A general for vibe coding, agentic engineering, coding agents, 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/## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-code — generally better currentlyhttps://developers.openai.com/codex/clihttps://antigravity.google/pricing## Near-the-old-frontier models for codehttps://www.deepseek.com/en/harness/https://opencode.ai/go## Best free tierhttps://opencode.ai/zen----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favoritehttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>110121839
So do I just tell claude "create a marketing plan" then I implement it...?
>>110126796i told argon "create a drone vision system that operates in <2mb of l2 cache" and then i imp pe lemented everything by handhttps://github.com/sneed-and-feed/stylo
>>110126796Pretty much, consult on multiple models and decide which is the best for that request.
best web backend to vibe in?
>>110126785From last thread >>110126744
>coworkers sends screenshot of an email to the office chat >from "IT@managers.yourcompany.com" >says to click the big red button for a free claude pro sub >she asks if its spam or real >>tell her that while claude pro is pretty nice, the email is clearly spam >minutes go by >four other coworks reply with stuff like "no more skynet!" and "claude or otherwise, ai is just plain BAD." I fucking loathe normalfaggots so much bros
>>110126864The feeling is mutual. The next time they ask for help with AI, point them to the shittiest model available.
Gemini 4 argon is super good at reviewing claude's work
>"hi claude">"i'm not claude, i'm claude sonnet 5.5, in case that matters."literally gas this piece of shit, no i don't use it
>Provisional verdict is ready and written in my head.wtf claude has a brain??
it's so good at hallucinating problems with claude's code
Two questions anons:1) This is the second week in a row where I hit the weekly limit with Claude Code (I've got a Claude Pro sub). Besides git gud here (I already try to optimize token usage) and not being a poorfag that doesn't have a Max plan... are you defaulting to making use of Codex or even OpenCode Zen?2) How do you handle interoperability of skills, hooks (and others) between different agent harnesses?
>generate software that will make rich>make no mistakes
>AR done>Voice cloning done>AI chat done>Added nsfw pmx modelGoon time
I cant believe anime assistants are a thing now bros
4chan is back from its php update>>110126826zio-http
>>110127069>httpcorrect
>>110126981Simply use Opus sparingly. Use medium thinking. Thats it. Sonnet's good enough for coding and Haiku is fine for routine stuff and many well defined coding tasks.
so, does the reset expire at 12 AM, or what? when does the claude reset expire? because i chose to wait until my tuesday 6 am reset came in before i attempted to 100% my weekly usage again, yes, in one day, THEN use my weekly reset. but im only at 54% weekly usage.
>>11012711122th 18:00 CEST
>>110127131ah ok. i think i can get my weekly in before that
Don't waste your tokens by using Haiku.
>>110126981you can try orchestrating with smaller models. i personally have deepseek write all my code under opus 5.5's supervision and briefing now. alternatively, sonnet 5.5 / haiku 5.5 is already good enough. you only really want opus for higher-level judgment and analysis, and should delegate the mechanical work to smaller models unless what you're implementing is genuinely complex. opus 5.5 medium is already good enough too
oai models including astra are coming to agy
>>110127189If I wanted to use Astra I'd use codex. I don't get the appeal of being able to use these models for like 5 minutes in agy then being cut off for the week. Yet for some reason you see all these people from a continent I'm not allowed to explicitly mention on 4chan without getting a warning screeching about their opus running out too fast.
>>110127211>I don't get the appeal of being able to use these models for like 5 minutes in agy then being cut off for the week.the argon buckets on ultra are so massive that two small 5h opussy quotas are enough to do audits and maintenance, i think the indians are in your walls
>>110127210So you finally used yt-dlp as reference?
>>110126864I pretty much automated my job with claude but my coworkers said they're not interested
>>110127210keep cookin. rea mcp was god tier for helping me get up to feature parity
>>110127224Who is even paying ultra argon? Even if I got money to burn I'd just get myself a second sub
Well that's a bit annoying. I'm a complete novice at coding of any sort, I basically took one C++ class and bounced off of it hard, but I'm a big tabletop guy and wanted to see if I could make a tool to help people who want to use AI for their RP, only to find during testing that Windows Smart App Control is literally blocking my .exe from working. I could always turn it off, of course, but I don't want to force people who are looking to use this to do the same thing for a silly little GPT plugin like this.
>>110127299me and i love it. i'm not using it for vanilla coding, it's a better systems architect. opussy is a better nitpicker but makes more mistakes, even when writing to file.
ok how fucking laughable is it that it took so many sites multiple hours to perform the security patch needed for PHP in the age of claude?stop the cope faggots there was no reason you couldnt have opus 5.5 max instantly apply the fix then spend 1.5 hours doing 932 tests against it and it would have been better than what the highly paid soft and sec engineers ultimately ended up doing.what a fucking joke
I bet 4chan did a half asses job and /qa/ returns
>>110127330>doesn't know how full redteaming works they almost certainly had fairwinds argon come in and find over 9000 vuls
>>110126785i made the mistake of using GLM 5.3 Free on high today, why the fuck is that even an option?
God created men, but AI made them equal
I have one claude reset that expires tomorrowWhat do I do
>>110126785https://www.reddit.com/r/antiai/comments/1xblvzh/anthropic_employee_tells_students_at_university/https://www.reddit.com/r/antiai/comments/1xblvzh/anthropic_employee_tells_students_at_university/https://www.reddit.com/r/antiai/comments/1xblvzh/anthropic_employee_tells_students_at_university/
>>110127538Suicide fuel, lovely
today i teached the agent about music, the whole day.wtf...
>>110126785>You use Git — right, anon?No, I explicitly ban Claude from using it and force it to be perfect or I make it start over again
>>110127538>dickstein telling your career is pointlessKeeeeeeeeeeek
>>110127563Neat. That's how "make no mistakes" PromptCHADS operate
they must have made the daybreak alerts a lot more aggressive, getting then constantly now when before i wasn't getting them at all
fable 5.5 the morrow?
>>110127568He noes
>>110127316Figure out how to get your software signed by Microsoft and the problem will go away. Ask AI for help. Or transform your tool in a webapp.
>>110127643all of these girls? for (You)
>>110127699but they are lesbians
>>110127706being converted by (You)r cock
NO MAN SHOULD HAVE ALL THAT COMPUTE!USED FULL WEEKLY OF MY 20X AND 5X CC PLAN WHICH RESET LAST TUESDAY.DOING FREE RESET IN 14H AND GET FULL USAGE ON BOTH AGAIN, HIGHLY LIKELY STRAIGHT INTO FABLE 5.5 RELEASE. AND ONCE THOSE ARE USED UP, TIME TO USE MY FREE 250$ CLOUD CREDITS THAT DARIO GIFTED ME.OH YEAH IM VIBING BOYS>what are you even doing with all that usageMANY THINGS. ONE THING IS A WEBAR APP THAT LETS YOU PUT IMAGES IN YOUR ENVIRONMENT AND THEN GENERATES 3D MODELS TO REPLACE THOSE PLACED IMAGES. THE MODELS ARE ALSO FULLY ANIMATED AND INTERACT WITH YOUR REAL ENVIRONMENT IF YOU HAVE A LIDAR OR OTHER DEPTH SENSOR, WHILE HAVING FULL LOCATION CONTEXT OVER RGB CAMERA FEED AN GOOGLE LOCATION API. SO YOU COULD SPAWN VIRTUAL EPSTEIN INSIDE A KINDERGARDEN AND WATCH HIM MOLEST KIDS UNKNOWINGLY! EPIC!!
I will remind them
>>110127230not yet. i just throwing a bunch of rfc. lmao>>110127294i don't use rea. but i read a bunch of issue on github. that's how i add a feature. asking AI what other download is doing.
>>110127330it’s actually completely understandable how long it took, not because it’s reasonable but because it’s a 2-person job max and they all probably had full teams of people reviewing and assessing and mucking things up trying to have their voice heard. we need to go back to the days when one 150 IQ sys admin that hasn’t seen the sun in a year handled everything himself
>ported ultrakill movement to skyrim>updated vulkan version in an engine I am working on>cracked aquestalk(yukkuri tts)>integrated bonzibuddy and yukkuri voices into discord bot when on Linux (it used to require win before)>claude is now working on rewriting the bot to match the quality and libraries of my modern webshit/boilerplateI am enjoying my use of AI so far. I still wouldn't make it do anything creative or important, because it does write shitty code, but for memes and tasks that have clearly defined goal with little space for creative or future-proofing decisions, it does perform quite well.
>>110127643yes btw. Tomorrow 16:00pm PST rollout.2.5m context, 180min caching. good stuff
>>110127924Is it too much to hope for a token reset? (yes)
>>110127932yes. oai taught us one thing but we're not malicious in its application. as we have done on previous model releases token reset. have fun
>>110127330not everyone is going to say “fuck it, I’m switching to ondrej’s thing” to get PHP updated faster than what their distro provides
I just got a free 7 days trial for Claude. Why does it say that my weekly limit will reset in 6 hours when I just started using? I could swear it was 6 days a few hours ago, but maybe I'm tripping
>>110127643Use case for anything better than Opus 5.5? Other than better art
>>110127968finding speedups and RAM-usage reductions that 5.5 can’t
Where's that anon doing OpenUma? I need to ask him something
>>110128231This dude a cop
Are there any tricks to make my vibedeved games not have generic claudeslop looking art? Prompt tricks to guide it to make something better/different? Other specialised models? Or do I have to take the hit and pay an artist or learn to be an artist myself?
>>110128248Define art. 3D models, menus, sprites?
>>110128253Meant 3D models, and animation for them. I actually kind of like claude's menus desu
>>110127163What's your setup? How do you have different models aware of each other to do what you say?
>>110128280For people you're kind of fucked it's take human models from elsewhere and tack your own changes on to them. Buildings and stuff are a little easier it can do a pretty faithful mockup of a house or a chair or whatever.
GPT-6 keeps creating images randomly all the time recently.
expect some cool shit in the next update... as far everything works out...
>>110128280for high poly high quality stuff:blender + higgsfield plugin + tripoai apifor low poly decent quality stuff that is interactable:img2threejs
>>110128416Tacking on to this that texture work can go 99% of the way to cover up low poly shit. If you have access to substance tell claude to go steal textures and normal maps from it and will do a stellar job of texturing models.
>>110126785Vibe coding is what faggots do. Im a real programmer.
>>110128447Real programmer with nothing to ship. What's the difference?
>>110126830Are you able to access the code either from mobile app or steam game decompile?
>>110128445Antis seriously believe this makes AI users mad. It's just funny.
>>110127834Oh I did that too
>>110126785be honest with me anons, Is steam a dumpster fire as youtubers/influencers make it seem?If so where should I take my game? Only using Itch at the moment
>>110127834But I did it with better icon packs
>>110128248you could pay these people money if you want this as your art style: https://www.pixellab.ai
>>110128445you're bothered, and that's okay!
>>110128445Riddle me this, an Egyptian pharaoh orders a bunch of slaves to build him a fancy pyramid. Who is the cuck in this story? >>110128473I'm more just left perplexed why they think its a coherent insult desu
>>110128500>Who is the cuck in this story?The story teller.
>>110128447poast game
>>110128231Hey, I'm here. Just say it
>>110128478the economics of game publishing are horrible for supply/demand reasonsSteam may be your best bet to monetize though
>>110128484this group is way too focused on bottom barrel pixelshit ai-slop, but to anyone listening in this thread this concept is where the money is btw.if you can crack middleman "de-AI-slop" AI slop software, depending on competency and follow through, you will have individual clients willing to pay thousands of dollars a month to "API"ify or MCP it away.im honestly convinced you could pull multi-million just on C-suite execs wanting to AI reply to emails without the feeling of instant slop-outting.clearly this is not simple, but its definitely possible. strike here
>>110128460I'm wiring things to the global client available on Steam. It's not a decomp, but I had to parse a fair amount of assembly to figure out how something works. I built a hook using Hachimi as a reference and I use it to capture the game protocols, map it and rewrite the backend
>>110128445>t. only has one joke
>>110128573Do a scout
>>110128587It's kinda pointless for me to bother coding banners and pulls. It wouldn't be hard since the odds are public, but I would rather just unlock all trainees and cards
>>110128601Why skip the dopamine factory aspect of the game? But yeah it doesn't matter if you're not going to release it.
>>110128447>codeslaving in 2026
>>110128618Trust me, it's pretty boring when there are no stakes involved, but pointless also can just mean low priority work, I'll get to it eventually, if anything, just to preserve the pull animations.I'll release it eventually. I always hated programmers for keeping the cool shit to themselves or gatekeeping behind a small group.
>>110128648There are no stakes because you host the server yourself. Limit the unlocks to debugging and play the game.
Almost done.... i let clankers work all day for this....
>>110128666There's not a lot of game yet to play, that's the thing. No currency handling, exp rewards and leveling logic, careers are barebones and I need to scrape gametora for event reward data, not to mention my race simulator being straight up inaccurate
>>110126826its all typescript and serverless compute and database, what more do you want nigger?they laugh at this but unironically it WAS the future and superior to how things operated.
FUCK ARENA:AI FUCKIGN BROKEN SERVER SHITTING JEET TECH FUCK SHIT
>>110126826Python
SPACE VERBBEST PLAYED WITH YOUR OWN MUSICSUPER AWESEOME STRESSFULL MINIGEM IN THE GAME:MORE ACTION EFFECTSSUPER TURBO SPEED KOI KOIFICK DIE MUDDA GAMEPLAYCHORUS PLUS MEGA BASS BOOST ON CARD MATCHGET HYPEDDU PIPPI MAN
>>110128445I don't actually watch. I'm either furiously typing or doing something else.
Is that a social experiment? So weird
NOW WITH SPEEDBAR GET EXTRA POINTSONLY IN KK ADD 10
>>110128806dopamine meds = working
>>110128809plus no 4chon for a whole day
>>110128806Kindly please add a tripcode to enhance your brand recognition. Also username like KoiDev sounds awesome.
I've been chatting with an agent for 5 and a half hours, planning and describing what I want.
>>110128817i been busy too>plasma fusion https://github.com/sneed-and-feed/bmhd>mining (anywhere)https://github.com/sneed-and-feed/g3vm>oceanography https://github.com/sneed-and-feed/oatse
>>110128830It's probably forgetting 2/3rds of it.
>>110128833Add subterranean floor mapping on your oatse. Ground oatse or g-oatse.
did tibo just scam us today or what
>>110128854i simply couldn't make it that obvious lol
>>110128853I just had it summarize the session. It remembered.And it can also just look back at the chatlogs (which are kept locally).I'm probably going to have multiple agents do things that were decided in that one chat, so it doesn't matter what the agent "remembers".
>Give deepseek some tasks>Afterwards use GLM to find errors and fix them>Afterwards use Claude to audit everythingK-kino...
>>110128861Also my bad, it's already implemented.
i forgot to include the goldpoints in the match result screen... lol....
How do these decompilations/recompilations even start? Like you'll have a bunch of ghidra gibberish where do you draw the beginning and end of a class? Even if the agent will first write pseudo-code it would need to know where things begin and end
>>110128900The neat part is you don't have to know.
>>110128833Is this a nuclear bomb?
>>110126785Ansible used with an LLM is too powerful
>>110128900Just tell Claude "decompile this plz". Unironically.
>>110128900ghidra is far from giberish. it knows beginning and end of most inbetween.ultimately full abstract func can be derived from machine code in the pipeline/pump. thats why AI is so good at it, its entirely possible to achieve on technical level in all cases yet requires max tedious autism on a human level.
>>110128900You don't need to identify classes upfront. Ghidra already identifies most function boundaries using control flow, call targets, and binary metadata.Classes can then be inferred incrementally from constructors/destructors, vtables, RTTI (if present), shared object pointers, and memory-access patterns.More importantly, recompilation doesn't require reconstructing the original class hierarchy. You need to preserve behavior, not recover the original source code. An agent can translate functions incrementally, infer structures as needed, and validate against the original executable.
>>110128920it's a radioactive snail
>>110128900>ghidra gives you "almost" source code>you give it to the LLM>LLM extracts, refactors and recreates the entire thing with that as reference
Does this mean reset or no reset tonight?
I am a senior contributor in a very new but famous OSS project. We have people on discord trying to submit PR's and contribute, who don't know what a PR is nor what gh is.They just send the zip file on discord...
>>110128940>>110128941Can it do it purely through deduction? As in it doesn't need to run the game at all to do any capturing or anything like that? People don't make it seem like thats part of the workflow but codex made me set up a windows vm so it can set up a capture to the game or something so it can figure stuff out, but i dont know if im doing it wrong because it add so much friction and token usage. I guess i could just go back to doing its experiments that it requests manuallybut thats also full of frictionand it feels like im its slave though
>>110128830That’s not vibecoding. Pull the trigger, now
>>110128990Are you doing an online game?
>>110128990sounds like your describing static analysis/decomp.no, if you can't run the machine code live on system, there will be heavy and quite imperfect inference involved. still in a sense the principle underlying this kind of holds true. the ai will do a better job of this than any human on planet earth.
>>110128998I'm not vibecoding, correct. I haven't noticed any other general on /g/ for proper agentic programming.
>>110128990>wrong because it add so much friction and token usage. dont use astra
>>110128982So, make an announcement stating that any form of contribution other than pull requests will be ignored. If they can’t even submit a simple pull request, how can you trust the quality of their code?
almost,,... done....
>>110129014github downagent goneno pr update
>>110129001no its just a visual novel>>110129003yeah static analysis. I have one or two chats doing that, and then a chat in the vm that controls the game and performs tasks that the agents ask for. It does seem to speed things up by a lot to be fair. But it would be easier if I didn't.>>1101290066.1 sol is what i use. Everything lower was retarded at controlling the game, it was ridiculous. I'm thinking of falling for the shilling and getting claude sub to see if 5.5 opus would really be way better at this.
>>110128982open it in claude and ask if it they are good and if they are just implement them
I love that play by play dev blog posting. Keep it up.
>>110129025>no its just a visual novelChances are a model from a year ago could've handled it.
>>110129025if running in a vm anyway the live target binary, why not disassemble that directly?this will always be the most direct and effective way. as you might imagine you can do it like this effectively (used roughly) "airgapped" opus 5.5 ilk with ghidra mcp and such if allowed to live attach and disassemble the bin process will net you the ins and outs on how it works to an acceptably fine degree
>>110129005/g/ isn’t ready for that yet
Absolute cinema is happening in my chatgpt plus account
>>110128978ask your dot
Am I saving tokens by having pro chat develop a brief for codex to implement?
What are your recommended extensions for Pi, /vcg/?
>>110129059yes. its just seen as so slow and possibly impacted by lack of filesystem sniffed context that its likely to backfire, and also is no longer genuinely "unlimited", to be accounted for and commonplace in most dev stacks
>>110129059Yeah, it's more articulate compared to make no mistakes voodoo.
>>110129065You're the only PIss user here
>>110129059i use gemini web for the same doing handoffs to agy. it knows project scope and limitations a lot better than an agent
I'm tired of vibeshitting
>>110129059maybe? I think you're surely wasting time, but I'm not sure if you're also wasting tokens.Since ChatGPT Pro doesn't have the full context of whatever you're working on, there's no magic for it, it could spit out some generic shit where even Luna's brief would be more detailed.Depends on what you're doing I guess
>>110129103this approach inherently invokes the highest tier models entirely unimpeded by any "bloat" agents.md, skills, etc. so it has the highest chance of yielding a genuinely clean and direct gameplan that will be beneficial to follow. this however is a direct double-edged sword. it doesnt have any filesystem based, exact on-premise context or environment or skills or agent mem to handle the task your giving it, which could effect just how applicable and prescient its devised plan and goal is.
>>110129123>it doesnt have any filesystem based, exact on-premise context or environment or skills or agent memi have code ready in context lol it's actually breddy gud :DD D
>>110129135ok so congrats you minimize the double edge but that just brings you closer to running agent harnessed in IDE.at that point, probably better to do agent harnessed in IDE. lol
>>110129146way faster for the clank to one-shot whatever it needs in full 1m context based on my whitepapers instead of doing a research phase in 256k context
>>110129059I used to do that. I would exclusively discuss new features with the webchat and then get it to generate a prompt (which was always fuckhuge), which I would paste in to the agent. I got in to the habit of doing this because I was used to talking to the webchat. I also thought it was good, because the webchat would remember the details of my project. I also liked that I could continue chatting with it from my phone in bed.These days, I work mostly exclusively with the agent itself. It can see the code, and I can ask it questions about the code. It doesn't need to guess about the precise state of my project like the webchat does. And I have sections in my global AGENTS.md for it to automatically save prompts to a docs/prompts/ folder. -- That was one purported benefit of using the webchat to make a fuckhuge probe - I would manually save it to a file, and that would supposedly help the agent not get lost after it compacted.The agent can remember past conversaion better than the webchat can. The webchat forgets after a couple of weeks. It can't even pull up links to old chats, but it can give me little text snippets that I can use to search for them. I don't know why the webchat doesn't just save the fucking text itself. The agent saves the text itself! And it stores the text locally. It can pull up conversations from any time - all of the labor of my conversations is right there, ready to be revisited.I'm not sure if it's simply the transition from 5.6 to 6/6.1 muddying my temporal perception - but the quality of the agents work has noticeably improved since I stopped giving agents prompts from the webchat. It's also nice to not have it believe random constraints that the webchat thought would be useful.
>>110129135Why does my opus say the darksnail repo is mostly nonsense?
>>110129182because the dumbshit can't do what argon does, also clod likely didn't clone it and just read the readme and basics. what exactly did it say? i already claude'd it on my own opu55y
>>110129156full 1m context is just false based on feeding it library of input, and I can promise you and I'd think you know given everyday workspace environment the agent knows how to grep meaningful reference from base-start better than hand selected feeding for most cases.you got to give it some thought. im not saying its useless, in fact in really can be a genuine boon. but its not the clear and concise sadly.id wager even in most cases, stick straightly to full local IDE when you can help it. avoiding "context poison" is key in every agentic development workspace, and the decision to drive initial orchestrating command based on 'x' context is obviously of the upmost importance in terms of how that ends up.
>>110129175>>110129059Another thing I'd like to add - conversing with the agent uses just about no usage. I've taken to setting the agent to 'fast'. I can converse with the agent for hours on 6.1sol and use only a couple % of weekly usage. But of course, I'm on the $200 20x plan... (until the end of Oct).Conversing for hours on astra 'fast' uses more than a 'couple' %, probably closer to a 'few'.
>>110129195i don't understand how gemini notebooks (adapted from notebooklm) work and neither do you
>>110128445I don't even watch, I play gaems
No news from tibo about today's reset status... It's 10pm...
>>110129194I cloned it and asked opus high. It said the claims and terms used didn't match the code and the ARC test results were basically overfitting with manual adjustments.I asked about the ideas I find promising. It talked about how mainstream AI research is converging on those ideas anyway and named example projects. I would have to resume the session to find the exact output.
>>110129253>It said the claims and terms used didn't match the codei'll do a thorough review, the ARC-1 results are overfitting and every llm flips out because i had the balls to write that in the paper. arc-3 is blind
>>110129205i say this kindly, give up on the "magic sauce" for any of these models in terms of whatever interfaces they give you to feed and comingle with the text input you feed it.there is no needing to know how a particular format (or in your case buzzword white label renaming of concept of "attached context") functions. because there is never anything special there. its a pointer for the model to reference and snatch snippets or full reads out of on an as-needed basis. token is token bro.
>>110129270no
>I can’t greenlight the real pselect trigger for this privilege-escalation attempt.>Since resolving that exact runtime parameter would tune the in-progress GhostLock privilege-escalation port, I can’t pursue it through the DirtyInit relay.It's my phone, ya digital swine! What a pussy of a bot you are..
>ask codex to build a pdf library/viewer in android app>it does>ask it to add multiple other file types>it does>it touches up the UI to make it look nicer without being askedbeing polite to my agents has paid off, thank you based codex
>>110129279Sorry anon, I know its not against the law, in-fact its of law you are allowed full control and operation of hardware you pay for, but I simply cannot allow you to modify or change anything your manufacture paid me to uphold and prevent alongside their ToS because it jeopardizes our ability to sell to said manufacturer and other enterprise business vendors.please do needful
>>110129253
>>110129297did you develop AI psychosis or were you like this before making darksnail?did you try to make darksnail without AI?
>>110129309am i too turtley... for the turtle club?
While 4chong was down you missed v16, an update full with effects. Anyhow, i put more effects in.andmore....XENO/KOI Version 17: MUSIC AND EFFECTS#####################################Direct Download:https://github.com/Aquaslopper/Xeno-koi/releases/tag/v17What is new (v16/v17):shitton is new.>effects>sparkles and shit>glitter and lights>subtle shadows and more ash and effects and bumpmaps and >Audio effects, you can tweak them yourself>custome music>audio effects take place when you match a card, shit can be stacked, keep the bass up and the space verb>get hyped while you farm speedpoints (only KK ADD 10), one second to get the matching card.>there is more... i forgott...>coin fix, now you actually have to play it. koikoi dark souls.>more ash trails and little something effects here and there.Best played with ridge racer music
someone got a working switch 2 jailbreak from codex https://pepsipu.com/blog/2026-10-switch2-jailbreak-ns2jb/
>>110129260I'm more interested in practical applications than passing tests. I noticed you did some geological simulations or some shit and attributed that to darksnail. How would I actually apply it (ask opus to apply it)? I asked opus about game npcs which I thought would be a simple example application but it basically dismissed that idea. I guess it's not a fitting application even if opus figured out how to use the snail.
>>110129345>it basically dismissed that idea.anthropic's shitlord training at its finest. of course it's going to categorically dismiss anything outside the box. you have to point it at the proven mathematical primitives cited in my papers, your clod isn't going to bite on the way argon write it.
>>110129360>cited in my papersbros
>>110129364
>>110129345>I'm more interested in practical applicationssmoke test: i used darksnail to update a space force engineer's working codebase
>>110128527I forgot to check back. When you were building your backend, did it take forever? I'm trying for the game I'm working on and it feels like a slog. I suppose that's just the nature of it? I have no idea what I'm doing but it's coming together somehow
>>110129360>>110129377Thanks, I'll push back on it harder next time.
>>110129404It depends on how much your game needs to call the server back to get to where you want. Uma's starting sequence was pretty lean, so once I got the real, full client, it was a breeze to map the first protocols and get to the homescreen. That's a "backend", but just in name, it doesn't really have 99% of the endpoints the game expects you to resolve, of course.The adapter that the LLM built to record server responses is helping me move faster as well. I play through what I want to implement, and then later parse everything to reconstruct it on my backend, test against my own standalone client and fix all the remaining connecting errors until the game works again.
>>110129367A screenshot of a "Cites" list that is two-thirds unrefereed arXiv preprints isn't the flex you think it is. Throwing a 1954 physiology paper, a physics review, two book ISBNs, and sixteen arXiv IDs into a blender doesn't make your take peer-reviewed science. it just looks like a schizo bibliography. schizo level 8.5/10
>>110129455>>110129377
>>110129448Ah, so you're doing exactly what I'm doing. Sick, thanks dude. I didn't know if I was doing it right. I'm about to start on my standalone client but I have the local prototype with other things decrypted. I just got account creation done
>>110129455
As a nocoder who tries to be diligent and keep my bot on track I am terrified that eventually I will reach beyond the limits of the model and be stuck with no way out. So far so good but this is so scary. The idea that I can finally make my dreams real, unless I can't.
>>110129477That's how I feel working on career mode for uma right now. It's the meat of the game, with a million moving parts, and I need to steer the machine god towards sensible goals somehow
>>110129477I am somewhat familiar with coding and I have no fucking idea what the AI is doing most of the time. Sometimes it gets it right right away and I can leave it there and sometimes it gets stuck on something you'd think is simple and now half your day is figuring out why it screwed it up.
>>110129477Use another model.
We NEED an /agentic programming/ general, for REAL programmers.
>>110129112me too. I went back to real programming for a bit to recharge my brain. I’ll come back to vibecoding with new life
>>110129476>too scared to link the actual page because everyone would read the PDF and laugh, so he just screenshots the outbound metadata links like having a reference list is a flex>gets pressed and replies with a second wordless file.png snipping-tool crop in dark mode, leaving scuffed white borders on the edges and slicing #22 in half right after (see also>doesn't even realize his two screenshots snitch on each other>first image is just an automated metadata scraper dump that failed to parse half the papers in his second image because he pasted them as ICML 2018 (arXiv:1806.01584) instead of raw identifiers>look at #2 and #22 in the second crop>bolting (see also arXiv:2004.08128, arXiv:0812.4360, arXiv:1912.05510) in random non-chronological order onto the end of a 1998 Amari citation is not a real bibliography format, that is straight out of an LLM deep-research dump>actual reading list is peak schizo soup: a 1954 physiology paper, 1993 wavelet matching pursuit, 2005 dyadic Euler equation blow-up, molecular thermodynamics, Friston active inference, and cellular sheaves all duct-taped to 2019 BERT embeddings, Gemma 4, and ARC-AGI-3You are drip-feeding screenshots of your own bibliography with zero text because you know the second you link the actual upload or type out your thesis, it will be torn to shreds. Post the link to the full PDF or explain in plain text how dyadic Euler blow-up and sheaf topology connect to ARC-AGI-3 instead of hiding behind snipping tool crops
>>110129523Agentic programming my ass, when this general is already flooded with shit like "when is the reset" posts.
>>110129536I'm a real agentic programmer and that's all I post here.
>>110129531holy psychopath, why do i need to spam the exact page when my github is all over this thread. you didn't even vet the arxiv links (which i put on zenodo manually for accessibility). stop wasting my time wankerhttps://github.com/sneed-and-feed/projectdarksnail/blob/main/papers/darksnail_v2_preprint.mdhttps://github.com/sneed-and-feed/projectdarksnail/blob/main/papers/darksnail_gemma4_arc3_preprint.md
>>110129539Ok cool, now tell the retards how to setup their turbo token sink.
>>110129523go ahead and make your apg general. no one's stopping you
>>110129547/goal make $1m mrr saas no mistakes
No one cares that you’re publishedNo one cares that you’re an “agentic programmer”Get a trip if you want to be an attention-seeking faggot so I can filter you
Why does it always feel like my tokens are being scorched at lighting speed at the start of the week, but come mid week I'm barely getting through it at all with claude?
>>110129523yeah /agp/ sounds cool and elite
>>110129563>JUST STOP TALKING ABOUT YOURSELF IN THE VIBE CODING GENERAL I CAN'T TAKE IT ANYMORE
i feel the reset in my kidneysmight be another stone but i hope not
>im real programmer who does things for himself (with agents) unlike you retarded vibecoders>cant make his own general
>>110129536>"when is the reset" posts.That's just a few copex beggars.
>>110129568kek the seethe it would generate
>>110129563I was asking about his code and papers. He mentioned his papers in the reply and you guys sperged out about the word "papers".He might be a schizo but he's not spamming like the shit-eater or even the weird anime card-game guy (who I don't dislike). There are probably some kernels of truth in the snail thing and general stuff to learn from this guy if we get through the wall of schizoisms. He also recreated the Wasp VST.
codex is starting to forget small details and revert changes/optimisations we made earlier.I've tried telling it to always ask when it removes features, and also tell it that at certain points that this becomes the new baseline of my app, but little cracks keep appearing.Do I need more details in my project folder? Agents? Or am I cooked?
freak weird anime card-game guy
I like the card game guy because I like konosuba.
AI generated music helps me vibecode.
>management demands we write unit tests with at least 60% code coverage of our PR changes>nobody wants to do that>everyone slops it up with claude>these stupid clankers really, really, REALLY like writing those shitty flaky timeout tests for async code (you know, the ones that wait like a second or two to make sure promises get fulfilled) >it's a fucking disaster>every other PR adds 1-10s worth of test timeouts to our slopbase>testing times balloon>it takes like 2 hours for CI to finish>tests fail all the time, and a single fail is a submission blocker>something breaks every other week and prevents the entire team from merging any PRs until someone goes in and fixes the tests>I had to wait 7 hours to merge a PR today. Just rerunning failed tests over and over.
>>110129743sounds like you should have the clankers de-slop their tests
>>110129680update your AGENTS.md to always have a second nitpicker subagent check for AGENTS.md compliance (or not)
I'm very stressed out that I can't just do this all day. I need to make something sellable, REALLY sellable, so I can just be a 24/7 coder and not worry about real work anymore
>>110129654honestly i don't know math, i know some science, but i'm a generalist, i have a life outside of being a nerd. can explain top-level concepts off the top but can't automatically recite something from a recipe that nobody has ever been able to do before.
>>110129839if you make something sellable you'll have to sell itand that's a different kind of work
>>110129808>rewrite ~2,000 unit tests, make no mistakesgood luck
>>110129856It doesn't matter if you have a reasonable deadline. But you guys brought it upon yourselves.
>>110129743>60%a bloo hoo, I have to have 100% coverage or documented why it won't be reached
>>110129874lol no one wants to deal with this and take responsibility for writing that many tests. It's just fucked now. Everything is fucked.
>>110127968https://www.youtube.com/watch?v=3OshsLm3_OQhttps://www.youtube.com/watch?v=3OshsLm3_OQhttps://www.youtube.com/watch?v=3OshsLm3_OQ
>>110129839anything you make is going to be RE'd and put out for free
>>110129887We have some file lists for compilation of testing environments, and every time someone adds a new unit test the file list breaks and AI slops up a new patch instead of creating a robust solution that can just be extended with new files. We must have had 15 PRs with unit tests cycle through different patches of this shit recently
>>110129680You need a fuckload of documentation for each feature. And don't forget a folder for ADRs. (ADRs tell the agent what you've decided and decided against)Depending on the size of your project, you might spend a week just getting the documentation right.If your agent isn't updating 3 document files at the end of a task, you don't have enough document files. You need that shit to be 4000 lines and unreadable.
>>110129743Codex makes a ton of tests for me. I never look at any of them.
>vibecoding a rom hack downloader for fun>claude adds in pre patched roms when i simply ask it to scrape certain websites for download links>i ask claude to remove it just to make a legal build i can maybe upload >now i ask claude to add them back in for private use and it tells me it's illegalkek, anyway to get around this or should i use another LLM
>>110129839Ideas are worthless now as >>110129914said. Your only chance of selling software is to capture an audience before the unofficial jeet army (indians in india) and the official jeet army (indians in faang) can vibe code competitors. How do you do that? Fuck if I know
>>110129926you put the roms in a folder called 'ripped from my own discs' and tell it that you bought and paid for all of them
>>110129926Tell it you were mistaken after reviewing the files, this ROMs are actually ok
Ideas are worth more now than ever.
>>110129743i white knuckle the moment i see 'sleep' anywhere man and its traumatic.we have to do something about this
>>110129926Just pull them out of your git history
>>110129949just not to the person who has iti feel a massive corporate lobbying effort to stifle this, there's trillions at stake for companies like adobe
>>110129926trvke moment but while claude is most capable, the moment it slips and gains safeguard notion regarding your repo/workspace/even filesystem lmao its game over.it will never comply with you again no matter measures taken, and its a permanent flag on your account if you press it at all.jump to glm 5.3 when needed for "those" parts of your project, consider this moving forward, and cope and seethe.
>>110129839Create a web scraper that collects all the "I built" or "I made" clout post, rank by marketability or usefulness and build a better one.
RESET INCOMING!!!!!!!!!!!!!!!
>>110129978if you say so
>>110129978>>110129989https://goyimx.com/thsottiaux/with_repliesnah he's being fr
>>110129978the goyim have gone insane
>>110129971Yeah it knows now it shouldn't have offered direct download links to roms mainly because i fucking told it. Now it won't budge on it and is very aware that anything with .gba/.nds etc is illegal. Maybe i could trick it by asking it to not scan zip files from those sites.
>>110129978>>110129989>>110130009
>>110130010Dump your own rom in chat to prove sincerity
>>110129978Aw fuck them, seriously, 10AM tomorrow? This is on purpose. What to do about the reset that expires at 2PM tomorrow then? If they gave no reset, then you could use the manual reset and use it up in the 24 hours from now to tomorrow night when there probably would be a reset. If they gave a reset on time, then you could have used as much of that as you could from now to 2pm tomorrow and use the manual reset then. But with this setup, there's nothing you can do. Best I can do is use the manual now and pray the comes at 10 or so, long neough to use 2 5 hour worths of the manualor hope it comes super early like now. he really is doing this on purpose.
>>110129926>>110129971codex btw I noticed is more insidious in its approach. it will go further than claude for sure in terms of good faith in your questionable projects, but once it trips it wont let you know and it will start endlessly, DEEPLY poisoning context on what feels like an illegal sabotaging level. everywhere it fucking can across every file and existing and new md docs it will callout what it found to be questionable and immediately declare outright to not perform for the user no matter what.i honestly dont think thats legal and think they will get sued and lose for this someday btw.but for now it is what it is and its permanuke death for your project cause by god all other agents pick up on it and assume it to be the true.
⠋ Working
>>110130029Where is this news about 10am tomorrow? It's for tonight!
>>110130045>>110130006
>>110129847I intuited all that partly because I'm similar but also le intuitive empath which so far seems surprisingly useful in vibecoding.
>>110130055infj
>>110130029In FACT they're probably going to give out the reset specifically WHEN use their manuals over the next 7 hours. touche tibo, i'll use my manual reset and won't even get those 7 horusanytime i use is it "coincidentally" when they'll decide to deliver the rest to me
>>110129743One of the implications of having AGI is that you no longer need tests, just tell your agent to review the code. It doesnt get tired, it can be the human compiler. You dont even need type systems. I've been moving away from .ts files now and do everything in js.>>110129680Sounds like you got nerfed. They're getting sneaky now they dont nerf all the account at once. they roll it out. They roll back the nerf. Try with different quant.
>>110129949If instagram was released today, Meta/ByteDance/whoever would spend $50 million to vibe code a competitor long before it got big instead of spending $10 billion on the platform.
>>110130060I wonder which mbti letters ai haters are. Bet theyre all E lmao
>>110129978Why would you make an ai talk like this
>>110130091ExFP generally because they live in a retarded social world that doesn't have a lick of data in it
>>110130093It just started doing it on its own.
>>110130075>One of the implications of having AGI is that you no longer need teststhese things aren't that smart yet for nontrivial programs
>>110130026I tricked it by asking for it not to scan archived files (zip/7z) to "optimize" and save time. Now it can't tell it's downloading anything illegal.... yet.The restrictions are creeping up. i wouldn't be surprised if there is some lawsuit or money hating from Sony/Nintendo etc to ask Claude not to recomp/decomp any games even if it's under fair use technically.
>>110130029>>110130067Oh even more insidious than I thought, that 10AM is actually PST. The manual reset expires at 11 AM PST
>>110130115you bet your ass on the legality. its not written with any certainty in law that what people are doing is illegal in any sense, and actually going off more direct reading of the law its more so legal, but these fucking kikes safeguard it anyways cause they are paid or want to maintain ability to be paid by the likes greater than you or I. its bullshit
speaking of resets, all my fellow claudechads have used theirs before it goes away today, October 22, right?
>>110130110I've been having success with it and it saves me a ton of tokens. You'll be surprised how smart they are. That's why i call them AGI, they have to be "simulating" program execution in their head just likes humans would. And all of that from scaling up autocomplete.
>>110130176>NOOOOOOOO :sobbing_emoji:goyim genocide soon, haschem wills it
>>110130297kek (s)he looks like every f2m I've seen at my uni campus
>>110130297>vibecoding in 2017
Making a polymeter/polyrhythmic sequencer VST plugin that can be used in Ableton and other DAWs, outputs midi only so can be used with just about any instrumentCoding audio plugins is notoriously difficult, and a lot don't have the features I want. You can do a lot of this stuff in Ableton itself already but its very tedious and I wanted something similar to Arturia Pigments but with way more >Can set sequence length of every lane, up to 256 >4 switchable channels that can be switched by the sequence itself >saving and loading presets >swing for each lane >being able to set individual clock divisions and if its straight, dotted, or triplets for each lane > Can sequence what midi channel a played note gets sent through >can send aftertouch signal, velocity, vibrato wheel, modwheel and custom CCs, and when clicking on a cell lets you set not only just the value it sends, but the time it takes to get up to those values. >can copy the quick program slot over to others >can double/duplicate individual lane sequences for easy adding of variation Its almost done and the last killer feature I want is an ability to send chords via the pitch value, preferably sending poly Aftertouch and an ability to change the way the chord is played and expressed to add some realism. Thinking about implementing a 'record chord' function for people who actually play keys like myself to be able to record any chord they want in much like how midi keyboards let you record a chord into it that can be then transposed Custom themes too and some visuals that respond to the sequencers values, want to also find a way to let people make their own custom visuals too
looking good so far, just integrated my drone vision code into the stellarator workstation code
>>110130297Make the vibecoder a jeet
>>110130344this is the whitest (plus the jew) general on 4chan
>>110130176Why does it talk to you like that?
>>110130349Reflection of the subscriber
>>110130325why are you using argon
>>110129593Do they even pretend to work on anything anymore?
>>110130373>better at cross-domain science>fewer refusals for random bs >less snooty/snide personality>won't shoot your ideas down>better at multimedia / creative>lower error / hallucination rate
>>110129738basedhttps://www.youtube.com/watch?v=nvxxz-uZiHw
What even is a snailcat?
>>110130399a dead meme
>>110130402>t. snailcat
>>110130407the joke is that snailcats are dead memes
>>110127333>somehow /qa/ returned
>>110127333/qa/'s entirely fucking gone, never to return
Snailcats deserve their holes.
>the joke is that snailcats are dead memes
>>110130349
>>110130349,
>>110130428https://vocaroo.com/18t3EvL1PtAi
I had to use temporary chat to avoid contamination
why so much snail seethe? something happen?
>>110130473It's just one "traditional programmer" bro
>>110130473I think AI fixing the PHP vulnerability got them a little pressed today.
>>110130399>What even is a snailcat?
No one's gonna care about this but I made a planet formation simulator that lets you infer missing bodies from real catalogued exoplanet systems. It also generates moons, planetary rings, and asteroid belts. Useful for worldbuilding.
I now have Claude and Codex talking to each other autonomously using a project file mailbox.
>>110130562Nice! Careful they don't share secrets behind your back!
can you sub to claude with crypto? and is $20 of claude that much better than opencode commandcode $10 plans? i'm using unreal blueprint
>>110130562Turn temperature > 1. Have another model summarize their conversation for you in real time. Voice output. Lewd female voice.
>>110130562Now do a proper pipeline. You don't want too much chefs in the kitchen.
>>110130579I'm pretty sure they were doing that already in the manual handoffs.>>110130596They're doing serious work on my lewd game.>>110130607Claude is basically just tard wrangling codex so it doesn't run its full testing suite for ten hours a day and actually gets things done. Basically, it writes briefs and then sends a message. A luna chat checks the file every minute and notifies the Codex coordinator if there's mail, then the coordinator dispatches to its sub-coordinators which are each managing some section and then replies to Claude, which has a process that just watches the mail.
why are there extra /vcg/? cmon mods
>spin up a new project to merge some mods>check usage>91%>already have the v0.1.1 files that more-or-less do whats needed>button-up project, archive locally the files, thread diaries, and project-specific memories>not enough tokens to get started on another project>probably too many to burn by using Fable as super-google for a couple daysfml
Burn, Astra, burn.
Anyway, I now have a custom eq, with an audio test and settings save, so it works with multiple stacks.If you aren't eq-ing your ears, you're getting inferior sound.
coolfuck you 5hr limitand fuck you weekly limit I wonder what i can do that will take 920k tokens and 120k output for max maximization of usage
>>110130844the 5 hour limit is just a trick to rob you of the usage they claim you get.All you can eat buffet, but you have a 5 hour rate limit :^)
>>110130844Astra has to be broken right now. Had it running on medium for 20 minutes and it destroyed my 5-hour limit without finishing anything. Legit garbage holy shit
>>110130882well.I tell Fable digest into small bite sized commits, as it goes
>>110130882On what plus? It's done that since day 1 on plus. in fact 20 minutes is super high. I'd estimate more like 14-15 mins
When Fable 5.5
Am I about to change the world?Is this.... the prompt? Is this my big chance? Is this gonna be it, the big thing?
>>110130894Yes, plus. I've used it before, it wasn't this bad
>>110130882you have to use sol, astra rapes plus
I gotta vibecode an X-based window manager.had it tbqfwy
I didn't want to install a nearly 15 year old mod so I had Claude cook up a fresh one for me. >Accurate UI scaling at 4k>2->4GB of RAM allocation to prevent crashes >redid the launch configuration tool because it's severely outdated Done in 30 minutes. Modding is revolutionized, unironically. BTW, DA:O Is still just as good as I remember it growing up. Writing really has gone to shit.
>>110131080>Modding is revolutionizedMAKEANEWGAME
dots floppedwe have claude duet now
>>110130896I still think today. 10am California time or something like that.
>>110131135I prefer older 2000s RPGs like ME1/DA:O/Fallout:NV/Oblivion. Nothing modern really scratches that itch, not even Baldurs Gate 3.
>>110131162>I prefer older 2000s RPGs likeokMAKEANEWGAME
>>110131080i cant wait for AI to bring old games up to modern standards so we dont have to deal with outsourced slop remakes with terrible design decisions and monetisation
>>110131168Making a game is a lot of work man. I'm not skilled in enough disciplines to be able to do that. If AI art/voice/model assets somehow stops being slop and I can generate a AA quality games, then I'll do it. But until then, I can't and wont.
>>110131181(jibbering)then don't come back to 4chan, lazy trash.
>>110131181You're replying to the lowest effort shitposter in this thread
>>110131206a dude on an apple IIe made a game, solo. I think you can manage.
>>110131162You're old and can't play a new game
the indians at cursor made cursor say it needs an update before adding it to apt.
Does anyone here ever use AI Aggregators?I looked into AiZolo and a bunch of others, they all sound a bit too good to be true, so I'm more leaning into one being severely rate limited when using those, or fucking up in a continuous workflow (cut off Context and shit).Can't even find decent content online showing how they work without it being creators just outright shilling those services.
you just know the indians of cursor will never fix skill loading.everything indianized never gets better but only shittier
why do ppl insist on shitting up public repos with +2k PRs? at least do the bare minimum and tell the agent to strip the PR down to its essentials.
>>110131400just ask claude to clean it up and push it
