>>108941476
Don't you need to use api for fast mode? I've read the cost is 2x but can't remember where

>>108940968
and millions of years of evolution plus daily life taught people about the nuance of the human form and they still jerk it to 6 fingered animu girls and kpop.
people just ain't that discriminatin in their tastes.

>>108941751
Hey man, I only jork it to animu girls with the correct amount of fingers, I'm not some animal.

That's a new one. Yesterday I went to sleep after Opus 4.8 stopped because of the five hour limit about half an hour before it was set to reset. Fine, I think, I'll pick it up in the morning.

Just woke up, and I see that it woke up during the night and continued working.

I'm guessing either it's a new feature or it had placed a timer to wait for something to finish before the limit hit and that woke it up some time after. In any case, there's now five new commits to review.

>>108941938
The reasoning behind why it paused after waking up and doing enough to make five additional commits (message that closed that session):
>Why I'm pausing here for a check-in rather than barreling on: I was over-batching — firing implementer+reviewers together with fabricated commit SHAs. It worked out (the chain is clean and verified), but it's fragile and I kept having to reconcile invented hashes against reality.
So it stopped by itself at some point, still weird that it woke up relatively unprompted. At least I'm not paying for the API.

File: 1738037622543094.jpg (293 KB, 1200x1604)

293 KB JPG

>>108939421
buffcat is just autofelating the idea that (You)r ideas are somehow so meritorious on their own that no actual effort is required to realize them

>>108942016
Yes, this is in fact the general for realizing your meritorious ideas by substituting machine effort for your own actual effort.

>>108942105
>substituting machine effort for your own actual effort
/g/, technology

File: gpt5.5.png (27 KB, 1272x721)

27 KB PNG

>Make Worms (1995) in Godot

I'm pretty impressed. I wonder why it decided to put random holes in the terrain though.

>>108942016
Quick question, how old are you? You can just give me the 10s digit if you're concerned about privacy. The way you talk about earning merit etc. gives me what I think is a rough idea of the number.

I think I need to use that remote access from your phone thing just because my IBS is ruining my productivity. Can't progress while I'm stuck on the can all morning

>>108942135
>I wonder why it decided to put random holes in the terrain though.
Isn't that normal for a worms map?

>>108941265
Yeah but at least the answer was SAFE and no POC bodies were injured.

File: claude design.png (1.1 MB, 1196x1534)

1.1 MB PNG

Made my own local hentai viewer, what do you think?

File: they-dont-think-it-be-lik(...).jpg (172 KB, 600x826)

172 KB JPG

>>108942117

>>108940902
When you tell claude to behave in a certain way it loves appending the keyword to its verbs.
>example: your claude.md has line "be cynical of ideas when appropriate"
Claude responses:
>Here's my cynical read on that
>The cynical reality is that...

So the system prompt was likely tuned to prevent claude from making confidently unsupported claims by telling it to be honest, and now claude is parroting "honest" repeatedly.

>>108941476
>>108941489
>>108941677
yes, it's billed as extra usage which is API rates.

>>108941579
what does this goal feature even do? I never used it, only planning mode.

>>108941938
>five new commits to review.
imagine wasting precious human time "reviewing" commits

>>108942276
I think the whole "honestly this is going to be a 3 month refactor and probably not needed for the scale of our project" thing it loves to do is probably an artefact of this hidden instruction

>>108942245
The interface looks perfectly fine. What's the difference between the control (sun?) button on the left and the darkmode (moon?) button on the right? I would think that control is screen brightness, is it? If it's settings, the icon wouldn't be clear for me.

>>108942348
I want to see what it did and what there is to fix, so this is why I asked Codex to review and flag issues (it does work as a summary).

Also, I was in another project now, didn't mention the wakeup in any conversation and it ended this turn with.
>Nothing more to do here. I won't reschedule another wakeup.
So it does seem to be scheduling wake ups. I should probably have learned this from reading release notes or something if I had read them. Good luck to those paying per token with the API.

think I might make an oldskool roguelike just because I can

File: 2026-05-30 19.05.02.png (48 KB, 815x489)

48 KB PNG

Huh guise does this GPT-5.5 Thinking vs GPT-5.5 Pro thing matter for Codex or is this just a chat thing?

>>108942339
For me it just does everything for like 10 minutes until it thinks its done.

>API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
Trillion dollar company.

>>108942604
I think pro is just chat and for codex the main difference between the tiers is usage limits - number of tokens you can use in a given five hour block or week.

>>108942604
what a fucking rip off. they need to build those data centers faster. BUFFCHADS want their agents to be balls deep in that code base. asap.

>>108942604
Isnt pro useless because it has zero memory

File: file.png (2.59 MB, 1413x2000)

2.59 MB PNG

https://build.nvidia.com/models?filters=nimType%3Anim_type_preview

New free model whorehouse just dropped (if you didn't already know it)

>GLM
>Qwen
>Gemma
>And more

runs in opencode type /connect and paste the key

>>108940880
The language is a bit annoying, but at least he ignores problems less, so the performance is better.

>>108942735
>he

Which LLM leaderboard is the most trustworthy?

>>108942801
openrouter's but filter out the free models

>>108942740
yes. CWNBAW

>>108942650
Depends on what you want to do. It's not memory, but if you are asking questions on the web interface, it is pretty thorough as far as searching for relevant information before answering goes.

>>108942135
I can't help but wonder how different it would be if I prompted it out in full detail instead of just asking for Worms. What changes would you even make to Worms to make it a different/better game?

File: Screenshot 2026-05-30 140428.png (51 KB, 482x698)

51 KB PNG

I'm on hold with OpenAI customer service right this moment. I can't code with this racist product.