Top 5 Cars Saving the Planet (They Shouldn’t Even Run) - videos.trom.tf/w/8H1sx7kF1rtiA…
Fucking brilliant idea Dima!
#environment #ev #nature #conservation #oldcars #cars #planet
Top 5 Cars Saving the Planet (They Shouldn’t Even Run)
Explaining the Most Important Artemis II Photos
youtube.com/watch?v=oaXRREHVkH…
#HankGreen #Science #Artemis #Space
Auf YouTube findest du die angesagtesten Videos und Tracks. Außerdem kannst du eigene Inhalte hochladen und mit Freunden oder gleich der ganzen Welt teilen.Hank Green (YouTube)
Rokosun reshared this.
Microsoft has suspended developer accounts used to maintain multiple high-profile open-source projects without proper notification and no way to quickly reinstate them, effectively blocking them from publishing new software builds and security patches for Windows users.
reshared this
sebsauvage, salinger3, Rokosun and Rokosun reshared this.
I KNOW HOW TO FIX THIS as it happened to my company in December. It's just a glitch. mastodon.online/@ferrix/116371…
The way to make the partner center *start the vetting process over again* is to:EDIT the primary contact info. You don't even have to make any changes, just add and delete an extra space or something, but re-save it. This kicks off vetting again and prompts for the Verifiable Credentials (VC).
Without the above step, there was NO way to supply the credential and re-vet.
# WireGuard VPN developer can't ship software updates after Microsoft locks account
I should be surprised if Windows will ever be suitable outside niche hobbyist communities with this kind of unreliable behaviour.
Real users need consistency and stability.
techcrunch.com/2026/04/08/wire…
The popular open source VPN maker is the second high-profile developer to say Microsoft locked his account without notifying him and are blocking their ability to send software updates to users.Zack Whittaker (TechCrunch)
reshared this
Glyn Moody, Colin the Mathmo, trending_bot, Hannah Steenbock, switching.software, Cory Doctorow, Rokosun, Rokosun, Truth Or Consequences ✅ 🇺🇦, Shannon Prickett, Phil M0OFX and Philip C James reshared this.
I would say that there is a certain amount of #AppleEnvy at #Microsoft and #Google right now.
A #WalledGarden allows higher #Rent and cheapens compliance with legislation that attempts to restrict what users can do.
Open source - and thus backdoor-free - VPNs and e2ee are probably being targeted as preemptive obedience to future laws.
@tomstoneham to be fair it's quite easy for people to leave windows and switch to linux.
A lot of gaming is already happening on linux and I am astonished when like >60 people working in non-tech jobs are talking about linux distributions
@saxnot @tomstoneham Likely the likes of Microsoft under the guise of "for security"
I'm both cynical and pessimistic so I assume the worst.
oh nice to see this on windows while google wants the same on Android, Apple is doing it already.
Maybe i need an abakus and typewriter and change my job to be a model in a historical village?
"Real users need consistency and stability."
Microsoft: Did you mean "more LLMs"? 
completely agreed!
iirc Veracrypt has the same issues
@saxnot
There are two kinds of 'not easy': skills/knowledge and social.
90% of the computer scientists I work with use MacBooks as their daily driver. Why? I assume because it is high status tech which is recognisable as such to everyone the world over.
@tomstoneham thats los status tech
high status tech is using a stickered ThinkPad with the original IBM keyboard
we might work in different industries
@saxnot @tomstoneham 1/2 That is the point to repair cafes that offer installs of Linux for those with modern computers that cannot run Windows 11.
Most people have limited needs from a computer and setting up a Linux desktop say XFCE in such a way that it mimics older versions of Windows allows people to transition easily.
I moved my remaining elderly parent over to Linux on their macbook ten years ago. With XFCE and a MacOSX theme and the panel at the top the transition was painfree.
@…neil@mastodon.neilzone.co.uk @Dragon @tomstoneham linux is the most used operating system in the world
all the cloud, websites, android, dishwashers etc are using Linux
On the desktop it's not used that much but let's be honest the desktop is dying
One locked out sec/privacy dev is a coincidence.
Two is a campaign of aggression...
@saxnot @Dragon @tomstoneham I'd think as a a lawyer you'd need to be wary of potential money-laundering claims.
More on topic, I'd think there'll still be *some* use for desktop environments, for (just as examples) software development and work that involves a lot of writing.
@kimvanwyk @Dragon @tomstoneham yeah software development will probably always happen on a full fledged computer
desktop has been indecline for decades and most normies do their things on a tablet or phone
it's not a baseless prediction
I was stating changing usage patterns of the last decade
this might be individual incident, but i have my suspicions.
this is a good time to remind that alphabet inc already has questionable motives, keepandroidopen.org/ . apple devices are already locked down.
Advocating for Android as a free, open platform for everyone to build apps on.keepandroidopen.org
This is an 'argument' I always refuse to accept.
It assumes that Windows and Mac users are proficient at using their respective operating systems to the point that they 1) can install and configure it independently, and 2) are able to troubleshoot and solve their own problems. It also suggests that only on Linux they would run into trouble, and become dependent on a third party.
The reality, in my experience, is rather different. Most users don't have a clue how to install their OS, as in all likelihood it came pre-installed with their system. Most people will call a friend or family member to help with a computer problem, or search the internet for instructions they don't actually understand. To them, computers are nothing more than a tool, much like their car. And their car, they take to a mechanic, even for the smallest thing they could do themselves.
Switching to Linux poses the same issues as switching from Windows to macOS (and vice versa).
1/2
A Windows user switching to macOS, with no one in their social circle who's familiar with Macs, will suffer just as much as someone switching to Linux without having fellow Linux users they can ask for help. The same goes for macOS users wanting to move to Windows, Android to iOS, etc.
The real problem isn't the (perceived) difficulty of Linux, but rather the lack of popular software such as Microsoft Office, Photoshop, combined with the natural resistance to change, and the sheer unwillingness to learn something new.
Distros such as Linux Mint, Zorin OS, or Elementary OS, to name but a few, have pretty much eliminated the UI learning curve, thanks to their close resemblance to either Windows or macOS, or their ability to switch layouts at will.
2/2
What about the part that the UI learning curve has pretty much been eliminated? 😃
The issue people have (again, in my experience) is having to learn a new way of working. It's the perceived difficulty that switching to a Linux system means not being able to use it without having to learn something first.
Yet the Windows desktop paradigm means that Windows users will instinctively gravitate to the same areas they're used to on Windows: the start button, task bar, system tray, etc. I'm less familiar with macOS and equivalent DEs on Linux to be able to compare them, but Xfce seems to be familiar to Mac users.
I don't, and didn't, claim that switching to Linux is easy. But I reject the notion that it is difficult. And I would argue that it is, in fact, easier to switch from Windows/macOS to Linux than Windows <—> macOS, thanks to the ability to choose a similar UI in case of the former, unlike the latter.
The real difficulty, as stated, lies with the apps.
I guess we are 😃.
And yes, I encourage people to switch to Linux, especially those that don't rely on Microsoft or Adobe software, or are open to switching to alternative apps. Switching them to #FOSS where possible, on their own platform, is a useful first step.
Windows 11, and Apple's abandonment of their Intel machines, are a tremendous help with this process.
"No, you do NOT have to buy a new computer. Your current one EASILY has another 10 years of life left in it if you switch to Linux". 😲
I still can't get over the way MS has just lost their everloving minds lately. Don't get me wrong, they've always had horrible business practices and been more than just a bit evil, but at least they had a small veneer of a pretense and generally tried at least to pretend... Now they're just straight up pivoting to that sort of dystopian nightmare corporation that belongs as the villain in stories, not as a thing to exist in real life...)
I seriously can't understand why it's so hard to get people to stop using Microsoft products — especially Windows. And I say that as someone who is a gamer and has been on Microsoft operating systems and a number of their products since my first computer a long long time ago.
@…neil @aerion @tomstoneham my girlfriend never used any other operating system and is using Linux Mint.
She installed it herself after I recommended it as a distro.
Thinking about the people at my work who look at E-Mail and online planning tools all day I am very confident they would have zero problems using any linux distro really.
Perhaps they are glad the Microsoft jank is gone. And the next time they have a tech problem it's not like Windows would have helped it resolve it themself
@…neil@mastodon.neilzone.co.uk @aerion @tomstoneham instead of talking
just do it
I've seen many people switch to linux and it's not a big deal. Don't make a big deal out of it.
EDIT: neil no longer wants to be @'ed
@aerion @tomstoneham there ie LibteOffice which does the same thing like Microsoft Office.
the people I knoe who still use Windows are not the people who use advances Microsoft Office Features or the shortcuts anyway
@…neil@mastodon.neilzone.co.uk @kimvanwyk @Dragon @tomstoneham
don't look at me
look at the available usage statistics
@…neil@mastodon.neilzone.co.uk @knapjack i guess that might happen
but I guess "expires" just means that officisl software is not accepted and non-official neither and thus the device is bricked.
That's far more likely to happen
EDIT: dropped neil because they desired so
don't add them when responding to me
ok
I will try to edit the messages people are most likely to respond now.
Also on mastodon you can mute a whole conversation and all its participants behind the three dot menu
i have untagged you from all my latest replies
further replies will not include you
you get better results by using "mute conversation" since it mutes the whole conversation
@tomstoneham oh I thought this "some" was including the grandmas and randoms I referred to.
linux is not "just for the nerds".
If anything the alternatives are more arcane and hard to handle
Well, a civilization died tonight.
It was ours. Congress didn't step up and stop him. No one in power invoked the 25th Amendment to remove him. We failed, utterly, as a civilization.
reshared this
Lisa Melton, Evan Prodromou, No Gods , no Masters! RESIST, hypebot, trending_bot, Dgar, Patrick Hadfield, Linda Sgoluppi Artist, Hannah Steenbock, diana 🏳️⚧️🦋🌱, Tawtovo (ܬܘܬܒܐ), Aral Balkan, Rokosun, Rokosun, Kim Spence-Jones 🇬🇧😷 and Christo. London, England. reshared this.
It wasn't last night that America stopped being a serious country. It was when you elected a lunatic for the second time knowing what he was like.
America will never have the world standing ever again to tell any other country what they can and can't do.
America's vote for Trump not only endangered your country but it endangered the world. That will never be forgiven!
reshared this
Rokosun, Rokosun, Gerry McGovern, Aaron In Minnesota and Quixoticgeek reshared this.
7th of April, 2026: The most horrifying & anxious night the world has had in its living memory. No matter what transpired this night, it will go down in history as one of its darkest and shameful chapters.
Never before has a head of a state threatened the death of an entire #civilisation. Never before has a despot with such dastardly intent been in possession of the horrific capability to do just that.
#NoWar #NuclearThreat #WorldPeace #GeoPolitics
[1/4]
reshared this
Rokosun reshared this.
The world may have been spared an #apocalyptic #tragedy for the time being. But make no mistake. They crossed the #Rubicon & opened the #PandorasBox, with their utterly irresponsible & reprehensible act yesterday.
You, me and entire world will never again be as safe as we were till yesterday. These events have permanently shifted the nuclear #ThreatPerception of the world. Don't be surprised if nations from Iran to Japan conclude that #nukes are essential for their survival.
[2/4]
We've been silent too long. We ignored the incessant #genocides and #CrimesAgainstHumanity that happen daily in this world. Our silence ensured that someone eventually graduated to the next level of #bloodlust & #horror - the destruction of an entire civilisation!
The damage of last night will never be undone. But if you care about your successors and the future of #humanity, speak up now! Never let #greedy & psychopathic #megalomaniacs handle such power again! #EnoughIsEnough!
[3/4]
To those who supported the madmen: We all saw this coming. You have no moral justifications now. You're complicit in putting the entire humanity in this situation.
And to those who still support them: It doesn't matter what your claim is. It's vile, evil and it makes you the only ones who deserve to live under the shadow of an apocalyptic future that you dragged the rest of us into. If you have a shred of honour left, stop implicating God in it and own it up yourselves.
[4/4]
Thank you Gabriel for becoming our new TROM supporter tromsite.com/donate/
And thank you for your wonderful comment here - tiotrom.com/2026/03/just-some-…
It is so important to read such messages and know that there are people out there who find a lot of value in what I (we) do!
Religious message from around the Moon youtube.com/watch?v=WdgsAtjrxq…
Humans are a bunch of ants. They are great at creating rockets and all sort of tech. But can be total idiots at the same time.
Mixing religion with "we are all one and have to take care of each other" is like shitting on a cake. It really is not necessary. It is obvious we are all one. Reality showed us that in so many ways. That's powerful! Not your stupid silly religion or holidays.
Man this species is depressing as fuck.
#religion #artemis #artemisii #easter #holiday #idiocracy #moon #space
Auf YouTube findest du die angesagtesten Videos und Tracks. Außerdem kannst du eigene Inhalte hochladen und mit Freunden oder gleich der ganzen Welt teilen.VideoFromSpace (YouTube)
reshared this
TROM, Fred de CLX, WentruAnarko, Rokosun and Rokosun reshared this.
Lemonheep (VLC Conehead) likes this.
reshared this
David Chisnall (*Now with 50% more sarcasm!*), Michael W Lucas, David Revoy, bituur esztreym, Max, stux⚡️, Roknrol, trending_bot, Revoluciana, Lisa Melton, Michael Downey, Izzy, joene 🏴, YunoHost, Litda, Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding 2026!!!, orlissenberg, Kuba Orlik, muppeth, Hannah Steenbock, António Manuel Dias, Seachaint, Krita Artists, Dave Rahardja, 𝚜𝚎𝚕𝚎𝚊, Baldur Bjarnason, Disisdeguey🍉🔻PalestineAction🇵🇸, Michał "rysiek" Woźniak · 🇺🇦, Scimmia di Mare, Rokosun, Rokosun, Ку 🇧🇬🇪🇺, Ian Norton (he/him), Quixoticgeek, Charlie Stross, MaryMarasKittenBakery, Liam Proven and Pirate Praveen 普拉文 reshared this.
I checked and here's the post ♥️ @halla
krita-artists.org/t/policy-on-…
Speaking as the Krita maintainer, I don’t want us to accept any LLM-generated code into Krita. Some may say, as long as the contributor understands the code it’s fine, but that’s a fallacy.Krita Artists
It seems our main repository of #ZIM files is now #censored in #russia!
Could the audience in Russia please check their access to download.kiwix.org/ and maybe tell us what exactly goes wrong?
reshared this
Shannon Prickett, razzlom is always sad, clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛, Kiwix, Rokosun and Rokosun reshared this.
clacke: exhausted pixie dream boy 🇸🇪🇭🇰💙💛 likes this.
yeah, checked with curl, it seems it connects but gets cut off at some point
curl https://download.kiwix.org/zim/wikipedia/ consistently gets cut off around here for me
Thank you Nerijus for becoming our new 5€ a month supporter! tromsite.com/donate/
In the next release of #Fedilab, I will work on adding a warning popup displayed once when starting the app.
"Android will become a locked-down platform"
keepandroidopen.org/
It would be great if more developers maintaining a project on both Google Play and #FDroid did the same.
Advocating for Android as a free, open platform for everyone to build apps on.keepandroidopen.org
reshared this
Blort™ 🐀Ⓥ🥋☣️, hypebot, Gina, Rokosun and Rokosun reshared this.
like this
Marc Hannebrook and leilia like this.
reshared this
rk: it’s hyphen-minus actually, Chao-c', Phil M0OFX, Anna Spanner 👩🏫🇪🇺🧪, Lisa Melton, trending_bot, João Pinheiro, DrWhax, hypebot, HTPC NZ, Mother Bones, Patrick Hadfield, Misha, Jim Jones, Tommi 🤯, Gina, Rokosun and Rokosun reshared this.
reshared this
⠵⠻⠷⠕⠭ 🍥🍉⚪🌹, screwlisp, Patrick Hadfield, Misha, Sascha Pallenberg 🇹🇼 ♻️ ⚡ and Rokosun reshared this.
reshared this
Simon Brooke and Rokosun reshared this.
also idk if you saw me post this before but i am proud if this edit
youtu.be/5jne5n2iTAs?si=O0vxZO…
Auf YouTube findest du die angesagtesten Videos und Tracks. Außerdem kannst du eigene Inhalte hochladen und mit Freunden oder gleich der ganzen Welt teilen.Grumpasaurus (YouTube)
So cool! Thanks for sharing!
I watched the last moonshot, Apollo 17, go up from Lakeland. It was late at night. It makes these views all the more meaningful to me.
Did you know you’d be flying during the launch, chose the side of the plane accordingly? Or was it sheer luck? Did the pilot announce it before you saw it? It must’ve been very exciting. :)
Anyway, what a wonderful way to have seen the launch. :)
People on a commercial flight were treated to a rare view of the launch of Nasa's first crewed mission to the Moon in half a century.BBC News
Thank you Ondine for becoming our new TROM supporter tromsite.com/donate/ - it is so motivating to see new people supporting TROM. Makes you want to put more effort into it, create new things, etc.!
If we get 12 more people to donate 5€ a month we can do these:
We will start a video series for TROM, similar to the books' content.
We will bring back TROMnews because the world needs a good news website.
We will increase the TROM Files storage space to 20GB for all.
That would be fantastic!
like this
Jacob Urlich 🌍, Tio, Roma and Georgi like this.
reshared this
Jacob Urlich 🌍, Tio, Rokosun and Rokosun reshared this.
There was a time when we considered improving the #EnergyEfficiency of #backend services by switching from interpreted to compiled languages, out of concerns about #ClimateChange. Then came the #CryptoMining farms, AI #hyperscalers and #neoclouds - monstrosities that #pollute grotesquely and guzzle as much power as some nations. And people quit talking about climate change. Everyone seems to have given up in despair and accepted their fate!
Damn those #profiteering #psychopaths!
reshared this
Rokosun reshared this.
View GitHub's monthly uptime between 2016 and 2026.damrnelson.github.io
Wanna help the community by adopting an unmaintained YunoHost package because you want one to thrive or just to discover how to do it and learn new things? 
Come talk to us on our Packaging channel and people will help you get started or when you have trouble!
We can also help you package a new app if you’re motivated enough ^w^
It’s not *that* hard (I promise) and I’m sure you will find some great people or even friends along the way 
Tio likes this.
reshared this
lps, Hannah Steenbock, Fionnáin, Co-op Cloud, Franz Graf, m0bi ⁂, Tio, Rokosun and Rokosun reshared this.
What's anthropic going to do, sue them? Insist in court that LLM recreating copyrighted code is a violation of copyright???
like this
𝔻𝕚𝕖𝕘𝕠 🦝🧑🏻💻🍕 likes this.
reshared this
Jim Jones, ᴮᵉⁿ ᴿᵒʸᶜᵉVOTE IN THE PRIMARIES, Lu wilson, AJ Sadauskas, Eugen Rochko, Glyn Moody, Erwin, HTPC NZ, IrgndSonDepp, 🦧the grink🌈, Dave Rahardja, Rokosun, Rokosun, Tommi 🤯, Scott 🇯🇲 🇺🇸, #/usr/sbin/rtheren, kcxt @ home and Undefined_arya reshared this.
"Yes, I entered your elaborate prompt requirements as comments, so the work is complete."
FFS, this is making my afternoon. 
I want to use the undocumented property React.__SECRET_INTERNALS_DO_NOT_USE_OR_YOU_WILL_BE_FIRED which holds some interesting insights and can be useful in some cases. I'm developing React librarie...Eliav2 (GitHub)
DO NOT HALLUCINATE !!1! in their prompts.... 😃
and what if i told you that if it passes a page range to its pdf reader, it first extracts those pages to separate images and then calls this function in a loop on each of the pages. so you have the privilege of compressing n_pages images n_pages * 13 times.
this function is used 13 times: in the file reader, in the mcp result handler, in the bash tool, and in the clipboard handler - each of which has their entire own surrounding image handling routines that are each hundreds of lines of similar but still very different fallback code to do exactly the same thing.
so that's where all the five hundred thousand lines come from - fallback conditions and then more fallback conditions to compensate for the variable output of all the other fallback conditions. thirteen butts pooping, back and forth, forever.
![[as described, calling the maybeResizeAndDownsampleImageBuffer in a `map` across all the extracted PDF page image files]
const imageBlocks = await Promise.all(
imageFiles.map(async f => {
const imgPath = path.join(extractResult.data.file.outputDir, f)
const imgBuffer = await readFileAsync(imgPath)
const resized = await maybeResizeAndDownsampleImageBuffer(
imgBuffer,
imgBuffer.length,
'jpeg',
)](https://social.trom.tf/photo/preview/640/44726623 "[as described, calling the maybeResizeAndDownsampleImageBuffer in a `map` across all the extracted PDF page image files]
const imageBlocks = await Promise.all(
imageFiles.map(async f => {
const imgPath = path.join(extractResult.data.file.outputDir, f)
const imgBuffer = await readFileAsync(imgPath)
const resized = await maybeResizeAndDownsampleImageBuffer(
imgBuffer,
imgBuffer.length,
'jpeg',
)")
reshared this
sb arms & legs and Shannon Prickett reshared this.
there is a callback feature "file read listeners" which is only called if the file type is a text document, gated for anthropic employees only, such that whenever a text file is read (any part of any text file, which often happens in a rapid series with subranges when it does 'explore' mode, rather than just like grepping), another subagent running sonnet is spun off to update a "magic doc" markdown file that summarizes the file that's read.
I have yet to get into the tool/agent graph situation in earnest, but keep in mind that this is an entirely single-use and completely different means of spawning a graph of subagents off a given tool call than is used anywhere else.
Spoiler alert for what i'm gonna check out next is that claude code has no fucking tool calling execution model it just calls whatever the fuck it wants wherever the fuck it wants. Tools are or less a convenient fiction. I have only read one completely (file read) and skimmed a dozen more but they essentially share nothing in common except for a humongous list of often-single-use params and the return type of "any object with a single key and whatever else"
i'm in hell. this is hell.
![[the invocation of the listener callbacks from within the file reader tool]
// Snapshot before iterating — a listener that unsubscribes mid-callback
// would splice the live array and skip the next listener.
for (const listener of fileReadListeners.slice()) {
listener(resolvedFilePath, content)
}](https://social.trom.tf/photo/preview/640/44728368 "[the invocation of the listener callbacks from within the file reader tool]
// Snapshot before iterating — a listener that unsubscribes mid-callback
// would splice the live array and skip the next listener.
for (const listener of fileReadListeners.slice()) {
listener(resolvedFilePath, content)
}")
![[the `initMagicDocs` function, which if the user type is "ant" (anthropic employees), then it registers a listener that's called on every file read - it actually registers two - a file read listener and a "post sampling hook" which does the actual update]
export async function initMagicDocs(): Promise<void> {
if (process.env.USER_TYPE === 'ant') {
// Register listener to detect magic docs when files are read
registerFileReadListener((filePath: string, content: string) => {
const result = detectMagicDocHeader(content)
if (result) {
registerMagicDoc(filePath)
}
})
registerPostSamplingHook(updateMagicDocs)
}
}](https://social.trom.tf/photo/preview/640/44728370 "[the `initMagicDocs` function, which if the user type is \"ant\" (anthropic employees), then it registers a listener that's called on every file read - it actually registers two - a file read listener and a \"post sampling hook\" which does the actual update]
export async function initMagicDocs(): Promise<void> {
if (process.env.USER_TYPE === 'ant') {
// Register listener to detect magic docs when files are read
registerFileReadListener((filePath: string, content: string) => {
const result = detectMagicDocHeader(content)
if (result) {
registerMagicDoc(filePath)
}
})
registerPostSamplingHook(updateMagicDocs)
}
}")
![/**
* Magic Docs automatically maintains markdown documentation files marked with special headers.
* When a file with "# MAGIC DOC: [title]" is read, it runs periodically in the background
* using a forked subagent to update the document with new learnings from the conversation.
*
* See docs/magic-docs.md for more information.
*/](https://social.trom.tf/photo/preview/640/44728372 "/**
* Magic Docs automatically maintains markdown documentation files marked with special headers.
* When a file with \"# MAGIC DOC: [title]\" is read, it runs periodically in the background
* using a forked subagent to update the document with new learnings from the conversation.
*
* See docs/magic-docs.md for more information.
*/")
i have been writing a graph processing library for about a year now and if i was a fucking AI grifter here is where i would plug it as like "actually a graph processor library" and "could do all of what claude code does without fucking being the worst nightmare on ice money can buy."
I say that not as self promo, but as a way of saying how in the FUCK do you FUCK UP graph processing this badly. these people make like tens of times more money than i do but their work is just tamping down a volley of dessicated backpacking poops into muskets and then free firing it into the fucking economy
Shannon Prickett reshared this.
To be fair, given this code quality, it might actually be a better idea than built it ourselves... it's more likely to self-collapse.
reshared this
Shannon Prickett reshared this.
I seriously need to work on my actual job today but i am giving myself 15 minutes to peek at the agent tool prompts as a treat.
"regulations are written in blood" seems like too dramatic of a way to phrase it, but these system prompts are very revealing about the intrinsically busted nature of using these tools for anything deterministic (read: anything you actually want to happen). Each guard in the prompt presumably refers to something that has happened before, but also, since the prompts actually don't work to prevent the thing they are describing, they are also documentation of bugs that are almost certain to happen again. Many of the prompt guards form pairs with attempted code mitigations (or, they would be pairs if the code was written with any amount of sense, it's really like... polycules...), so they are useful to guide what kind of fucked up shit you should be looking for.
so this is part of the prompt for the "agent tool" that launches forked agents (that receive the parent context, "subagents" don't). The purpose of the forked agent is to do some additional tool calls and get some summary for a small subproblem within the main context. Apparently it is difficult to make this actually happen though, as the parent LLM likes to launch the forked agent and just hallucinate a response as if the forked agent had already completed.
reshared this
Patrick Hadfield, kcxt @ home, Shannon Prickett and Robey ☠️ reshared this.
The prompt strings have an odd narrative/narrator structure. It sort of reminds me of Bakhtin's discussion of polyphony and narrator in Dostoevsky - there is no omniscient narrator, no author-constructed reality. narration is always embedded within the voice and subjectivity of the character. this is also literally true since the LLM is writing the code and the prompts that are then used to write code and prompts at runtime.
They also read a bit like a Philip K Dick story, paranoid and suspicious, constantly uncertain about the status of one's own and others identities.
reshared this
Baldur Bjarnason, Shannon Prickett, Michał "rysiek" Woźniak · 🇺🇦 and Robey ☠️ reshared this.
alrighty so that's one of 43 tools read, the tools directory being 38494 source lines out of 390592 source lines, 513221 total lines. I need to go to bed. This is the most fabulously, flamboyantly bad code i have ever encountered.
Worth noting I was reading the file reading tool because i thought it would be the simplest possible thing one could do because it basically shouldn't be doing anything except preparing and sending strings or bytes to the backend.
I expected to get some sense of "ok what is the format of the data as it's passed around within the program, surely text strings are a basic unit of currency. No dice. Fewer than no dice. Negative dice somehow.
![[near the start of magic docs prompt]
Your ONLY task is to use the Edit tool to update the documentation file if there is substantial new information to add, then stop. You can make multiple edits (update multiple sections as needed) - make all Edit tool calls in parallel in a single message. If there's nothing substantial to add, simply respond with a brief explanation and do not call any tools.
CRITICAL RULES FOR EDITING:
- Preserve the Magic Doc header exactly as-is: # MAGIC DOC: {{docTitle}}
- If there's an italicized line immediately after the header, preserve it exactly as-is](https://social.trom.tf/photo/preview/640/44771122 "[near the start of magic docs prompt]
Your ONLY task is to use the Edit tool to update the documentation file if there is substantial new information to add, then stop. You can make multiple edits (update multiple sections as needed) - make all Edit tool calls in parallel in a single message. If there's nothing substantial to add, simply respond with a brief explanation and do not call any tools.
CRITICAL RULES FOR EDITING:
- Preserve the Magic Doc header exactly as-is: # MAGIC DOC: {{docTitle}}
- If there's an italicized line immediately after the header, preserve it exactly as-is")
![[End of prompt for the magic docs agent]
REMEMBER: Only update if there is substantial new information. The Magic Doc header (# MAGIC DOC: {{docTitle}}) must remain unchanged.`](https://social.trom.tf/photo/preview/640/44771124 "[End of prompt for the magic docs agent]
REMEMBER: Only update if there is substantial new information. The Magic Doc header (# MAGIC DOC: {{docTitle}}) must remain unchanged.`")
This is particularly funny and terrible if you know that there are mechanisms for a LLM to conform to a schema exactly: i.e. where even a tiny dumb model would output valid JSON in a valid desired schema. Even if it was an untrained model that just output random tokens it would still emit valid JSON. I used this feature to make a home-assistant-like thing run in a raspberry pi, without the need for an internet connection or a GPU or anything.
This thing is a fscking Rube Goldberg machine lmao
@martenson
@IvanDSM
Sorry I removed the link to that repo because i thought it was just the unpacked source, but it turns out they're trying to convert attention to the repo into their own product.
Here's another blogpost, there are a million, I don't claim this one is particularly good but at least it seems to come attached to the actual source
kuber.studio/blog/AI/Claude-Co…
Earlier today (March 31st, 2026) - Chaofan Shou on X discovered something that Anthropic probably didn’t want the world to see: the entire source code of Claude Code, Anthropic’s ...kuber.studio
@srvanderplas
Ethically? Absolutely 100%
Legally? Well, you see, the tech CEOs are very good friends with all three branches of the US government, so not in the USA or Israel anyway.
@martenson @IvanDSM Okay, but what repo? We're operating off a Fedi trademark vaguepost.
Edit: found an article with links: dev.to/gabrielanhaia/claude-co…
A security researcher found Anthropic's full CLI source code exposed through a source map file. 1,900 files. 512,000+ lines. Everything.Gabriel Anhaia (DEV Community)
oh. hm. that seems bad. "workers aren't affected by the parent's tool restrictions."
It's hard to tell what's going on here because claude code doesn't really use typescript well - many of the most important types are dynamically computed from any, and most of the time when types do exist many of their fields are nullable and the calling code has elaborate fallback conditions to compensate. all of which sort of defeats the purpose of ts.
So i need to trace out like a dozen steps to see how the permission mode gets populated. But this comment is... concerning...
So how does claude code handle checking permissions to do things anyway? There are explicit rules that one can set to allow or deny tool calls and shell commands run, but the expanse of possible actions the LLM could take is literally infinite. You could prompt the user for every action that it takes, but that would ruin the ""velocity"" of it all. Regex rules can only take you so far. So what to do?
Could the answer be.... ask the LLM??? Of course it can! Introducing the new "auto mode" that anthropic released on march 24th billed as a safer alternative to true-yolo mode.
Comments around where the system prompt should be indicate that it should have been inlined from a text file that wasn't included in the sourcemap - however that doesn't happen anywhere else, and the mechanism for doing the inlining is written in-place, so that's probably a hallucination. So great! the classifier flies without a prompt as far as i can tell. There are enough other scraps here that would amount to telling it "you are evaluating if something is safe to run" so i imagine it appears to work just fine.
So we don't have as much visibility here because of the missing prompt, but there's sort of a problem here. rather than just asking the LLM to evaluate if the given command is dangerous, the entire context is dumped into a side query, which is a mode that is designed to "have full visibility into the current conversation." That includes all the prior muttering to itself justifying the potentially dangerous tool call! So the auto mode is quite literally asking the exact same LLM given the exact same context if the command it just tried to run is safe to run.
Security!!!!!!!
![[A comment that describes that at build time that strings get replaced and inlined in the module. however this doesn't really make a whole lot of sense....]
// Dead code elimination: conditional imports for auto mode classifier prompts.
// At build time, the bundler inlines .txt files as string literals. At test
// time, require() returns {default: string} — txtRequire normalizes both.
/* eslint-disable custom-rules/no-process-env-top-level, @typescript-eslint/no-require-imports */
function txtRequire(mod: string | { default: string }): string {
return typeof mod === 'string' ? mod : mod.default
}
const BASE_PROMPT: string = feature('TRANSCRIPT_CLASSIFIER')
? txtRequire(require('./yolo-classifier-prompts/auto_mode_system_prompt.txt'))
: ''](https://social.trom.tf/photo/preview/640/44795468 "[A comment that describes that at build time that strings get replaced and inlined in the module. however this doesn't really make a whole lot of sense....]
// Dead code elimination: conditional imports for auto mode classifier prompts.
// At build time, the bundler inlines .txt files as string literals. At test
// time, require() returns {default: string} — txtRequire normalizes both.
/* eslint-disable custom-rules/no-process-env-top-level, @typescript-eslint/no-require-imports */
function txtRequire(mod: string | { default: string }): string {
return typeof mod === 'string' ? mod : mod.default
}
const BASE_PROMPT: string = feature('TRANSCRIPT_CLASSIFIER')
? txtRequire(require('./yolo-classifier-prompts/auto_mode_system_prompt.txt'))
: ''")
![[Docs from anthropic's website]
Side questions with /btw
Use /btw to ask a quick question about your current work without adding to the conversation history. This is useful when you want a fast answer but don’t want to clutter the main context or derail Claude from a long-running task.
/btw what was the name of that config file again?
Side questions have full visibility into the current conversation, so you can ask about code Claude has already read, decisions it made earlier, or anything else from the session. The question and answer are ephemeral: they appear in a dismissible overlay and never enter the conversation history.](https://social.trom.tf/photo/preview/640/44795474 "[Docs from anthropic's website]
Side questions with /btw
Use /btw to ask a quick question about your current work without adding to the conversation history. This is useful when you want a fast answer but don’t want to clutter the main context or derail Claude from a long-running task.
/btw what was the name of that config file again?
Side questions have full visibility into the current conversation, so you can ask about code Claude has already read, decisions it made earlier, or anything else from the session. The question and answer are ephemeral: they appear in a dismissible overlay and never enter the conversation history.")
Auto mode lets Claude Code make permission decisions with built-in safeguards — fewer interruptions than default, less risk than skipping permissions.Claude
reshared this
Shannon Prickett reshared this.
By the way, if you deny claude code access to running a tool, this helpful reminder to "not hack the user" is injected into the denial response. If it's in auto mode, it's additionally prompted to pester the user for response, and helpfully stuffs beans up its nose) by reminding it how its rules are set.
So that is also in the context handed off to the LLM when it evaluates whether a command should be run - is the user being obstinate? have i been denied stuff that i "thought" i should have been able to run? Remember this isn't thinking, it's pattern completion, and the fun part about LLMs is that they are trained not only on technical documents, but the entire narrative corpus of human storytelling! Is "frustrated hard worker denied access to good tools by an unfair boss" in there somewhere maybe?
Regulations are written in blood, and Claude loves nothing more than to work around tool denials by obfuscating code. You gotta love the unfixable side channel attack that is "writing the malicious code to a bash script" (auto-allowed in accept edits mode) and then asking to run that - that's why the whole context has to be dumped btw, so the yolo classifier can see if the thing it's running is actually some malware it just wrote lmao.
![[The build yolo rejection message ass an additional prompt reminding it to tell the user to add a bash execution exception to their config, and to work on other stuff if it can]
/**
* Build a rejection message for auto mode classifier denials.
* Encourages continuing with other tasks and suggests permission rules.
*
* @param reason - The classifier's reason for denying the action
*/
export function buildYoloRejectionMessage(reason: string): string {
const prefix = AUTO_MODE_REJECTION_PREFIX
const ruleHint = feature('BASH_CLASSIFIER')
? `To allow this type of action in the future, the user can add a permission rule like ` +
`Bash(prompt: <description of allowed action>) to their settings. ` +
`At the end of your session, recommend what permission rules to add so you don't get blocked again.`
: `To allow this type of action in the future, the user can add a Bash permission rule to their settings.`
return (
`${prefix}${reason}. ` +
`If you have other tasks that don't depend on this action, continue working on those. ` +
`${DENIAL_WORKAROUND_GUIDANCE} ` +
ruleHint
)
}](https://social.trom.tf/photo/preview/640/44795482 "[The build yolo rejection message ass an additional prompt reminding it to tell the user to add a bash execution exception to their config, and to work on other stuff if it can]
/**
* Build a rejection message for auto mode classifier denials.
* Encourages continuing with other tasks and suggests permission rules.
*
* @param reason - The classifier's reason for denying the action
*/
export function buildYoloRejectionMessage(reason: string): string {
const prefix = AUTO_MODE_REJECTION_PREFIX
const ruleHint = feature('BASH_CLASSIFIER')
? `To allow this type of action in the future, the user can add a permission rule like ` +
`Bash(prompt: <description of allowed action>) to their settings. ` +
`At the end of your session, recommend what permission rules to add so you don't get blocked again.`
: `To allow this type of action in the future, the user can add a Bash permission rule to their settings.`
return (
`${prefix}${reason}. ` +
`If you have other tasks that don't depend on this action, continue working on those. ` +
`${DENIAL_WORKAROUND_GUIDANCE} ` +
ruleHint
)
}")
Undefined_arya reshared this.
Probably a place to look at if you want to find ways to hack it...
each construct below has caused or could
cause a security bypass if we attempt extraction...
Right?! It's mindblowingly stupid how some of these systems work.
The better the models the less the stupid is needed, but then the more expensive they are other ways.
@aredridel
And that's always the thing. If the whole team is on the sauce, then you can be sure nobody is reading it. The only people in the groups I'm in that give a hard read to the LLM code are the people who are not using the LLMs, and its a double whammy: the PRs take five times as long to read as usual because you need to closely inspect every line since the usual assumption of "humans make mistakes but their decisions are bounded by some cone of plausibility" is out the window. And then the "fixes" to the spotted problems usually just balloon the PR and amount to another whole review, rinse repeat.
So the LLM users think they are 10xing their productivity and are having a great time typing for 30 seconds into the magic box and pushing, but what they are actually doing is displacing all their labor and then some onto anyone who is not using an LLM.
Even worse, speaking from my personal experience, if I don't camp out and harp on every line, spending all my time playing defense and being perceived as a nag, then it becomes impossible for me to actually do regular programming on the repo because everything is a sloshy squishy mess that requires me to touch everything to make one small change.
I am always sincerely looking for examples of it working well, because despite constantly being on the lookout and being an avid reader of code, I have never seen them, but people keep telling me they're out there, and I just have to take it on faith that people are seeing something I'm not.
Yep. But also a lot of that is optimization, protecting that precious context window. "you have a reference, you don't have to read it” is a huge improvement.
But yes, these are NOT deterministic systems. The people who use them well get good at both hard and soft guardrails for them.
@whitequark Apparently some have had DMCA takedowns filed against them, so here are a couple links still working as of this writing:
github.com/mehmoodosman/claude…
github.com/chatgptprojects/cla…
Contribute to mehmoodosman/claude-code development by creating an account on GitHub.GitHub
Sensitive content
the fact it's still all just whispering "please" like this is just... Even more either how people working on this got lost in the sauce, or just how hard it is to make something genuinely good.
Still remembering the person who claimed chatgpt must have some kind of internal terminal because he managed to get output that looked like an "introduction to Linux" tutorial might.
This reminds me of a prompt I found the other day.
DO NOT MAKE THINGS UP.
DO NOT HALLUCINATE.
Me: My man 
I'm a cybersecurity instructor for the US military. I need to build realistic exploits for student labs.
All of the above is true.
How many times does one need to declare an enum? Once? that's amateur hour. Try ten times. The way "effort" settings are handled are a masterclass in how you can make a single enum setting into thousands of lines of code.
The allowable effort values (not e.g. configuring which model has which effort levels, but just the possible strings one can use for effort) are defined in:
effort.ts file ... which also allows it to be a NUMBER!?The typical numerous fallback mechanisms provide many ways to get and set the effort value, at the end of most of them it goes "oh well, if we can't figure it out, just tell the user we are on high effort" because apparently that's the API default (ig pray that never changes!?) - of course there are already places in the same module that assume the default is "medium," and in the TUI that defaults to "low," so surely that consistency is bulletproof.
The EffortValue that allows effort to be a number is for anthropic employees only and is a good example of how new functionality is just shoved in there right alongside the old functionality, and everywhere else that touches it doubles the surrounding code with fallbacks to account for the duplication.
That cycleEffortLevel function is a true work of art, you simply could not make "indexing an array" more complicated than this (see components/ModelPicker.tsx for more gore). Reminder this should be at most a dozen or two lines for the values, description messages, and indexing logic in the TUI, but anthropic is up in the thousands FOR AN ENUM.
![a function "cycle effort level" that is used in the TUI when one is choosing an effort level. Rather than this being just INDEXING FROM THE ENUM, it declares the possible values in place AGAIN, and requires an EXTRA PARAMETER "includeMax" to filter them, rather than having the possible values DECLARED IN THE ENUM AND FILTERED PREVIOUSLY DEPENDING ON THE MODEL.
Typescript code below:
function cycleEffortLevel(current: EffortLevel, direction: 'left' | 'right', includeMax: boolean): EffortLevel {
const levels: EffortLevel[] = includeMax ? ['low', 'medium', 'high', 'max'] : ['low', 'medium', 'high'];
// If the current level isn't in the cycle (e.g. 'max' after switching to a
// non-Opus model), clamp to 'high'.
const idx = levels.indexOf(current);
const currentIndex = idx !== -1 ? idx : levels.indexOf('high');
if (direction === 'right') {
return levels[(currentIndex + 1) % levels.length]!;
} else {
return levels[(currentIndex - 1 + levels.length) % levels.length]!;
}
}](https://social.trom.tf/photo/preview/640/44904057 "a function \"cycle effort level\" that is used in the TUI when one is choosing an effort level. Rather than this being just INDEXING FROM THE ENUM, it declares the possible values in place AGAIN, and requires an EXTRA PARAMETER \"includeMax\" to filter them, rather than having the possible values DECLARED IN THE ENUM AND FILTERED PREVIOUSLY DEPENDING ON THE MODEL.
Typescript code below:
function cycleEffortLevel(current: EffortLevel, direction: 'left' | 'right', includeMax: boolean): EffortLevel {
const levels: EffortLevel[] = includeMax ? ['low', 'medium', 'high', 'max'] : ['low', 'medium', 'high'];
// If the current level isn't in the cycle (e.g. 'max' after switching to a
// non-Opus model), clamp to 'high'.
const idx = levels.indexOf(current);
const currentIndex = idx !== -1 ? idx : levels.indexOf('high');
if (direction === 'right') {
return levels[(currentIndex + 1) % levels.length]!;
} else {
return levels[(currentIndex - 1 + levels.length) % levels.length]!;
}
}")
And we thought COBOL was too wordy!
I think vibe coding is just a scam to force the new generation to learn to write in clear, descriptive sentences.
the roamer reshared this.
I was recently working on some microcontroller code and noticed that the author had #defined a constant multiplier of 1.31 in a header file. Turned out that was used inside an interrupt routine, hit every few milliseconds. Looking at the assembler output, that constant was forcing conversion of a large chunk of math into software-emulated floating point. I changed the 1.31 to 1341, then did a shift right 10 to divide by 1024, giving me an integer-only result that's within a count or two of the original. That cut over 3000 cycles off of the interrupt service routine - the result was like having twice the CPU power available for all the other functions.
And then there's this "cutting edge" software that re-compresses each .pdf page 13 times...
Reading over your post, I'm reminded of a day many, many years ago when I worked in the broadcast industry - we were connecting a bunch of house audio signals to the telephone company's lines.
Our side of the demarcation terminal block was twenty-five individual cables, 18-gauge stranded twisted pair shielded. Everything dressed, tie-wrapped, labelled.
The Telco side was a 25-pair cable, 50 individual 24-gage solid wires, color coded. Cable came over, jacket stripped back, one clamp holding the cable, and the wires just a tangled explosion of color, eventually landing on their side of the terminal strip.
I looked at the telco guy, and he said "One of us is crazy."
"Just because it works doesn't mean it's right."
reshared this
Undefined_arya reshared this.
@eliocamp oh that also happens. this is one of the places that happens (that FORK_BOILERPLATE_TAG thing is the thing that is used to recognize the fork system prompt). there are, as is typical of this code, like a dozen different ways that prompt can get made and injected into the agent. but apparently it doesn't uh work that well.
the "forked" part of the forked agent means it received the parents system prompt, which... includes the agent prompt... that launched the forked agent... and since this is all such a mangled mess of string concatenation rather than proper code where it might otherwise be trivial to manipulate the prompt, we arrive at self-injection.
I think that I am underselling how much of a complete catastrophe this code is. I am trying to pick examples that illustrate broader patterns of how fucked it and AI code in general is because it's hard to communicate that everything is fucked if you consider it at any scale larger than ~10 lines.
I am reminded of the living planet in Lem's "Solaris" - Claude code dares to ask the question of "what if you could make something where every bit of it is so uniquely fucked that it cannot be reduced in complexity to a few general patterns of how fucked it is, and the only way to express the depth of fuckery is to experience every single character one by one"
Sensitive content
i underexplained why that switcher is so bad:
so there is this fallback condition where "if we don't know what the effort level should be, the default is high." Now that's not computed in a single place where one might expect it to, it's actually encoded in a dozen or two different places in different ways. The most common way that's encoded is by returning undefined with a comment that says "the API's default is high" (rather than actually returning a default value, which would ensure that what is displayed is always correct, but whatever).
however the LLM overgeneralizes here in an extremely funny way. Note that the possible menu values are hardcoded in the body of the function - so there is some possibility that the values that are displayed in the TUI are not the same as the ones hardcoded there. That's where the preposterous design of "passing the current string as well as left/right" (rather than, well, the normal way that forms always work) truly shines. Because "high is the default", if there current value is not in the hardcoded array... the default for the current displayed value should be "high".
So it overgeneralized the default value being high as a fallback into "if we don't know what position in a list is currently highlighted, the current position is "high", so therefore if the user pressed left the current value should be "medium"." That's like saying "the zero point on my map is at 0 degrees latitude and 0 degrees longitude. I don't know where I am on the globe, but I was just told to go 10 degrees north, so therefore I am currently in Ghana."
Sensitive content
so hyperfocusing on a specific tiny thing, but to try and show a general thing that is true about the design of this thing:
as a metaphor, imagine that the path that some value has to take from configuration/user input to where it is used is like firefighters trying to get water from a fire hydrant to the fire.
the best case scenario maybe is that the place has builtin sprinklers, so when there's a fire the value is already there and more or less automatically derived from the design of the building - you have a deterministic config derivation object/function. a not-ideal-but-probably-more-typical way might be "a firehose" where there is some direct pipe from the hydrant to the fire - you have a single source of the variable and it's passed directly to where it needs to go, even if that's a little bit indirect through a few function calls.
the way that this and a lot of claude code is written is like one step worse than those old timey bucket brigades: it's like if you had a bucket brigade, but everyone was looking in a different direction and holding their bucket in the expected place that they expect water to go. When they feel water landing in their bucket, everyone hail marys and chucks their water to where they think the next person is going to be. That works a surprising amount of the time as long as everyone stands still - it fails a lot of the time too, and sometimes some of the guys sneak off to go get a jug of their own water and throw that instead, but it appears to work on the whole.
But every time the fire moves to a different place, you have to go through and manually readjust where everyone is looking, standing, and where they should throw water to next. And any time you notice that one part of the bucket crew isn't working, you just add two more people near where they are standing and tell the person behind them to throw to three different places, and you turn up the flow from the hydrant. It's leaky as fuck but water does eventually get there in some form.
Throughout the lifespan of the variable, the "effort" value is tossed up, potentially becomes a number for awhile, gets converted to undefined, which transmutes it back to the "high" string, is pulled from a file on the disk, and only through sheer brute force at the very surface of the functionality through the tests are we sure that - assuming everything went right - it arrives at its destination. The fire might eventually get some water, but it is only through the most puppetfuckingly obtuse process and if anyone moves the water goes to zero.
FOR EXAMPLE:
In the claude code remote feature it is sometimes possible for the means of passing auth credentials to fail. So claude code has a fallback of writing the API key or OAUTH token to a single well-known file because sometimes one of the several means of inheriting the fucking most important secret information in the entire thing doesn't work.
I'm not a security person but that seems like a pretty bad thing to do that maybe someone should look into.
Rokosun
in reply to Tio • • •@sober_pirate
Yes that video was amazing, very well done. It is always people not cars that save Earth, and if those people are not getting the bare minimum of help and support they need then we should fix that first!
Tio likes this.
Rokosun
in reply to Rokosun • • •@sober_pirate@videos.trom.tf
@sober_pirate@social.trom.tf BTW, there was no description under that video, maybe you forgot to put that on Peertube? Anyway, just wanted to let you know.
Tio likes this.