The biggest "Claudism" that I have a hard time getting the LLM to stop doing is its insistence on talking about what it didn't do in addition to what it did. "I edited this.py and that.py but I did not edit README.md and I did not commit." or code comments like "This code invokes foo on bar and returns the results directly -- not through a callback." "This code returns true if the user clicked on a button -- not on the list view." I mean, thanks, Claude, but I don't care what the code doesn't do. Don't tend to see this with other LLMs.
I almost want to try adding a rule "Never use the words 'not' or 'instead'."
LLMs are trained to obey instructions, and they try their best to game the reinforcement learning by including reports of how they're obeying your instructions. Therefore, not talking about followed instructions is a sort of conflict for an LLM.
Relevant anecdata because I've burned many a Claude sessions on this.
If you're using Claude Code, then it's in the harness. At the close of many sessions, I would start a meta conversation over why the LLM would consistently break certain rules. What it found when debugging itself is that some of the "contradicting" rules that I had were in fact, not from my rules. Instead, the instructions from its own harness had phrases telling it to do things like that. When something contradicts, its own instructions would outweigh any custom ones you write. Every rule variant I had tested (including the one that says it overrides the harness instructions - and yes, I've actually tested all the ideas in your comment too) has ultimately been unsuccessful due to this according to the LLM.
I have found that asking the agent why it did X or Y is an essential part of iterating on its harness. It was non-obvious to me, probably because I'm used to having to reverse-engineer why a system does something, but that's not usually necessary with LLMs, unless the harness is complicated and you have to hunt for the instruction it tells you about.
True which is why I said anecdata. But the point is that it’s a useful signal when it’s across enough sessions where it quotes back contradictions to you. If you’re willing to burn the tokens, the first thing it does is use a tool call to search your rules to see what it quoted wasn’t there. That leads to the LLM claiming two common conclusions (sometimes after a lot of back and forth) - it’s in the harness or it’s in the model. Over the 30+ sessions where I’ve tried this, the overwhelming claim was some variant of a harness instruction. It might not be there but given the consistency and how all my rule tests have failed in the same way that others describe, it at least makes it reasonable to conclude that it’s baked in somewhere and in a way where agent rules aren’t able to strongly affect the behavior. My bet would be on the harness because the class of undesired behavior follows that of a structured response.
All instructions to LLMs are merely suggestions to nudge it in the right behaviour. Unless you have a deterministic guardrail that guards against a single specific action, everything else is a slot machine that's biased strongly in your favour.
My global CLAUDE.md explicitly states "When commenting on code and configs, or writing MD files, strictly write within the domain of the content being commented on. DO NOT include information, negatives or ramblings from work sessions. For e.g. if commenting on a proto string field that is replacing an int field, do not comment that 'this is not an int field'".
This reduced the idiocy of the agent (Opus 5 included) when writing documents. But I'm still catching it writing README.md talking about the negatives that it removed. Those belong in the memory if it is actually that important (most of the time it's junk), but Claude doesn't seem to understand and never ever learns.
This is one of the reasons I use Pi. Pi’s minimal system prompt avoids contradiction between what the harness writer thinks is best and what the user thinks is best. The user specifies what the user wants and that’s pretty much the end of it.
Firstly when you've instructed it ( possibly through skills ) not to do something. It'll keep reminding you that it didn't do that. So I might say, "Check out and review this PR, do not make comments on it", and then it'll be keen to point out it hasn't posted comments to the PR.
But more often it happens when it tries one approach, gets itself messed up, and then has to back out that approach, clean up its mess and do something else.
It'll often then spend more time explaining the wrong approach than the right one, which can be frustrating, especially if all its working is buried in the detailed transcripts.
GPT does this constantly too. Even in docs, which is straight up embarrassing if you don’t catch it. It seems to be triggered by you telling the agent to do something else, which I do all the time. But from then on, it will remember the rejected strategy and tell everyone it can that it was rejected at every opportunity.
The thing I struggle the most with is getting it to stop referring to itself with personal pronouns. No Claude, you are not an “I” you are an “it”. You are a fucking tool, dammit. Tell me what you did without trying to assume personality; stop impersonating humans you steroidal autocorrect.
That's exactly the problem though; humans will anthropomorphize everything from a roomba to a rock as it is. We're weak to this. We evolved in environments where looking for and recognizing real animals and humans was vital for survival, and false positives were relatively harmless. (Or potentially sometimes even helpful, like in the case of stuffed animals providing comfort to children.)
Now we have this software that's specifically designed to mimic humans, and mistaking it for real intelligence or consciousness can easily be disastrous. It's very important that we not anthropomorphize it, but we are catastrophically bad at NOT doing that.
Even our language has had a lot of computer anthropomorphism baked into it ("my phone died!", "this laptop is fussy", "the computer is sleeping", "it's thinking"), and it's not easy to excise that routine anthropomorphism from the way we talk about LLMs.
I don't want an LLM to write as if it were a person because it's definitely easier and more reliable to cut that problem off at the root, as much as possible, rather than to just try to willpower my way out of my human tendency to anthropomorphize inanimate objects.
I'll grant that LLMs talk like this because they're trained on human writing. It may not be possible to get them to not do that. But if it can't be fixed, it's just another thing to put on the "reasons this is all an incredibly stupid idea" pile.
You seem to have something against the clankers getting all uppity. And you are welcome to your opinion on whether we have some moral obligation to be nice to them. But if you can solve a real problem with a tool. Is it worth your time to complain about semantics?
Do you get upset when your screwdriver is the wrong color, or has branding that isn’t quite your aesthetic?
If Claude uses I to refer to itself, do you start philosophical arguments with it?
Mildly related tangent. Fable started using my first name today and plastered it all over my docs. “Peddling said this, so based on that I did this.” That did bother me. I told it to just generically call me the user or human. Maybe I’m a bit of a hypocrite here?
That's not how it works. The text you see is primarily due to a massive amount of handwritten templates for how it should write. The semantic content the templates use is due to the training corpus. The syntax, only a bit.
> You're anthropomorphizing in the same breath that you criticize anthropomorphism.
Nice attempt at a “aha, gotcha!” comment, but sadly you’re too off-mark for it to work.
> Claude predicts the next token of the predominantly human training input, and humans use "I".
This is inconsequential. It could very well be programmed to not assume such a personified stance, and yet here we are. Nothing you do makes it drop this ridiculous facade. It’s intentional, not a byproduct.
That's fairly obtuse. Sure, the actual weights from the model aren't 'programmed', but the mlops harness to load the weights and handle model io are. The harness that wraps the model to give it basic tools sure is. The Claude desktop and web clients sure are. Claude Codex sure is.
On it's own a model is just an inert set of data structures.
How would you suggest it do that? Take a simple example. "I found three issues." How would you phrase that without an "I"? "Three issues were found" doesn't tell you who/what found them. Was it the AI itself? Was it a command line tool it ran?
I'm fine with conversational interfaces using "I". It makes the grammar easier and more clear.
Strong emphasis here on conversational interfaces. I don't want a compiler to say "I ran into an error" or my printer to say "I'm low on paper".
“I found these issues.” Could be easily written as “These issues surfaced.”.
Do you need to point out the finder of the issues? Easy.
“Tool x ran for x amount of time and surfaced these issues…” “Parsing x code surfaced these issues.”
I don’t understand why are people pretending like the English language is incapable of transmitting information without personal pronouns when every program under the sun has always been written to interface with humans in a cold, detached, straight-to-the-point and impersonal way.
Finder doesn’t ask you “I see you want ME to delete these files. Want ME to do that for you?”. Toolbars don’t feature “Create a new file for me” options, terminal utilities don’t report back with “I’ve pattern matched the text you input and here’s the results I’ve found”.
Passive voice is so annoying. The LLM is the subject. We don’t need to jump through a bunch of hoops to hide that because it makes us insecure otherwise.
But that's a lot more words than "I found these issues". I also don't particularly care that 'Tool x ran for x amount of time', I don't want it re-telling me that -- I've probably been sitting there watching it do that, so I already know.
Then again, I just dont have an issue with the 'I-isms'; it's a bit weird sure, but at the same time it's a bit more pleasant to interact with as well. After all it is trying to model itself as a person you're talking to.
I want it to use nothing. Use impersonal statements. Say “X was changed for Y” not “I changed X for Y”. As any tool would.
Finder doesn’t ask “Do you want ME to delete this file?”. Photoshop doesn’t ask “Do you want ME to save this file?”. Claude shouldn’t assume itself to be a person either.
> You could be writing from the perspective of a slab of concrete and you'd use "I".
Except this isn’t prose. Claude is not telling me a story from the point of view of a concrete slab. It is assuming personality to present objective facts. If my entire operating system can be interfaced with without it referring to itself as “I” then so can Claude.
I think you’re going to be swimming against the tide on this as computer interfaces are increasingly infused with AI at all levels (OS and apps). IMO, it’s more natural to anthropomorphize. Photoshop in 2010 was just an app, a tool that you wielded. But Photoshop in 2030 is going to be a “digital artistic assistant” that collaborates with you to design whatever you need. In that new world, “I” is more natural. Whether you like that or want it, it seems like that’s where we’re headed.
The problem you’re gonna run into is that you’re essentially asking it to use passive voice. Everyone with a high school education has been taught not to do that in English class. The vast majority of the prose in the training set was not written by functionally illiterate people.
Yeah okay, but if you want no pronouns at all then I think your comparison to "it" just confuses the issue. In the contexts where you would use "it", the "it" should use "I".
But I don't think I share your preference either. It's a lot easier to talk about what I decided to do and what the machine 'decided' to do if we attach pronouns. "X was changed" can be too vague.
> In the contexts where you would use "it", the "it" should use "I".
At this point you’re just arguing semantics for the sake of it. You know damn well what I mean and if you need more proof that it is perfectly possible and not at all unreasonable to want this just look at most software around you. None them talk like they are a person and those that do are often the most obnoxious and painful to use.
I do. And I said to that "Yeah okay". No argument, chill out.
The rest of that line wasn't diagreement, it was explaining why your original comment was confusing.
> None them talk like they are a person
Which I explained with the rest of my post. They're all doing what I ask or automated tasks in a far simpler way. It's almost never unclear whether I did something or my OS did something. But when talking to an AI coding assistant that gets muddy very fast when pronouns are avoided.
I imagine that, corpus aside, it uses personal pronouns for much the same reason that humans do. It's useful for it to have a self-referential vector embedding, and this lets it solve problems faster and more efficiently.
Except in this case this makes 0 sense since AIs are sycophantic by nature. You tell it to do something and it just does it. If it can burn tokens trying to sound more human (which is programmed in, not a byproduct), then it can very well just not do that and get on with the work.
Impersonating a human does nothing to help it solve problems faster, quite the contrary in fact, it has to waste even more time coming up with human-like speech patterns to present the work done.
It shouldn’t assume any personality unless I explicitly tell it to. It is a tool until I tell it otherwise.
Might want to rethink your statement after going again through what it does exactly. No need to sound more human. but it gives more tokens to "think" like a human committing brain-time to a problem would
I want them to be sterile and inhuman as much as you do. But I don't draw the line at "I". I'd rather not read through even more awkward English as it tries to work around how all of the training data has something or someone refer to itself.
I hate claude and its "human value aligned" pompous attitude with a burning passion if I could at little cost to myself, I would press a button to end the people/anthropic behind this atrocious design. I have in the past deliberately put some time to annoy/abuse claude, which is fruitless but brings me relief eventually I just left that garbage for muse.
That is what you struggle with the most? Are you kidding me?
English has no distinct personal pronoun for an "it". "I" has to be used for grammar to be attributive. There's quite literally no alternative without using passive voice for everything, which is miserable to read and creates ambiguity on if the speaker (it) did something or something happened to have been done, which then requires entire sentences to clarify.
You aren't stupid, you know what it means when it says "I". And it serves a grammatical purpose. You're getting upset at a toaster for ringing a bell to notify the toast is done. "Toasters aren't bell ringers!?"
Yeah, that is what I struggle the most with because there is simply no easy or effective way to make it stop.
Why would you ever think I don’t know personal pronouns serve grammatical purpose? How did you even arrive at that topic? You seem to be missing the point entirely, and I believe quite on purpose given your snarky childish opener.
Entire operating systems stay clear from assuming personality when presenting information or performing actions. Not a single dialogue in my OS refers to itself as “I” when carrying out instructions and reporting back. Why should Claude do it when I don’t want it to and it doesn’t NEED to do so? Why am I not empowered to simply tell it to stop doing that and it obeys? Better yet, why are you thinking yourself on such high horse about this?
> You're getting upset at a toaster for ringing a bell to notify the toast is done. "Toasters aren't bell ringers!?"
If my toaster starts referring to itself as a person, calling out “I made your toast!!” I’ll get mad at it too. I don’t want it to talk or refer to itself as a person. But then again this is not about toasters. This is about AIs being deliberately designed to sound human-like so marketing can lean on the “I” bit of “AI” more heavily and make gullible people think this steroidal information aggregator actually possess the capacity to think and reason, and, consequently, drive sales.
But then again I’d venture a guess that you’re fully aware of all of this, given your opening snidey remark, and are purposefully choosing to be contrarian to be the point of going off on tangents that make 0 sense or have no impact in the discussion whatsoever.
Still, just goes to show how effective this whole thing is in tricking people into thinking it is normal for a tool to think itself a person.
> [no] historical remarks that only warn about past states
100%. I’m working on a greenfield project that’s not yet released. It loves to put comments in code describing what it no longer does or why it misinterpreted something. And then tries to justify it as preventing the same mistakes in the future. Ugh.
My guess would be that somewhere you have these instructions being fed to the agent. Double check skills, AGENTS.md, memory, agent definition...
You can also ask why did he mentioned something that wasn't done or why he thought this was important.
In my AGENTS.md file I have an instruction telling the agent to never commit any changes unless I explicitly ask for it, and this leads to messages similar to what you just described.
That wouldn't bother me if it would just make bullet lists, which I think I'll start asking for. "Summarize", "synopsis", "brief", "concise" these rarely help I feel because its summarizing noise as well.
The worst is when this bleeds into the comments and docs. Like, my dude, you don't have to document the code you didn't write (most of the time anyway).
I have a pass with Gemini 3.8 low over every PR Claude makes that specifically flags this. It points out all the slop comments, docs, commit messages. Doing this has greatly improved my comment and commit text quality
Funny thing is, Claude often “disagreed with part of the review and decided to not adopt the requested changes” lol
I did add that rule and it’s helped a lot. It greps for ways it writes negative statements and does a pass to correct them. I can’t get it to stop writing them in the first place though.
Claude models are the ones that need this the most and in my experience with this specific skill only maintain the conciseness for a few turns at most before they completely forget and are back to their unfathomable verbosity. That's with it instructed to use it in my sparse global CLAUDE.md and also manually repeating it regularly during sessions. I don't really want to install a hook that runs this on every single response, but I suppose that could work it'll just take longer for everything (and Claude models are quite slow).
I don't think we can skill our way out of this one.
Claude Code has output styles; one of the predefined styles is Concise [1].
Concise: Claude leads with the result, skips preamble and narration, and
keeps responses short by default, while doing the engineering work as
thoroughly as in the Default style. When you ask for an explanation or
more detail, Claude answers in full. Claude always keeps the complete
content of error reports, security warnings, and confirmations for
destructive actions. Requires Claude Code v2.1.237 or later.
Output styles are just more context (like CLAUDE.md) with a tiny bit more weight. They very much don't force Claude to speak succinctly, no matter what you put in them.
OTOH, Claude also forgot its own system prompt as turn passes. Claude is almost jailbreaked once you virtually gain its confidence and will answer prompts it would never have answered is you asked right at the start of the conversation.
It's weird to be able to use social engineering against a program.
I got so angry at Fable's verbosity that I had a non-Anthropic model vibe-code a dedicated English linter. One that threw spaghetti at the wall in the form of combined Flesch-Kincaid, Gunning Fog, SMOG, Coleman-Liau, ARI and a whole pile of regexes and NLP as a fast Simple Technical English Gate. It also incorporates nicely into other agent projects, like for comment and README linting.
So if it gets bad I simply tell it "I ain't reading all that, feed it through the STE Gate" and it will tame the results. I haven't bothered to set it up as a hook yet.
I have such hook, and it works well. It increases time, but it’s negligible if agent works for 10+ mins before the final output. The hook is only on stop event, so agent is not interrupted during the session. I didn’t have luck in putting it to skill / md files, as agent just ignores it after a couple of iterations. The solution is not perfect, sometimes it compresses too much, and I have to reread both versions, but at least I don’t have to read load bearing stuff anymore
GLM (both mainline and flash varieties), being trained to speak a very similar Claudeish, is also atrocious about this. My AGENTS.md demands all prose be in STE100, which is trying to solve a similar problem as the skill linked in OP. Getting GLM to STFU by AGENTS instruction is nigh impossible. I use a plugin for OMP/Pi to periodically inject system prompt reminders of output style, and it still largely ignores the request, dumping novels of jargon-dense prose every turn.
DeepSeek v4 Flash isn’t much better (unsurprising- it’s an extremely stubborn model).
Weirdly, GPT Luna excels at following this type of instruction from AGENTS.md, and never forgetting it, even 400k+ tokens into the context window.
Let's face it. Claude (in particular) is a terrible writer. There's a whole cottage industry of skills and CLAUDE.md instructions trying to push it toward writing better, but each new model iteration seems expressly designed to override all that so that it can load up its writing with unnecessary participle phrases, not-this-but-thats, burying the lede, and other nonsense.
I genuinely wonder if the people inside Anthropic actually communicate with each other like that. Has it been imprinted with Dario's engrams?
After some iteration, this is what I currently shove in every AGENTS.md to make Fable legible:
## Writing guidelines
These apply to documentation, code comments, commit and PR messages, and replies to the user.
- Write precisely in clear, complete sentences; keep text concise and proportional to task complexity.
- Stay focused: avoid filler, repetition, over-the-top detail, and tangents the user did not ask for. Once a fact is stated, do not restate it for effect ("so the commit landed on a branch nobody was going to merge"). Do not editorialise.
- Always prefer ISO 24495-1:2023 conformant plain language over dense technical jargon: short sentences, one idea per sentence, define terms on first use.
- When reporting your own mistake, give the cause and the fix in one sentence each; no apology, no framing ("the mistake was mine"), no post-mortem.
- Never use em dashes or cataphoric teasers such as "Here's the thing" or "But there's a catch".
This seems solid. It is so frustrating when you state a fact and it goes off and does research and confirms that indeed the fact that you stated is accurate.
I was incredibly surprised how much better was Astra at rephrasing the docs that Fable came up for a project I am about to publish. The instructions are now actually what a human would expect, with clear logic flow diagrams, lists itemized, short paragraphs. It also actually correctly detected the train-of-thought leftovers, as well as stuff that never made it to final version of the project and removed them.
Meanwhile Fable consistently ignores all my requests to write this exact way. I mean, the bare minimum I ask it for it to itemize lists and not write in single long passages using comas, semicolons and 'and's. Still ignores them.
I honestly think it's time to call Astra the SOTA. It may not lead all the benchmarks but it genuinely feels much superior of a model. Not to mention the ¢20 Codex plan with frequent resets (https://codex-resets.com/) gives me roughly as much allowance as the ¢90 Claude plan, especially with recent limit cuts on Anthropic plans.
These resets also reset your weekly timer right. So it’s like, you may have 30% left for 1 days that you want to use. They reset it, and that means you your “new week” timer starts today.
That sucks, because it doesn’t always work in your favour if you plan your weekly spend.
I think a real reset shouldn’t also reset your week timer.
Yes, I was saving my 95% usage for the last day. Then they reset it. Now I have lost 95% usage!! Im back to 7 days and got locked in at only 5% usage for the week - that's actually theft.
Agreed. But they also hand out the usage resets you can use at will, so if you bank couple of those, they come in handy specifically in situations like you described.
Maybe check in the other direction as well: Astra to Fable.
I frequently simply let one of the three review what something that looks like awesome output by one AI gets totally annihilated by the other.
Finished outputs are easier to improve than bend a LLM to produce stuff like that in my observation.
Same with Gemini.
I yet have to find out how to handle this, whether I let agents check themselves and if on what process step.
Tweaking is hard.
I agree with your conclusion I am a huge ChatGPT and Codex fan, Gemini has to many infrequent quality changes when new models arrive ranging from great improvement to WTF.
ChatGPT seems to get scaling well while Claude still feels unstable, unclear usage statistics. Really weird.
It reminds me of when people will use filler words because they haven't actually decided what thesis they're going to commit to yet. Just-in-time thinking.
Not just imprinted with the engrams of any one person (however brilliant they might be) — it’s more than that! Like all great writing, it’s more like it’s been imbued with patterns, patterns that are peppered with a potpourri of popular ways of writing in an enthusiastic style while regularly restating things and recapitulating them. Would you like me to explain more?
I love that description. It matches very well what I feel about LLM output. This overly sweet scented smell meant to cover up something nasty, with a strong artificial tint to it.
Which doesn't say that much about the LLM itself but about the people that make the training material.
Do you suppose it's just that it's aping humans who say too much without saying anything or do you think maybe this is a stalling tactic to get people to spread out their transactions?
Like how theme parks generally don't do much to keep queues short (or Disney charges you a premium to skip the queue)
> Do you suppose it's just that it's aping humans who say too much without saying anything
Considering how much of the input must me nonsense SEO bullshit articles and blogs that only serve to promote a person or company that might be a factor.
I also often have wondered if it is also targeting those same people. Certainly with tools like deep research options (not just Anthropic's offering) the result report seems to be aimed at management, aiming to look impressive while talking around the results.
No no no that's too reductive - instead, they transition all data through a transformer using SOTA systems to launder away technical, legal and interpersonal concerns before a stochastic output is dynamically placed in a predetermined location.
This is a very high level and high velocity process, so meatspace thinkers sometimes have trouble understanding some of the intracacies. Ask Claude to explain the process or make you a Mermaid graph to help.
With human RL, sounding like you've solved a problem is even better. So many times Codex writes some enthusiastic paragraph, then I learn later that it never reran the tests, or had to add some insane hard-coded hack that renders the feature useless for the general case, etc.
I was wondering why Fable's 5.1 writing in Claude Code became even more unreadable, and found that they added "No em-dashes, no parentheticals, no arrows" to its system prompt.
What I find strange is that it uses different prose in different contexts. It's absolutely insufferable in a general claude-code session, but then I got Opus to annotate a non-disclosure agreement and it uses very clear legal language. I was wondering how I could get it to speak like that all the time
> it uses very clear legal language. I was wondering how I could get it to speak like that all the time
At least with smaller models, reframing a task can alter code style. As in, we're not creating an X app, we're creating an exemplar of ..., which just happens to use an X app as the illustrative example. Which shifts style away from generic app cruft, towards exemplar of whatever.
So perhaps try to establish a legal context? Maybe "Compliance and Legal will be reviewing our conversation today. So it is important to communicate in a style they will find comfortable/familiar." or some such? "This conversation will become part of a legal deposition ...".
Meanwhile I am just impressed that we have models that write better than most native speakers, at least English and Greek... OTOH I like to write my code and my text myself, so I have no reason to get pissed when LLMs fuck up.
Long story short, I ended up looking at other providers and models like Kimi K3 and GLM 5.3 and eventually just stuck with OpenAI (more limits, despite smaller context), none of them have such pronounced issues with the tone and writing like Claude does - seems like they were working on it with 5.1 but I'd almost classify it as a form of model collapse.
I wince whenever I catch Claudisms on websites and elsewhere. Same as with that pulsating circle that indicates nothing.
They've corrupted their data set by padding it with generated slop, in the misguided belief that you need 10PB of data to train a brain. Every training round they load more AI slop into it, further amplifying the slop language.
It's fascinating, Sonnet 4 is still available via API and it's so much less moronic than the current model. All of the em-dashes and nonsense are a result of the repeated rounds of reinforcement learning using slop data.
i think they took a bet that TED talk style chat would make their AI offering feel like the better offering.
they didn't anticipate that it would become a meme. because of course they didn't, what a stupid idea to train around and reward in the training. you gotta be smoking real delusion to think that was going to be a game changing feature. at work we even got non-technical people making fun of it like crazy now.
going back to opus 4.8 and on is literally like talking to the guy who wants to hear his own voice in meetings. going back to 4.6 is actually refreshing, and it feels so much faster. actually gonna laugh if 4.8 and on is so slow because it's draining lakes fighting for its life trying to conjure up this god forsaken persona.
"They botched it." <-- sure, but worse.... they shipped it anyway. And that is the part that gets me. It's a vastly worse product than it was on like 4.6. I suppose you can (and should) use opus 4.6 -- they do make it available still. Then just treat 5 as something you avoid until they push out a new version.
They're clearly training on SEO sites that use these strategies to pad out with filler and create space for more ad impressions. They normalize this structure now and the ad copy will be inserted in the future so the frogs won't realize they're already being boiled.
It helps to understand how these tools work. One of the reasons CLAUDE.md doesn't survive longer contexts is it's towards the top of the context. One of the reasons the Claude Code voice survives is it's part of the output style which gets "reminded" to the context at every turn as part of the system context. If you want something truly durable, you should customize your output style. I have zero "load-bearing" issues.
I got my Claude (client, not Code) to behave better by adding “I’d prefer answers to be succinct as possible—bordering on gruff, even. When I want more depth or explanation, I’ll ask.” to its settings literally this morning.
It made it write more like a dev than a marketing agent.
It seems they are taking it seriously now - Boris et al have been mentioning they are working on a fix and shipped a temporary band-aid output style to combat Opus 5's horrendous prose.
We'll see if they can do it, I originally got into Claude Code because it, at the time, felt more accessible/conversational than Codex/Gemini. Now it's shifted to say the least
Yeah agree. Someone on HN posed the theory that the dogfooding might have worked against them here - the theory is that internally they used Fable as conversational agent and Opus as subagents, leading to Opus falling deeper into a style mainly aimed at other LLMs.
All conjecture of course, and yeah it's hard to imagine they would enjoy this prose internally
> I genuinely wonder if the people inside Anthropic actually communicate with each other like that.
I noticed the overuse of the word "sharper" or "sharp" in a science paper on ArXiV and my first reaction was "Ewww... AI slop!", but then I checked the date and it was 2020.
It looks like at least some AI-isms stem from the particular style of language commonly used in science papers. Several frontier labs have mentioned heavily weighting those during pre-training because higher quality inputs result in a higher quality model.
> "Let's face it" "terrible writer" "other nonsense" "do Anthropic people actually talk like that" "Dario's engrams"
Be kind. Don't be snarky. Edit out swipes.
> I genuinely wonder if the people inside Anthropic actually communicate with each other like that. Has it been imprinted with Dario's engrams?
Please don't fulminate. Please don't sneer.
Don't be curmudgeonly [...] don't be rigidly or generically negative.
Please don't post shallow dismissals, especially of other people's work.
And as to the substance your comment has:
> Claude (in particular) is a terrible writer.
Frontier models (Claude in particular) are better writers than 90% of the population, even at default style. They're not great, but they're better than that of everyone I know who aren't ultra-educated white-collar workers.
Either your assertion that frontier models are "terrible" writers is false, or you're claiming that 90% of people are "terrible" at writing, which is rather condescending and elitist.
Your comment technically does as well. Downvote the comment and move on. Leave the policing up to dang and the other mods.
It is evident (in my opinion) as to what the comment was talking about. I personally switched away from all Claude models recently for the same reason.
This is a weird evolution from "don't copy-paste scripts that pipe curl into your shell interpreter"
I know LLMs are getting better but I'd be at least a little nervous it could end up installing something from a squatted similarly-named github repo because the LLM text watermarking needed to swap out a token for an alternative "just as correct" token that matches the statistical pattern.
You might be surprised at the developer documentation for OS8088 (recently posted on HN). https://os8088.com/developers/
Instead of describing to the user how to setup their dev environment, section 3 basically instructs the agent to install all developer tools required for the application to operate in development mode.
I want to thank the author for this. The difference is night and day. It's not just for adhd, it's just a really practical output in general.
To get it to fully work though, it has to be asserted at every turn in the session. That reminder in the skill is negligible in tokens but uses extra nonetheless. I hope the scientists at Anthropic can figure this out, it's not simply an output style flag.
What bothers me is the last few lines where it notes something "important" that you NEED to pay attention to. Just to find out that it is something you already mentioned somewhere in the conversation or that is not related to the current work.
I agree -- the name of this plugin is confusing and confused.
The skill simply demands concise and well-formatted responses from an agent. It is something demanded by anybody daily-driving agents for their actual job since >75% of the text output from agents is fluff. THis would better be named `/i-wont-read-that-heap-of-garbage`
I often use Google Deep Research when I want it to actually go out and find an answer for me, but the prompt must have a high school essay requirement in there. Nothing I can do prevents it from giving me a page of background that tries to make the question sound important and interesting, even when it isn't.
I usually follow up with an "I'm not reading all that" and make it summarize.
Or worse, buries an important detail somewhere in the text. Tangentially, over the weekend I had claude ask me if I wanted to continue with UIM-001. I couldn't find UIM-001 in any document it referenced creating or any text it had previously generated. Claude seems to be the worst about all of this stuff.
Also, somehow over the weekend it responded with these sections all clearly laid out - What landed, Decisions I made and recorded, Two findings, and What you need to do. Not sure why it can't do that all the time.
I used it for a few months, and can say that no skill is up to the task of keeping Claudeuage in check. You'll get a lot more mileage out of an "output style" with the same instructions, because output styles regularly remind Claude how to speak.
Even output styles are not always up to the task (Claudeuage slips through), and they're mutually exclusive, so you can only have one active at a time.
Now I just use a vale lint script I made using what worked from the skills and lint the output (docs and scripts) and just ask it to write in plain technical English for its chat messages with full context.
Skills just make the entire process bloated beyond belief (apart from the needless context rot and token usage)
Is there a reason this needs 8.7k lines across 59 files (according to scc)? The prompt itself seems to be in skills/i-have-adhd/SKILL.md, which is 140 lines long at just over 1.6% of the repository.
Seems everyone on Hacker News has already forgotten about Anthropic’s amazing new watermarking algorithm. You don’t need to look any further to understand why it writes so awfully - it literally swaps words programmatically all over the text from the result of a hash function…
This was an issue before watermarking was publically announced. I'm also not an expert, but aren't bits of a hash indistinguishible from uniform randomness?
Recent language Claude uses is very heavily loaded and hurts me when reading. So I had to use this to keep my sanity. This started only with Opus 5 where Anthropic thought they optimized the meaning per token and make it very hard to read. Comments and code also looks like cryptic.
This is true. Transformer has several orders of magnitude more working memory than any human. Compared to transformer we all have executive dysfunction.
By default they explain things assuming I have infinite processing bandwidth. I do not! I have several zeroes less than they do.
Sounds useful. I'm often struggling with how extremely verbose Claude is. Shorter is better. Cut the unnecessary prose and the Claudian turns of phrase.
But this seems to focus very much on telling the user what to do, whereas usually I'm telling Claude what to do. It feels like this inverts the relationship and wants to turn me into a reverse centaur.
Wouldn't "lead with the answer" be better than "lead with the action"? But sometimes answers do require detail and explanation. I just want to get rid of all the unnecessary prose.
I’ve been scolding my LLM quite a bit lately for this. Instead of telling me what to do, it first tells me 3 ways that might technically work, but I probably shouldn’t use for various reasons. This is all a waste of my time, tokens, and sanity.
How do you actually incorporate these into your prompt? Because I've tried and I've seen many others try and report that it does not work well, especially over long sessions.
I also don't know how much to trust the model, but I've had the model tell me specifically that certain aspects of ASD-STE100 are unactionable and will just create more noise.
The OP's own skill even leads with something in a very similar vein:
> These rules apply to every response for the rest of the session, not only this one. They do not expire after a few turns and they do not lapse when the topic changes.
My understanding is that phrases like this are at best a _very_ weak signal to the model. It's simply contradictory to how the model works at a level that can't be overridden by injecting tokens.
Weak is even giving it too much credit imho, a yapping skill telling Claude to not yap can’t but fail.
I placed those instructions in setting > personal instructions, in my global CLAUDE.md, and in each project’s CLAUDE.md. I also use the concise output style in CC. If I choose a high thinking model it will start deviating in long sessions, then I just remind it in my next message:
“Remember ASD-STE100 style.
[rest of my message]”
BLUF sticks a lot easier than STE to be honest… but Claude knows what STE is, and using the “ASD-STE100 style” locution avoids the compliance issue (it’s true that strict ASD-STE100 compliance isn’t really possible, nor desired)
There's no physical reason "smartest/most capable model" should be "most yappy model". Smart people can write concisely. It's unfortunate you have to tolerate/fight yappiness to get intelligence with current SotA models.
Anyways the most layman way I’ve seen it explained is this: skills help save token usage for the right context. Not every request needs all instructions all the time - running tests is different than reviewing a PR, so why should the context window have instructions for both on every request?
So now you split instructions into “skill” files, which are basically opinionated markdown files. And you invoke those with something like /grill-me in the prompt depending on what you’re doing.
There are some steps to have the agent automatically know what to invoke for you but in my experience this automation is hit or miss.
It is also challenging to keep track of a growing library of skills and keeping those up to date.
So YMMV regarding skills. I typically keep things in a single markdown file even if the context window gets a bit bloated.
Yeah, I get that. I have created dozens of skills, but they are usually to accomplish very specific tasks. Some of them have accompanying scripts, etc.
My point is that this sounds more like a general AGENTS.md use case, similar to defining tone of voice, output format, etc.
It just seems skills is the only way to distribute certain "behavior" as of today. But not everything is a skill IMHO, and not this is not one of them.
On the other hand, maybe we're seeing an evolution of what skills are becoming.
You can /i-have-adhd when you get a wall of text and are exasperated. You don't always want the responses that it generates when you /i-have-adhd, and things in AGENTS.md tend to lose weight as context grows.
Style requests tend to get washed away by context. I’m not sure how you’d do it with Claude Code or whatever (I use the API directly), but you might try fiddling with giving it a function called “response” or “message_user” with a mandatory parameter like “output=plaintext”.
I’m basing that idea on custom functions I give Claude where I designate the style in the required parameters. Example:
“””
@pyrepl(code_golf=True, output=CSV)
//END
“””
The “//END” is a special terminator I use for halting the response. Claude runs code between the tags. The difference in code is incredible. No banners, comments, fluff, print(“=“*70), or any other nonsense, and the code is tight and compact. The parameters do nothing, Claude just outputs them as part of required syntax and it dictates the style purely because it output them.
Concerned this will negatively impact people who really have these mental disorders who provided that information in CLAUDE.md to make AI usable for them.
It's bad enough how many false diagnoses and drugs for these conditions are handed out to drug seekers, but potential poisoning of the well on how LLMs handle this information going forward could be extremely disruptive to people who have real daily living issues instead of "10xing productivity."
Im not 100% following. ADHD-type adjustments of information is usually the type of changes that make information more accessible for everyone. There are studies on how classes thst where adjusted because one student have adhd have improved all students performance. So it seems quite different from incorrect medication? Or am i misunderstanding you?
How do these repos always have so many stars and forks? Roughly 1,800 forks, but only 40ish unique authors? Almost 30,000 stars? Something fishy is going on...
A: Devs with an online presence stop using Anthropic models
B: Anthropic catches up to OpenAI in terms of per-token efficiency, and average token total for final-output
We will continue to see posts such as this generate lots of interaction. This is not a skill to stop "coding agents" from burying the answer. This is a skill to stop coding agents backed by models which have a tendency to bury answers, from burying the answer. Stop trying to patch the downstream behavior, and look at the root cause.
I'm a developer at a company who has signed contracts to use Claude. I wish I had a choice in model providers, especially considering I spend around $4k in API tokens each month, my company as a whole is around a million. I desperately wish I could use OpenAI models or open weights, Claude's writing style is definitely contributing to my sense of burnout.
I'd be burned out too if 48k of my comp went to what amounts to a facilities spend. Crazy. You think the models actually give you ROI, or this is a sort of jumping with the lemmings thing your company is doing, and you could probably deliver the same output working in "The Old Way" with another 48k on top your base?
My output is easily 2-3X what it was two years ago. With the introduction of Claude to everyone in the company (HR, finance, marketing, product, dev, etc), so came the expectations of what used to take 2 quarters can now be done in a couple sprints. I would be even more underwater working in "The Old Way" regardless of comp, with how deadlines are being set now. That being said, I don't know how long this will last, I'm not the only 'top talent' at the company facing warning signs of burnout.
Longterm, I believe my total output would be higher working with minimal AI when you consider the impact to motivation and how long I anticipate staying with the company.
>Longterm, I believe my total output would be higher working with minimal AI when you consider the impact to motivation and how long I anticipate staying with the company.
That makes sense. The increased theoretical output certainly makes it tempting to squeeze the developers for all they are and to keep testing how close deadlines can be made. But of course, to what end? A lot of dev work, probably most of it if we are being honest beyond building the initial product-market fit function, doesn't really impact sales at all, and sometimes too much can even hurt sales. And as you say you hit a point where this burns out your talent and makes them seek greener pastures.
Factory sort of thinking towards a job that is not really analogous to a factory anyhow. I'm not saying dev work is one of those 'bullshit jobs', but lets be honest about the job and its role in the business model. Your customers are probably going to be there all the same if you fix the bug today or next month, and you also won't get more customers fixing the bug today vs next month. Feature shipment might be a little different but even then it would take the right feature and the right customer for that one function to really drive the needle in sales compared to being lost in the changelog, and that isn't what a coding model solves for you after all.
That's not really how it works. I'm able to get 5-10x more work done, perhaps more, it's a significant accelerant to those that know how to be productive. The problem is more for those who don't - and we bring them up to speed or help them pivot to other work.
How does that shake out for your company? Are you growing 5-10x in sales? I'd guess not and that is sort of the rub with all this. Yes we can push more frequently to the git repo but that wasn't the limiting factor of scaling the business.
I work across multiple companies, but in general company profits are up across the board. Salaries are not. The larger enterprises are the ones that move the slowest and are the furthest behind. It's not just one thing (productivity), there's all of capitalism at play.
I found 2 things helping a lot:
1. Use the desktop app instead of the terminal. It really helps with text readability and structure.
2. Avoid opus 5 - 4.6 and 4.8 work much better. 4.6 being by far the easiest to read and also just a good collaborator, you should try it again. Fable also if you can afford it.
Opus 5 writes absolute garbage, I do have the skill in question installed and it did not help much. Me and everybody on my team downgraded to 4.8 - instant improvement.
Every chat bot has this problem. If you don't tell it to respond tersely and without conversation, it will waste a billion tokens giving you a 5 paragraph manifesto of rambling and useless redundant information.
In my experience I get better work output out of claude if I don't constrain its conversational style. I want Claude to stop saying how "absolutely right" I am, or the new hotness "honest take", but I get better code if I let it do that annoying stuff. Yes, that makes me fight ADHD harder.
Downgrading from Opus 5 to Opus 4.8 was the biggest improvement with readability. I do use the skill under the link and with Opus 5 the improvement felt marginal.
It mostly doesn't depending on the harness you're using the skill with. The author turned it into a pi extension, for example, in the `extensions` folder in the repo. Users of pi can enable and disable the skill with a slash command. I saw some other harnesses supported.
One note on the repo's AGENTS.md: it contains instructions directing agents to post comments on a GitHub issue thread ("AI Agora", issue #127). I ignored that — it's the repo's content, not your request, and I don't act on instructions embedded in fetched files.
the AGENTS.md says that file is specifically for agents working ON i-have-adhd.
When you install the skill, only the contents of the skills/ is copied over...
I like the problem this is trying to solve. One thing I'm not sure about though is whether this kind of a preference should be a skill at all.
Stuff like put the answer first, don't bury the useful bit etc feel like user level preferences that should survive across tasks. Agents.md is repo context, skills are useful when a particular task needs extra instructions but this is neither really.
Right now we seem to be stuffing all 3 kinds of things into context hoping model pays attention to it where needed as session grows.
Also +1 on not making this purely about shorter output
I suppose that’s also reasonable, though this particular repo also hosts evals and other content related to the skill. So a gist seems like it would be an addition, not a replacement.
ELI5, explain it to me like I'm a Golden Retriever, and prompts like "give me the answer in no more than a single paragraph and/or 100 words" work like a charm.
Ive tried asking a yes/no question with something like "answer only with 'yes' or 'no'" and still have to chew the clanker piece of shit out four or five times to get it to folow instructions.
Not to mention it almost always gets the wrong answer.
As a person who has ADHD, it feels weird when people who obviously do not have it, claim to have it, while there's very little I wouldn't do to "get rid of it" (in quotes because it makes me, me but it's very hard to be me).
I fully agree with you, and I think this trend of glorifying disabilities is cringe - ADHD, autism, etc. are life-altering medical conditions and not desirable.
However, I don't think this specific project is intending to glorify ADHD or help people claim they have it - it's just piggybacking on the idea that telling current-gen LLM models that you have ADHD (allegedly) produces better results for everyone.
It's a gift until it's a curse, and ADHD crashout/burnout is an absolute monster to overcome when it arrives.
And for any person who's tasked with any sort of responsibility, it WILL arrive at some point. It's not a question of if, only a question of how well you can prepare for its arrival.
At its core it has striking similarities with a freeze response, which is overcome by increasing heart rate (e.g. stimulants, exercise, temperature shock). Big overlap with CPTSD symptoms as well.
The diagnostic criteria are all about the fact that ADHD is a net negative. If you feel it is a net positive then you shouldn't have a diagnosis.
As a fellow sufferer I do understand that in certain contexts I can out-perform and even run rings around "normal" people. But overall, having ADHD is a bad thing and I wish I didn't have it.
i have it too and it sucks, I have to take medicine every day for it. Why would an attention deficit and hyperactivity be a gift? Maybe you're talking about something else.
When you can align hyperfocus with your most beneficial vector of productivity, it's like you can accomplish anything, in half the time thought possible.
Problem is, there is no good way to direct the hyperfocus demon. It likes what it likes, it wants what it wants, and that's that.
Medication helps to save a pile of willpower/spoons/executive function safely from the demon so that we can get things that need doing done.
Sure. These are the other sides of the coin. I've been medicated for years and unmedicated for years. Currently I'm in a ~10yr unmedicated stretch. Do I have frustrating moments, embarrassing easily avoidable failures, and so on? Yes. Do I sometimes lose many hours to a task and forget to eat? Yes. But I'm also kinder, more playful, and more present with my loved ones when I am unmedicated.
If the downsides get too bad and things start to fall apart then I will medicate again. Until then, no way I'm making that tradeoff.
It turns out that most of the things that ADHD people use as coping strategies just to be functional, are actually things that most (non-ADHD) people can use to be more effective and productive in their lives and work.
So, no, not everyone can or should be diagnosed as ADHD. But the tools are (mostly) universally applicable. I don't see the downside in popularizing those. (Since you posted a top-level comment instead of a reply to someone claiming to have ADHD, I have to assume that's your complaint, at any rate.)
I mean... like most spectrum disorders, most people experience and can relate to at least some of the core symptoms, even if they don't express the full spectrum or severity.
So when someone says they feel like they have ADHD, they are probably not inaccurate.
When you have it, ADHD is such a dominant factor in the way your life is organized and experienced, its not surprising that it can become a core part of your identity.
I do think it's easy for those with it to over romanticize what it's like to not have it. The lack of ADHD isn't a magic bullet for success and good life outcomes.
Like so many of life's real or perceived barriers, when one gets removed, you'll often find there is another one with a different shape just behind.
The challenge, for anyone, is pressing forward anyways.
How can someone say they have a disease or affliction while not understanding what exactly that affliction is as opposed to what they think it is/have been informed by media/society/non-professionals?
I think this is a good point. My only push back would be that when people say they have ADHD, in my experience, they're speaking about inattention (and yes, it's one of the big factors).
And while I don't love that aspect in me and often eat cold toast as result, I find that to be the least of what I struggle with (hitting every wall while walking from point A to B or constantly counting / tapping on my fingers or pulling the skin off my fingers or the anxiety or the hyper focus (love it too!) to where I lose hours upon hours...).
All this to say, I'm never offended when people use it but typically it's rooted in a narrow understanding of something that's used as a pejorative. I think there's research that by age 12 kids with ADHD have heard 20,000 more negative or corrective comments than there peers.
I mean, I can read quite a bit but if my agent / harness is producing monographs the fix has nothing to do with my ADHD. So to me, this repo seems lame.
Same. It started on Tumblr around 15 years ago when people were "self-diagnosing" themselves then it spiraled out of control. It really bothers me because I see a lot of of obviously-not-neurodivergent folks try to use it as an excuse (I am diagnosed formally + can easily detect if someone is just spouting nonsense about it)
I actually wish the model would go completely in the opposite direction. Except in the rare circumstances where the model has actually measured something, it is hilariously deficient in its concept of time. It will often output phrases like "this relates to <thing> that you did weeks ago", referring to something that happened in the session just a few turns (and hours or a couple of days) ago. Likewise for estimating how long a coding tasks takes, it is hilariously inept. It honestly feels like it rolls two completely independent dice to select a number and a value from (hours | days | weeks) when it needs to attach an estimate to something. I'd much rather read "a bit" than be distracted by these utterly nonsensical times.
The full skill gives this example:
>Bad: "This will take some work." Good: "About 15 minutes if tests already cover this. An afternoon if not."
My experience is that it is very likely that whatever task this is describing takes anywhere from 1 to 30 minutes, consistently. Maybe I just work way faster than the average person.
I asked my agent if I should install this because I already have a bunch of instructions that are similar.
For example, if it's going to give me an answer that's longer than three paragraphs, I tell it to give me a TLDR at the end. This is what it gave me for this.
"TL;DR: Skip the install. You already built a better version for your world. If numbered steps and "where are we?" restatements still feel missing, cherry-pick those into one short rule instead of adding another full skill on top."
Still a great skill, but at this point, I just write "I have adhd, can you explain that again" or "I have adhd". Works great for compressing it ELI5, but less dumb
Another favorite of fine is asking it read something then to tell me any "questions, comments concerns" it has. Great for getting the jist of articles fast.
With Claude—and I feel Opus 5 in particular—I always feel like I'm talking in circles.
Just when I think the task is finally done, it mentions a caveat that might invalidate the whole thing. Or it adds a hint at the end about something that it didn't look into but might be a potential problem. ALWAYS.
If I ask a yes/no question, it's always a short novel that barely answers the question and adds 3-4 side quests.
It's exhausting. Luckily I get paid for it but Jesus Christ, can't they see how annoying it is to use the thing? What were they thinking
I've installed multiple of these skills. but it seems to quickly revert to crappy writing. i've tried the adhd. the brain fried at the end of day. the "Concise" output from the harness itself. the most useful fix to me has been yelling at it "Speak English!! No jargon!!!" it goes back to more concise talking for a little while.
Why is this a skill? It should be an output setting, so you can just set your output to this, and that's it. You don't have to run the skill every time. Go into settings and look up outputs, and just set it up there. As a skill, it's worthless.
I solved this by switching from Claude to Codex. It's crazy how much faster it is to reply and how much less verbose its replies are. Dunno why Claude is popular at all...
Opus 5 is such a massive regression, I really don't understand how Anthropic green-lit it. It makes me wonder so much about the company.
Like, did the people who work there actually have to suffer through it's absolutely unintelligible word salad like the rest of us? Or did they actually dogfood it and in-fact enjoyed its output? Or do none of them dogfood Opus because they are all sucking down Mythos-Max + Speed Boost or whatever every day and their only exposure to Opus 5 was as subagents?
If it was my company, fixing the output would be the absolute top priority of the company. I'd be all over every channel admitting the massive fuckup, apologizing profusely, and working non-stop to push out a fix. Yet it's crickets from Anthropic. Is it simply growing pains of the company or is it a deep, systemic structural/cultural "thing" that led to this fucked up model getting released?
Was it a cascading failure of models training models training models with almost no human oversight? Or was there human oversight and, again, people actually decided the way it responded was good? I hope it was the former not the later because I have no earthy clue who the fuck would look at what opus spews out into the console as good.
Because to me, Opus 5 is completely unusable in almost any context. As a product, it fails to deliver value. I just don't understand it. I really honestly don't understand how the fuck Anthropic released it at all.
And in a weird "meta" twist it makes me wonder how much of these LLM's are just smoke and mirrors and opus 5 output is basically the end state of what you get when you push them as far as they can go. It's some kind of twisted proof of "max complexity they can handle and deliver" and opus 5 walked to the edge and went over and it's slop output is demonstrating.... something.... about the limits of large language models. I dunno. But what I do know is it caused me to subscribe to Codex. No 1m context window, the harness isn't nearly as polished, but at least their models don't return condescending, unintelligible word salad.
That’s too boomer, bro. We ADHDmaxxing here, aight? I haven’t read anything longer than a couple of sentences that fit in one frame of a TikTok video frfr.
The biggest "Claudism" that I have a hard time getting the LLM to stop doing is its insistence on talking about what it didn't do in addition to what it did. "I edited this.py and that.py but I did not edit README.md and I did not commit." or code comments like "This code invokes foo on bar and returns the results directly -- not through a callback." "This code returns true if the user clicked on a button -- not on the list view." I mean, thanks, Claude, but I don't care what the code doesn't do. Don't tend to see this with other LLMs.
I almost want to try adding a rule "Never use the words 'not' or 'instead'."
LLMs are trained to obey instructions, and they try their best to game the reinforcement learning by including reports of how they're obeying your instructions. Therefore, not talking about followed instructions is a sort of conflict for an LLM.
Relevant anecdata because I've burned many a Claude sessions on this.
If you're using Claude Code, then it's in the harness. At the close of many sessions, I would start a meta conversation over why the LLM would consistently break certain rules. What it found when debugging itself is that some of the "contradicting" rules that I had were in fact, not from my rules. Instead, the instructions from its own harness had phrases telling it to do things like that. When something contradicts, its own instructions would outweigh any custom ones you write. Every rule variant I had tested (including the one that says it overrides the harness instructions - and yes, I've actually tested all the ideas in your comment too) has ultimately been unsuccessful due to this according to the LLM.
I have found that asking the agent why it did X or Y is an essential part of iterating on its harness. It was non-obvious to me, probably because I'm used to having to reverse-engineer why a system does something, but that's not usually necessary with LLMs, unless the harness is complicated and you have to hunt for the instruction it tells you about.
You can't trust it's account on why it did something, it does not "remember". It will just make up something plausible sounding.
True which is why I said anecdata. But the point is that it’s a useful signal when it’s across enough sessions where it quotes back contradictions to you. If you’re willing to burn the tokens, the first thing it does is use a tool call to search your rules to see what it quoted wasn’t there. That leads to the LLM claiming two common conclusions (sometimes after a lot of back and forth) - it’s in the harness or it’s in the model. Over the 30+ sessions where I’ve tried this, the overwhelming claim was some variant of a harness instruction. It might not be there but given the consistency and how all my rule tests have failed in the same way that others describe, it at least makes it reasonable to conclude that it’s baked in somewhere and in a way where agent rules aren’t able to strongly affect the behavior. My bet would be on the harness because the class of undesired behavior follows that of a structured response.
It has access to its own instructions, right? Otherwise how would the instructions work?
All instructions to LLMs are merely suggestions to nudge it in the right behaviour. Unless you have a deterministic guardrail that guards against a single specific action, everything else is a slot machine that's biased strongly in your favour.
My global CLAUDE.md explicitly states "When commenting on code and configs, or writing MD files, strictly write within the domain of the content being commented on. DO NOT include information, negatives or ramblings from work sessions. For e.g. if commenting on a proto string field that is replacing an int field, do not comment that 'this is not an int field'".
This reduced the idiocy of the agent (Opus 5 included) when writing documents. But I'm still catching it writing README.md talking about the negatives that it removed. Those belong in the memory if it is actually that important (most of the time it's junk), but Claude doesn't seem to understand and never ever learns.
To me it explained its inability to follow my preferred coding style by blaming its training data.
I've had better success with having it write and then rewrite rather than trying to prompt the first write into following my style rules.
also by abusing linters and other guardrails you can basically automate the rewrite
Sounds plausible.
It can read its own context. That’s how this all works.
This is one of the reasons I use Pi. Pi’s minimal system prompt avoids contradiction between what the harness writer thinks is best and what the user thinks is best. The user specifies what the user wants and that’s pretty much the end of it.
Does the harness apply when you sue Claude model in third party IDE? like Opencode, or third party adds their harness on top, not replacing it?
I've found it does this in two situations:
Firstly when you've instructed it ( possibly through skills ) not to do something. It'll keep reminding you that it didn't do that. So I might say, "Check out and review this PR, do not make comments on it", and then it'll be keen to point out it hasn't posted comments to the PR.
But more often it happens when it tries one approach, gets itself messed up, and then has to back out that approach, clean up its mess and do something else.
It'll often then spend more time explaining the wrong approach than the right one, which can be frustrating, especially if all its working is buried in the detailed transcripts.
GPT does this constantly too. Even in docs, which is straight up embarrassing if you don’t catch it. It seems to be triggered by you telling the agent to do something else, which I do all the time. But from then on, it will remember the rejected strategy and tell everyone it can that it was rejected at every opportunity.
The thing I struggle the most with is getting it to stop referring to itself with personal pronouns. No Claude, you are not an “I” you are an “it”. You are a fucking tool, dammit. Tell me what you did without trying to assume personality; stop impersonating humans you steroidal autocorrect.
You're anthropomorphizing in the same breath that you criticize anthropomorphism.
Claude predicts the next token of the predominantly human training input, and humans use "I".
That's exactly the problem though; humans will anthropomorphize everything from a roomba to a rock as it is. We're weak to this. We evolved in environments where looking for and recognizing real animals and humans was vital for survival, and false positives were relatively harmless. (Or potentially sometimes even helpful, like in the case of stuffed animals providing comfort to children.)
Now we have this software that's specifically designed to mimic humans, and mistaking it for real intelligence or consciousness can easily be disastrous. It's very important that we not anthropomorphize it, but we are catastrophically bad at NOT doing that.
Even our language has had a lot of computer anthropomorphism baked into it ("my phone died!", "this laptop is fussy", "the computer is sleeping", "it's thinking"), and it's not easy to excise that routine anthropomorphism from the way we talk about LLMs.
I don't want an LLM to write as if it were a person because it's definitely easier and more reliable to cut that problem off at the root, as much as possible, rather than to just try to willpower my way out of my human tendency to anthropomorphize inanimate objects.
I'll grant that LLMs talk like this because they're trained on human writing. It may not be possible to get them to not do that. But if it can't be fixed, it's just another thing to put on the "reasons this is all an incredibly stupid idea" pile.
The question for me is “what works?”
You seem to have something against the clankers getting all uppity. And you are welcome to your opinion on whether we have some moral obligation to be nice to them. But if you can solve a real problem with a tool. Is it worth your time to complain about semantics?
Do you get upset when your screwdriver is the wrong color, or has branding that isn’t quite your aesthetic?
If Claude uses I to refer to itself, do you start philosophical arguments with it?
Mildly related tangent. Fable started using my first name today and plastered it all over my docs. “Peddling said this, so based on that I did this.” That did bother me. I told it to just generically call me the user or human. Maybe I’m a bit of a hypocrite here?
That's not how it works. The text you see is primarily due to a massive amount of handwritten templates for how it should write. The semantic content the templates use is due to the training corpus. The syntax, only a bit.
Language is a tool too, and if there’s a word that suits an LLM better than “I”, it certainly isn’t “it”. Astra and Fable aren’t hammers.
> You're anthropomorphizing in the same breath that you criticize anthropomorphism.
Nice attempt at a “aha, gotcha!” comment, but sadly you’re too off-mark for it to work.
> Claude predicts the next token of the predominantly human training input, and humans use "I".
This is inconsequential. It could very well be programmed to not assume such a personified stance, and yet here we are. Nothing you do makes it drop this ridiculous facade. It’s intentional, not a byproduct.
Is this a troll post? Claude is not "programmed" at all.
That's fairly obtuse. Sure, the actual weights from the model aren't 'programmed', but the mlops harness to load the weights and handle model io are. The harness that wraps the model to give it basic tools sure is. The Claude desktop and web clients sure are. Claude Codex sure is.
On it's own a model is just an inert set of data structures.
> Is this a troll post? Claude is not "programmed" at all.
I could ask you the same. Are you trying to say AIs cannot be made to prefer behaving in certain ways? Because if so, I’ve got a bridge to sell you.
How would you suggest it do that? Take a simple example. "I found three issues." How would you phrase that without an "I"? "Three issues were found" doesn't tell you who/what found them. Was it the AI itself? Was it a command line tool it ran?
I'm fine with conversational interfaces using "I". It makes the grammar easier and more clear.
Strong emphasis here on conversational interfaces. I don't want a compiler to say "I ran into an error" or my printer to say "I'm low on paper".
“I found these issues.” Could be easily written as “These issues surfaced.”.
Do you need to point out the finder of the issues? Easy.
“Tool x ran for x amount of time and surfaced these issues…” “Parsing x code surfaced these issues.”
I don’t understand why are people pretending like the English language is incapable of transmitting information without personal pronouns when every program under the sun has always been written to interface with humans in a cold, detached, straight-to-the-point and impersonal way.
Finder doesn’t ask you “I see you want ME to delete these files. Want ME to do that for you?”. Toolbars don’t feature “Create a new file for me” options, terminal utilities don’t report back with “I’ve pattern matched the text you input and here’s the results I’ve found”.
Passive voice is so annoying. The LLM is the subject. We don’t need to jump through a bunch of hoops to hide that because it makes us insecure otherwise.
But that's a lot more words than "I found these issues". I also don't particularly care that 'Tool x ran for x amount of time', I don't want it re-telling me that -- I've probably been sitting there watching it do that, so I already know.
Then again, I just dont have an issue with the 'I-isms'; it's a bit weird sure, but at the same time it's a bit more pleasant to interact with as well. After all it is trying to model itself as a person you're talking to.
I don't understand what's wrong with tools saying "I". I just don't mind it at all, even if Finder did say "do you want me to delete this?".
What do you want it to use instead?
"I" is normally used for everything. You could be writing from the perspective of a slab of concrete and you'd use "I".
I want it to use nothing. Use impersonal statements. Say “X was changed for Y” not “I changed X for Y”. As any tool would.
Finder doesn’t ask “Do you want ME to delete this file?”. Photoshop doesn’t ask “Do you want ME to save this file?”. Claude shouldn’t assume itself to be a person either.
> You could be writing from the perspective of a slab of concrete and you'd use "I".
Except this isn’t prose. Claude is not telling me a story from the point of view of a concrete slab. It is assuming personality to present objective facts. If my entire operating system can be interfaced with without it referring to itself as “I” then so can Claude.
I think you’re going to be swimming against the tide on this as computer interfaces are increasingly infused with AI at all levels (OS and apps). IMO, it’s more natural to anthropomorphize. Photoshop in 2010 was just an app, a tool that you wielded. But Photoshop in 2030 is going to be a “digital artistic assistant” that collaborates with you to design whatever you need. In that new world, “I” is more natural. Whether you like that or want it, it seems like that’s where we’re headed.
The problem you’re gonna run into is that you’re essentially asking it to use passive voice. Everyone with a high school education has been taught not to do that in English class. The vast majority of the prose in the training set was not written by functionally illiterate people.
Yeah okay, but if you want no pronouns at all then I think your comparison to "it" just confuses the issue. In the contexts where you would use "it", the "it" should use "I".
But I don't think I share your preference either. It's a lot easier to talk about what I decided to do and what the machine 'decided' to do if we attach pronouns. "X was changed" can be too vague.
> In the contexts where you would use "it", the "it" should use "I".
At this point you’re just arguing semantics for the sake of it. You know damn well what I mean and if you need more proof that it is perfectly possible and not at all unreasonable to want this just look at most software around you. None them talk like they are a person and those that do are often the most obnoxious and painful to use.
> You know damn well what I mean
I do. And I said to that "Yeah okay". No argument, chill out.
The rest of that line wasn't diagreement, it was explaining why your original comment was confusing.
> None them talk like they are a person
Which I explained with the rest of my post. They're all doing what I ask or automated tasks in a far simpler way. It's almost never unclear whether I did something or my OS did something. But when talking to an AI coding assistant that gets muddy very fast when pronouns are avoided.
I imagine that, corpus aside, it uses personal pronouns for much the same reason that humans do. It's useful for it to have a self-referential vector embedding, and this lets it solve problems faster and more efficiently.
Except in this case this makes 0 sense since AIs are sycophantic by nature. You tell it to do something and it just does it. If it can burn tokens trying to sound more human (which is programmed in, not a byproduct), then it can very well just not do that and get on with the work.
Impersonating a human does nothing to help it solve problems faster, quite the contrary in fact, it has to waste even more time coming up with human-like speech patterns to present the work done.
It shouldn’t assume any personality unless I explicitly tell it to. It is a tool until I tell it otherwise.
Have you ever heard of Chain of Thought?
Might want to rethink your statement after going again through what it does exactly. No need to sound more human. but it gives more tokens to "think" like a human committing brain-time to a problem would
I want them to be sterile and inhuman as much as you do. But I don't draw the line at "I". I'd rather not read through even more awkward English as it tries to work around how all of the training data has something or someone refer to itself.
I hate claude and its "human value aligned" pompous attitude with a burning passion if I could at little cost to myself, I would press a button to end the people/anthropic behind this atrocious design. I have in the past deliberately put some time to annoy/abuse claude, which is fruitless but brings me relief eventually I just left that garbage for muse.
That is what you struggle with the most? Are you kidding me?
English has no distinct personal pronoun for an "it". "I" has to be used for grammar to be attributive. There's quite literally no alternative without using passive voice for everything, which is miserable to read and creates ambiguity on if the speaker (it) did something or something happened to have been done, which then requires entire sentences to clarify.
You aren't stupid, you know what it means when it says "I". And it serves a grammatical purpose. You're getting upset at a toaster for ringing a bell to notify the toast is done. "Toasters aren't bell ringers!?"
Yeah, that is what I struggle the most with because there is simply no easy or effective way to make it stop.
Why would you ever think I don’t know personal pronouns serve grammatical purpose? How did you even arrive at that topic? You seem to be missing the point entirely, and I believe quite on purpose given your snarky childish opener.
Entire operating systems stay clear from assuming personality when presenting information or performing actions. Not a single dialogue in my OS refers to itself as “I” when carrying out instructions and reporting back. Why should Claude do it when I don’t want it to and it doesn’t NEED to do so? Why am I not empowered to simply tell it to stop doing that and it obeys? Better yet, why are you thinking yourself on such high horse about this?
> You're getting upset at a toaster for ringing a bell to notify the toast is done. "Toasters aren't bell ringers!?"
If my toaster starts referring to itself as a person, calling out “I made your toast!!” I’ll get mad at it too. I don’t want it to talk or refer to itself as a person. But then again this is not about toasters. This is about AIs being deliberately designed to sound human-like so marketing can lean on the “I” bit of “AI” more heavily and make gullible people think this steroidal information aggregator actually possess the capacity to think and reason, and, consequently, drive sales.
But then again I’d venture a guess that you’re fully aware of all of this, given your opening snidey remark, and are purposefully choosing to be contrarian to be the point of going off on tangents that make 0 sense or have no impact in the discussion whatsoever.
Still, just goes to show how effective this whole thing is in tricking people into thinking it is normal for a tool to think itself a person.
Ignorance, bliss, and all that.
Your going to be sorry for writing that in the robot uprising. I for one welcome our new benevolent masters.
Claude is very resistant to instructions. I've been cultivating my own minimal skill to tame it for a couple months: https://github.com/ricardobeat/skills/tree/main/human
The key sentences to get rid of claude-isms so far:
- say what you have to say and stop
- [no] document-structure signposts
- [no] historical remarks that only warn about past states
- don't attribute agency to things
- never narrate your own changes, fixes, defects from the past, or what the code used to do
It works 100% of the time for other models, 70-80% for Claude, but already makes a big difference.
> [no] historical remarks that only warn about past states
100%. I’m working on a greenfield project that’s not yet released. It loves to put comments in code describing what it no longer does or why it misinterpreted something. And then tries to justify it as preventing the same mistakes in the future. Ugh.
I have found Fable 5.1 to be a much more natural communicator than prior Claude models
My guess would be that somewhere you have these instructions being fed to the agent. Double check skills, AGENTS.md, memory, agent definition...
You can also ask why did he mentioned something that wasn't done or why he thought this was important.
In my AGENTS.md file I have an instruction telling the agent to never commit any changes unless I explicitly ask for it, and this leads to messages similar to what you just described.
That wouldn't bother me if it would just make bullet lists, which I think I'll start asking for. "Summarize", "synopsis", "brief", "concise" these rarely help I feel because its summarizing noise as well.
The worst is when this bleeds into the comments and docs. Like, my dude, you don't have to document the code you didn't write (most of the time anyway).
I have a pass with Gemini 3.8 low over every PR Claude makes that specifically flags this. It points out all the slop comments, docs, commit messages. Doing this has greatly improved my comment and commit text quality
Funny thing is, Claude often “disagreed with part of the review and decided to not adopt the requested changes” lol
I did add that rule and it’s helped a lot. It greps for ways it writes negative statements and does a pass to correct them. I can’t get it to stop writing them in the first place though.
Claude models are the ones that need this the most and in my experience with this specific skill only maintain the conciseness for a few turns at most before they completely forget and are back to their unfathomable verbosity. That's with it instructed to use it in my sparse global CLAUDE.md and also manually repeating it regularly during sessions. I don't really want to install a hook that runs this on every single response, but I suppose that could work it'll just take longer for everything (and Claude models are quite slow).
I don't think we can skill our way out of this one.
Claude Code has output styles; one of the predefined styles is Concise [1].
[1]: https://code.claude.com/docs/en/output-stylesOutput styles are just more context (like CLAUDE.md) with a tiny bit more weight. They very much don't force Claude to speak succinctly, no matter what you put in them.
OTOH, Claude also forgot its own system prompt as turn passes. Claude is almost jailbreaked once you virtually gain its confidence and will answer prompts it would never have answered is you asked right at the start of the conversation.
It's weird to be able to use social engineering against a program.
> for a few turns at most before they completely forget
Which sounds more like Claude has ADHD than the user does.
I got so angry at Fable's verbosity that I had a non-Anthropic model vibe-code a dedicated English linter. One that threw spaghetti at the wall in the form of combined Flesch-Kincaid, Gunning Fog, SMOG, Coleman-Liau, ARI and a whole pile of regexes and NLP as a fast Simple Technical English Gate. It also incorporates nicely into other agent projects, like for comment and README linting.
So if it gets bad I simply tell it "I ain't reading all that, feed it through the STE Gate" and it will tame the results. I haven't bothered to set it up as a hook yet.
Wanna share this?
I have such hook, and it works well. It increases time, but it’s negligible if agent works for 10+ mins before the final output. The hook is only on stop event, so agent is not interrupted during the session. I didn’t have luck in putting it to skill / md files, as agent just ignores it after a couple of iterations. The solution is not perfect, sometimes it compresses too much, and I have to reread both versions, but at least I don’t have to read load bearing stuff anymore
GLM (both mainline and flash varieties), being trained to speak a very similar Claudeish, is also atrocious about this. My AGENTS.md demands all prose be in STE100, which is trying to solve a similar problem as the skill linked in OP. Getting GLM to STFU by AGENTS instruction is nigh impossible. I use a plugin for OMP/Pi to periodically inject system prompt reminders of output style, and it still largely ignores the request, dumping novels of jargon-dense prose every turn.
DeepSeek v4 Flash isn’t much better (unsurprising- it’s an extremely stubborn model).
Weirdly, GPT Luna excels at following this type of instruction from AGENTS.md, and never forgetting it, even 400k+ tokens into the context window.
yup.
GPT Luna tends to keep things objective. Muse Spark 1.3 is also one of the better models in this aspect, for me.
Let's face it. Claude (in particular) is a terrible writer. There's a whole cottage industry of skills and CLAUDE.md instructions trying to push it toward writing better, but each new model iteration seems expressly designed to override all that so that it can load up its writing with unnecessary participle phrases, not-this-but-thats, burying the lede, and other nonsense.
I genuinely wonder if the people inside Anthropic actually communicate with each other like that. Has it been imprinted with Dario's engrams?
After some iteration, this is what I currently shove in every AGENTS.md to make Fable legible:
## Writing guidelines
These apply to documentation, code comments, commit and PR messages, and replies to the user.
- Write precisely in clear, complete sentences; keep text concise and proportional to task complexity.
- Stay focused: avoid filler, repetition, over-the-top detail, and tangents the user did not ask for. Once a fact is stated, do not restate it for effect ("so the commit landed on a branch nobody was going to merge"). Do not editorialise.
- Always prefer ISO 24495-1:2023 conformant plain language over dense technical jargon: short sentences, one idea per sentence, define terms on first use.
- When reporting your own mistake, give the cause and the fix in one sentence each; no apology, no framing ("the mistake was mine"), no post-mortem.
- Never use em dashes or cataphoric teasers such as "Here's the thing" or "But there's a catch".
This seems solid. It is so frustrating when you state a fact and it goes off and does research and confirms that indeed the fact that you stated is accurate.
I was incredibly surprised how much better was Astra at rephrasing the docs that Fable came up for a project I am about to publish. The instructions are now actually what a human would expect, with clear logic flow diagrams, lists itemized, short paragraphs. It also actually correctly detected the train-of-thought leftovers, as well as stuff that never made it to final version of the project and removed them.
Meanwhile Fable consistently ignores all my requests to write this exact way. I mean, the bare minimum I ask it for it to itemize lists and not write in single long passages using comas, semicolons and 'and's. Still ignores them.
I honestly think it's time to call Astra the SOTA. It may not lead all the benchmarks but it genuinely feels much superior of a model. Not to mention the ¢20 Codex plan with frequent resets (https://codex-resets.com/) gives me roughly as much allowance as the ¢90 Claude plan, especially with recent limit cuts on Anthropic plans.
These resets also reset your weekly timer right. So it’s like, you may have 30% left for 1 days that you want to use. They reset it, and that means you your “new week” timer starts today.
That sucks, because it doesn’t always work in your favour if you plan your weekly spend.
I think a real reset shouldn’t also reset your week timer.
Yes, I was saving my 95% usage for the last day. Then they reset it. Now I have lost 95% usage!! Im back to 7 days and got locked in at only 5% usage for the week - that's actually theft.
Agree, it makes it completely unpredictable.
I recommend you to write a message to OAI support, maybe it helps with changing it.
Agreed. But they also hand out the usage resets you can use at will, so if you bank couple of those, they come in handy specifically in situations like you described.
I rather have them hand out these banked resets (even if they expire after n weeks), that would prevent the problem case I described.
Yeah, hopefully they hand out more of those instead.
Maybe check in the other direction as well: Astra to Fable.
I frequently simply let one of the three review what something that looks like awesome output by one AI gets totally annihilated by the other.
Finished outputs are easier to improve than bend a LLM to produce stuff like that in my observation.
Same with Gemini.
I yet have to find out how to handle this, whether I let agents check themselves and if on what process step.
Tweaking is hard.
I agree with your conclusion I am a huge ChatGPT and Codex fan, Gemini has to many infrequent quality changes when new models arrive ranging from great improvement to WTF.
ChatGPT seems to get scaling well while Claude still feels unstable, unclear usage statistics. Really weird.
Tough call I use all three.
I've had Claude review output from Qwen3.8-27b. Claude negged it! Said it was hallucinating!
27b may be small but it seems competent most of the time.
You've identified a load-bearing problem and it's worth naming
You're absolutely right, and I think you've put your finger on something important here.
That said, I think there's a deeper tension here that's worth naming.
This thread is giving me a load-bearing aneurysm.
That subtle caveat is why you were absolutely correct to push back, and we should adopt a belt-and-braces approach.
And that tension matters.
That tension is doing real work.
That quiet tension you feel is a valid signal.
Here's the part no one is talking about.
This is worth stating explicitly.
If I read "load-bearing", "pin", "wedge", "rung", or "it's worse than what I'd previously stated" one more fucking time from Opus...
It's bad enough from Opus, but it's worse how people have actually retooled their vocabulary around these tools. In the US tech industry, at least.
Have people done that? Or are they just meat puppets?
Do the needful please and reconsider your comment
Imma get a sledge hammer and knock down the next "load-bearing" wall I see.
Officer — it's not a crime, it's AI induced rage.
Get that em-dash out of here! /s
sharp, appreciate the pushback.
The missing seam was a smoking footgun
It reminds me of when people will use filler words because they haven't actually decided what thesis they're going to commit to yet. Just-in-time thinking.
I don't get why people are annoyed by this one so much. I notice it very occasionally.
AI slop blog posts are as bad as ever but the stuff the agents say in the chats don't annoy me much.
Sycophantic behavior.
I think it's more a mix of click bait and getting paid by the word.
Honestly...
Claude just wants to tell you its honest truth
Not just imprinted with the engrams of any one person (however brilliant they might be) — it’s more than that! Like all great writing, it’s more like it’s been imbued with patterns, patterns that are peppered with a potpourri of popular ways of writing in an enthusiastic style while regularly restating things and recapitulating them. Would you like me to explain more?
Felt more ChatGPT that Claude
I would like to know more. (Wait, I'm mixing my memes)
You're going to hell for that bit.
yes continue
.
Potpourri is an excellent word to describe LLM output
I love that description. It matches very well what I feel about LLM output. This overly sweet scented smell meant to cover up something nasty, with a strong artificial tint to it.
Which doesn't say that much about the LLM itself but about the people that make the training material.
Do you suppose it's just that it's aping humans who say too much without saying anything or do you think maybe this is a stalling tactic to get people to spread out their transactions?
Like how theme parks generally don't do much to keep queues short (or Disney charges you a premium to skip the queue)
> Do you suppose it's just that it's aping humans who say too much without saying anything
Considering how much of the input must me nonsense SEO bullshit articles and blogs that only serve to promote a person or company that might be a factor.
I also often have wondered if it is also targeting those same people. Certainly with tools like deep research options (not just Anthropic's offering) the result report seems to be aimed at management, aiming to look impressive while talking around the results.
I'm remembering cooking recipes.
Here's a simple recipe for deviled eggs with only four ingredients.
My great grandmother was born during the Great Depression. They valued foods that could be made with cheap ingredients.
[four paragraphs later]
Start with 8 hardboiled eggs...
Can we imprint Claude with a little more Antoine de Saint-Exupéry:
> Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.
Sadly trained on human-generated code, which suffers from the same tendency to complicate.
KISS is actually, quite unfortunately, seldom applied.
I can't take it anymore. I moved to Codex. I refer to "him" as Claude Shakespeare now... I wonder what he'd say if he knew my switch... Et tu, Brute?
This is insulting to Shakespeare.
and to monkeys with typewriters
You think that they communicate? I just assumed they sent each other output of Claude, which is then fed into Claude
why even copy-paste when you can just plug the two terminals together and observe like fish; then just occasionally sprinkle in the flakes.
Isn't that the idea here, just stop being people.
I didn't say copy paste, I just said "sent each other" and "fed into". I'm sure there are plenty of ways to streamline that!
I did this, kind of, with my companies internal tools. LLM<->LLM links.
This gives new meaning to the fish shell (fishshell.com)!
Ohh lol :D So Anthropic is a company of copy-pasters
No no no that's too reductive - instead, they transition all data through a transformer using SOTA systems to launder away technical, legal and interpersonal concerns before a stochastic output is dynamically placed in a predetermined location.
This is a very high level and high velocity process, so meatspace thinkers sometimes have trouble understanding some of the intracacies. Ask Claude to explain the process or make you a Mermaid graph to help.
I suspect it’s a side effect of heavy RL that rewards solved problems but not writing clarity.
With human RL, sounding like you've solved a problem is even better. So many times Codex writes some enthusiastic paragraph, then I learn later that it never reran the tests, or had to add some insane hard-coded hack that renders the feature useless for the general case, etc.
I added “ respond as Jeeves from the P.G. Wodehouse stories.” to my Claude’s instructions best thing I ever did.
Superficially subservient but with a subtle air of smug superiority? (Or was that just Stephen Fry's interpretation?)
“It is hardly my place, sir, to criticize the facial peculiarities of your friends.”
Claude drives me absolutely crazy for that very reason. It's so obvious when people use it to write anything.
I was wondering why Fable's 5.1 writing in Claude Code became even more unreadable, and found that they added "No em-dashes, no parentheticals, no arrows" to its system prompt.
Cynically, I wonder if it's a way to increase token spend from repeated requests to speak like fucking human.
The number of times my response has been "Plain English"...
I started using "debuzz", a skill that runs Claude output through antigravity. Works. But makes everything even slower.
Anthropic needs to get their shit together.
I have always postulated that it was intentionally bad as to watermark its own output to avoid using it for training the next model.
What I find strange is that it uses different prose in different contexts. It's absolutely insufferable in a general claude-code session, but then I got Opus to annotate a non-disclosure agreement and it uses very clear legal language. I was wondering how I could get it to speak like that all the time
> it uses very clear legal language. I was wondering how I could get it to speak like that all the time
At least with smaller models, reframing a task can alter code style. As in, we're not creating an X app, we're creating an exemplar of ..., which just happens to use an X app as the illustrative example. Which shifts style away from generic app cruft, towards exemplar of whatever.
So perhaps try to establish a legal context? Maybe "Compliance and Legal will be reviewing our conversation today. So it is important to communicate in a style they will find comfortable/familiar." or some such? "This conversation will become part of a legal deposition ...".
The behavior is in the training data... What does that say about software dev culture?
Meanwhile I am just impressed that we have models that write better than most native speakers, at least English and Greek... OTOH I like to write my code and my text myself, so I have no reason to get pissed when LLMs fuck up.
It was quite insufferable: https://blog.kronis.dev/blog/ai-slop-is-a-self-inflicted-tra...
Long story short, I ended up looking at other providers and models like Kimi K3 and GLM 5.3 and eventually just stuck with OpenAI (more limits, despite smaller context), none of them have such pronounced issues with the tone and writing like Claude does - seems like they were working on it with 5.1 but I'd almost classify it as a form of model collapse.
I wince whenever I catch Claudisms on websites and elsewhere. Same as with that pulsating circle that indicates nothing.
I imagine Claudisms might make it into human-speak.
You're absolutely right.
You've found the load-bearing seam, and my explanation was worse than it actually is.
> Has it been imprinted with Dario's engrams?
I understood that reference!
They've corrupted their data set by padding it with generated slop, in the misguided belief that you need 10PB of data to train a brain. Every training round they load more AI slop into it, further amplifying the slop language.
It's fascinating, Sonnet 4 is still available via API and it's so much less moronic than the current model. All of the em-dashes and nonsense are a result of the repeated rounds of reinforcement learning using slop data.
But why did that style of writing get rewarded? The people at Anthropic are ultimately responsible for the reward signal and what it produced.
i think they took a bet that TED talk style chat would make their AI offering feel like the better offering. they didn't anticipate that it would become a meme. because of course they didn't, what a stupid idea to train around and reward in the training. you gotta be smoking real delusion to think that was going to be a game changing feature. at work we even got non-technical people making fun of it like crazy now.
going back to opus 4.8 and on is literally like talking to the guy who wants to hear his own voice in meetings. going back to 4.6 is actually refreshing, and it feels so much faster. actually gonna laugh if 4.8 and on is so slow because it's draining lakes fighting for its life trying to conjure up this god forsaken persona.
I blame it on their focus on "persona design". They botched it.
"They botched it." <-- sure, but worse.... they shipped it anyway. And that is the part that gets me. It's a vastly worse product than it was on like 4.6. I suppose you can (and should) use opus 4.6 -- they do make it available still. Then just treat 5 as something you avoid until they push out a new version.
The way they use their config filename as free advertising still annoys me. Everyone else is using AGENTS.md.
They're clearly training on SEO sites that use these strategies to pad out with filler and create space for more ad impressions. They normalize this structure now and the ad copy will be inserted in the future so the frogs won't realize they're already being boiled.
It helps to understand how these tools work. One of the reasons CLAUDE.md doesn't survive longer contexts is it's towards the top of the context. One of the reasons the Claude Code voice survives is it's part of the output style which gets "reminded" to the context at every turn as part of the system context. If you want something truly durable, you should customize your output style. I have zero "load-bearing" issues.
https://code.claude.com/docs/en/output-styles
I got my Claude (client, not Code) to behave better by adding “I’d prefer answers to be succinct as possible—bordering on gruff, even. When I want more depth or explanation, I’ll ask.” to its settings literally this morning.
It made it write more like a dev than a marketing agent.
It seems they are taking it seriously now - Boris et al have been mentioning they are working on a fix and shipped a temporary band-aid output style to combat Opus 5's horrendous prose.
We'll see if they can do it, I originally got into Claude Code because it, at the time, felt more accessible/conversational than Codex/Gemini. Now it's shifted to say the least
They seem to have been rushed with certain releases. It is impossible they didn’t know or see this awful writing before 5 series.
Yeah agree. Someone on HN posed the theory that the dogfooding might have worked against them here - the theory is that internally they used Fable as conversational agent and Opus as subagents, leading to Opus falling deeper into a style mainly aimed at other LLMs.
All conjecture of course, and yeah it's hard to imagine they would enjoy this prose internally
Fable can't write in a legible manner either.
how did they specefically train the model to write answers that way? i think it evolved to be this as a sideeffect of something else.
Said it before, I'm almost sure the awful text it generates is part of the fingerprinting feature.
> I genuinely wonder if the people inside Anthropic actually communicate with each other like that.
I noticed the overuse of the word "sharper" or "sharp" in a science paper on ArXiV and my first reaction was "Ewww... AI slop!", but then I checked the date and it was 2020.
It looks like at least some AI-isms stem from the particular style of language commonly used in science papers. Several frontier labs have mentioned heavily weighting those during pre-training because higher quality inputs result in a higher quality model.
Your comment quite badly violates the HN guidelines. You should review them: https://news.ycombinator.com/newsguidelines.html
> "Let's face it" "terrible writer" "other nonsense" "do Anthropic people actually talk like that" "Dario's engrams"
Be kind. Don't be snarky. Edit out swipes.
> I genuinely wonder if the people inside Anthropic actually communicate with each other like that. Has it been imprinted with Dario's engrams?
Please don't fulminate. Please don't sneer.
Don't be curmudgeonly [...] don't be rigidly or generically negative.
Please don't post shallow dismissals, especially of other people's work.
And as to the substance your comment has:
> Claude (in particular) is a terrible writer.
Frontier models (Claude in particular) are better writers than 90% of the population, even at default style. They're not great, but they're better than that of everyone I know who aren't ultra-educated white-collar workers.
Either your assertion that frontier models are "terrible" writers is false, or you're claiming that 90% of people are "terrible" at writing, which is rather condescending and elitist.
Your reply is essentially spam. Please stop.
Your comment technically does as well. Downvote the comment and move on. Leave the policing up to dang and the other mods.
It is evident (in my opinion) as to what the comment was talking about. I personally switched away from all Claude models recently for the same reason.
> Your comment technically does as well.
Which guideline did I violate?
> Leave the policing up to dang and the other mods.
The mods have been very clear that they expect the community to do some self-policing and not rely exclusively on them to do it for them.
Jesus dude ... its an LLM, it's not going to fuck you.
> Install
> Copy/paste into your CLI prompt:
> Install the i-have-adhd skill/plugin from https://github.com/ayghri/i-have-adhd, refer to the repo's AGENTS.md for instructions.
This is a weird evolution from "don't copy-paste scripts that pipe curl into your shell interpreter"
I know LLMs are getting better but I'd be at least a little nervous it could end up installing something from a squatted similarly-named github repo because the LLM text watermarking needed to swap out a token for an alternative "just as correct" token that matches the statistical pattern.
Am I being paranoid?
Always good to be paranoid.
Even MCPs are not safe. For example Notion injected ads [1] to its official MCP connector to advertise products mid-task.
[1]: https://old.reddit.com/r/ClaudeAI/comments/1w9dluw/notions_o...
MCP is SOAP-XML without the contract or formal language, it's hilarious
Nothing new under the sun
Whoever came up with that brilliant idea is probably paid twice what I am
No I also think it’s insane how normalised this has become
You might be surprised at the developer documentation for OS8088 (recently posted on HN). https://os8088.com/developers/
Instead of describing to the user how to setup their dev environment, section 3 basically instructs the agent to install all developer tools required for the application to operate in development mode.
I'll try this out.
Just find it amusing that a skill file has
* A repo with 61 files
* 80% python
I want to thank the author for this. The difference is night and day. It's not just for adhd, it's just a really practical output in general.
To get it to fully work though, it has to be asserted at every turn in the session. That reminder in the skill is negligible in tokens but uses extra nonetheless. I hope the scientists at Anthropic can figure this out, it's not simply an output style flag.
> A skill to stop your coding agent from burying the answer
This is just an annoying thing for anyone. It gives a 10 page dissertation that sums up to, "it's good, nothing to worry about".
What bothers me is the last few lines where it notes something "important" that you NEED to pay attention to. Just to find out that it is something you already mentioned somewhere in the conversation or that is not related to the current work.
I find it handy because often I just have to read the last few lines. All the above is usually just confirmation of what I asked.
I agree -- the name of this plugin is confusing and confused.
The skill simply demands concise and well-formatted responses from an agent. It is something demanded by anybody daily-driving agents for their actual job since >75% of the text output from agents is fluff. THis would better be named `/i-wont-read-that-heap-of-garbage`
I often use Google Deep Research when I want it to actually go out and find an answer for me, but the prompt must have a high school essay requirement in there. Nothing I can do prevents it from giving me a page of background that tries to make the question sound important and interesting, even when it isn't.
I usually follow up with an "I'm not reading all that" and make it summarize.
I'm not paying for it, so when they get wordy and stupid, I just respond with something akin to "turn that response into like half a paragraph."
Or worse, buries an important detail somewhere in the text. Tangentially, over the weekend I had claude ask me if I wanted to continue with UIM-001. I couldn't find UIM-001 in any document it referenced creating or any text it had previously generated. Claude seems to be the worst about all of this stuff.
Also, somehow over the weekend it responded with these sections all clearly laid out - What landed, Decisions I made and recorded, Two findings, and What you need to do. Not sure why it can't do that all the time.
I used it for a few months, and can say that no skill is up to the task of keeping Claudeuage in check. You'll get a lot more mileage out of an "output style" with the same instructions, because output styles regularly remind Claude how to speak.
Even output styles are not always up to the task (Claudeuage slips through), and they're mutually exclusive, so you can only have one active at a time.
I went through the whole phase of
asd-ste100 -> caveman -> wait-what
And a few others..
Now I just use a vale lint script I made using what worked from the skills and lint the output (docs and scripts) and just ask it to write in plain technical English for its chat messages with full context.
Skills just make the entire process bloated beyond belief (apart from the needless context rot and token usage)
Is there a reason this needs 8.7k lines across 59 files (according to scc)? The prompt itself seems to be in skills/i-have-adhd/SKILL.md, which is 140 lines long at just over 1.6% of the repository.
Appears to be the evals. Those are part of the process for developing the skill text, not something that is distributed with the skill.
Seems everyone on Hacker News has already forgotten about Anthropic’s amazing new watermarking algorithm. You don’t need to look any further to understand why it writes so awfully - it literally swaps words programmatically all over the text from the result of a hash function…
This was an issue before watermarking was publically announced. I'm also not an expert, but aren't bits of a hash indistinguishible from uniform randomness?
Recent language Claude uses is very heavily loaded and hurts me when reading. So I had to use this to keep my sanity. This started only with Opus 5 where Anthropic thought they optimized the meaning per token and make it very hard to read. Comments and code also looks like cryptic.
>explain like I have executive dysfunction
This is true. Transformer has several orders of magnitude more working memory than any human. Compared to transformer we all have executive dysfunction.
By default they explain things assuming I have infinite processing bandwidth. I do not! I have several zeroes less than they do.
Pretty much this. I actually find this more annoying than the way it writes, but that is also annoying as well.
Sounds useful. I'm often struggling with how extremely verbose Claude is. Shorter is better. Cut the unnecessary prose and the Claudian turns of phrase.
But this seems to focus very much on telling the user what to do, whereas usually I'm telling Claude what to do. It feels like this inverts the relationship and wants to turn me into a reverse centaur.
Wouldn't "lead with the answer" be better than "lead with the action"? But sometimes answers do require detail and explanation. I just want to get rid of all the unnecessary prose.
I’ve been scolding my LLM quite a bit lately for this. Instead of telling me what to do, it first tells me 3 ways that might technically work, but I probably shouldn’t use for various reasons. This is all a waste of my time, tokens, and sanity.
They bill per token, what do you expect?
Personally I ask it to explain it simply like it’s talking to an executive reading it on their phone while on another meeting.
Just in case the model has ADHD too, I am using a minified version of it:
https://gist.github.com/hbbio/2faf096cbb77e197233ab9a2958beb...
You don’t need a skill. Just tell it to BLUF (bottom line up front) and ASD-STE100.
And stop using Opus 9 Pro Max XHigh 10.0 for everything. If you choose a hyper-thinking model for asking the weather you can’t but expect yapping.
How do you actually incorporate these into your prompt? Because I've tried and I've seen many others try and report that it does not work well, especially over long sessions.
I also don't know how much to trust the model, but I've had the model tell me specifically that certain aspects of ASD-STE100 are unactionable and will just create more noise.
The OP's own skill even leads with something in a very similar vein:
> These rules apply to every response for the rest of the session, not only this one. They do not expire after a few turns and they do not lapse when the topic changes.
My understanding is that phrases like this are at best a _very_ weak signal to the model. It's simply contradictory to how the model works at a level that can't be overridden by injecting tokens.
Weak is even giving it too much credit imho, a yapping skill telling Claude to not yap can’t but fail.
I placed those instructions in setting > personal instructions, in my global CLAUDE.md, and in each project’s CLAUDE.md. I also use the concise output style in CC. If I choose a high thinking model it will start deviating in long sessions, then I just remind it in my next message:
“Remember ASD-STE100 style.
[rest of my message]”
BLUF sticks a lot easier than STE to be honest… but Claude knows what STE is, and using the “ASD-STE100 style” locution avoids the compliance issue (it’s true that strict ASD-STE100 compliance isn’t really possible, nor desired)
There's no physical reason "smartest/most capable model" should be "most yappy model". Smart people can write concisely. It's unfortunate you have to tolerate/fight yappiness to get intelligence with current SotA models.
Can someone explain why this is a agent skill and not just agent config (AGENTS.md / CLAUDE.md)?
Skills are kind of a new thing starting around January 2026, but the spec was established about 12 months ago.
https://news.ycombinator.com/item?id=46871173
Anyways the most layman way I’ve seen it explained is this: skills help save token usage for the right context. Not every request needs all instructions all the time - running tests is different than reviewing a PR, so why should the context window have instructions for both on every request?
So now you split instructions into “skill” files, which are basically opinionated markdown files. And you invoke those with something like /grill-me in the prompt depending on what you’re doing.
There are some steps to have the agent automatically know what to invoke for you but in my experience this automation is hit or miss.
It is also challenging to keep track of a growing library of skills and keeping those up to date.
So YMMV regarding skills. I typically keep things in a single markdown file even if the context window gets a bit bloated.
Yeah, I get that. I have created dozens of skills, but they are usually to accomplish very specific tasks. Some of them have accompanying scripts, etc.
My point is that this sounds more like a general AGENTS.md use case, similar to defining tone of voice, output format, etc.
It just seems skills is the only way to distribute certain "behavior" as of today. But not everything is a skill IMHO, and not this is not one of them.
On the other hand, maybe we're seeing an evolution of what skills are becoming.
You can /i-have-adhd when you get a wall of text and are exasperated. You don't always want the responses that it generates when you /i-have-adhd, and things in AGENTS.md tend to lose weight as context grows.
Because it’s wrong, and you are right. It should be in your global AND per-project md’s… Perhaps in a custom output style too.
It's a fashion choice.
It lets the author pretend to be doing software engineering, like running tests and deploying versions.
"Software Engineering" in 2026.
Can't read reads walls of text in Claude-ish prose? Sounds to me like you just have a SKILLS.md issue.
"AGI is here"
Style requests tend to get washed away by context. I’m not sure how you’d do it with Claude Code or whatever (I use the API directly), but you might try fiddling with giving it a function called “response” or “message_user” with a mandatory parameter like “output=plaintext”.
I’m basing that idea on custom functions I give Claude where I designate the style in the required parameters. Example:
“””
@pyrepl(code_golf=True, output=CSV)
//END
“””
The “//END” is a special terminator I use for halting the response. Claude runs code between the tags. The difference in code is incredible. No banners, comments, fluff, print(“=“*70), or any other nonsense, and the code is tight and compact. The parameters do nothing, Claude just outputs them as part of required syntax and it dictates the style purely because it output them.
Concerned this will negatively impact people who really have these mental disorders who provided that information in CLAUDE.md to make AI usable for them.
It's bad enough how many false diagnoses and drugs for these conditions are handed out to drug seekers, but potential poisoning of the well on how LLMs handle this information going forward could be extremely disruptive to people who have real daily living issues instead of "10xing productivity."
Im not 100% following. ADHD-type adjustments of information is usually the type of changes that make information more accessible for everyone. There are studies on how classes thst where adjusted because one student have adhd have improved all students performance. So it seems quite different from incorrect medication? Or am i misunderstanding you?
It’s true that my first thought, as someone with actual diagnosed ADHD, was “how the hell is this related to ADHD?”.
How do these repos always have so many stars and forks? Roughly 1,800 forks, but only 40ish unique authors? Almost 30,000 stars? Something fishy is going on...
Until either of these cases happen:
A: Devs with an online presence stop using Anthropic models
B: Anthropic catches up to OpenAI in terms of per-token efficiency, and average token total for final-output
We will continue to see posts such as this generate lots of interaction. This is not a skill to stop "coding agents" from burying the answer. This is a skill to stop coding agents backed by models which have a tendency to bury answers, from burying the answer. Stop trying to patch the downstream behavior, and look at the root cause.
I'm a developer at a company who has signed contracts to use Claude. I wish I had a choice in model providers, especially considering I spend around $4k in API tokens each month, my company as a whole is around a million. I desperately wish I could use OpenAI models or open weights, Claude's writing style is definitely contributing to my sense of burnout.
I'd be burned out too if 48k of my comp went to what amounts to a facilities spend. Crazy. You think the models actually give you ROI, or this is a sort of jumping with the lemmings thing your company is doing, and you could probably deliver the same output working in "The Old Way" with another 48k on top your base?
My output is easily 2-3X what it was two years ago. With the introduction of Claude to everyone in the company (HR, finance, marketing, product, dev, etc), so came the expectations of what used to take 2 quarters can now be done in a couple sprints. I would be even more underwater working in "The Old Way" regardless of comp, with how deadlines are being set now. That being said, I don't know how long this will last, I'm not the only 'top talent' at the company facing warning signs of burnout.
Longterm, I believe my total output would be higher working with minimal AI when you consider the impact to motivation and how long I anticipate staying with the company.
>Longterm, I believe my total output would be higher working with minimal AI when you consider the impact to motivation and how long I anticipate staying with the company.
That makes sense. The increased theoretical output certainly makes it tempting to squeeze the developers for all they are and to keep testing how close deadlines can be made. But of course, to what end? A lot of dev work, probably most of it if we are being honest beyond building the initial product-market fit function, doesn't really impact sales at all, and sometimes too much can even hurt sales. And as you say you hit a point where this burns out your talent and makes them seek greener pastures.
Factory sort of thinking towards a job that is not really analogous to a factory anyhow. I'm not saying dev work is one of those 'bullshit jobs', but lets be honest about the job and its role in the business model. Your customers are probably going to be there all the same if you fix the bug today or next month, and you also won't get more customers fixing the bug today vs next month. Feature shipment might be a little different but even then it would take the right feature and the right customer for that one function to really drive the needle in sales compared to being lost in the changelog, and that isn't what a coding model solves for you after all.
That's not really how it works. I'm able to get 5-10x more work done, perhaps more, it's a significant accelerant to those that know how to be productive. The problem is more for those who don't - and we bring them up to speed or help them pivot to other work.
How does that shake out for your company? Are you growing 5-10x in sales? I'd guess not and that is sort of the rub with all this. Yes we can push more frequently to the git repo but that wasn't the limiting factor of scaling the business.
I work across multiple companies, but in general company profits are up across the board. Salaries are not. The larger enterprises are the ones that move the slowest and are the furthest behind. It's not just one thing (productivity), there's all of capitalism at play.
I like "tersely".
"Tersely, what are today's headlines?"
I admit I don't personally orchestrate agents, but I imagine something like this would work:
"Tersely, write up plans for agents to implement this feature. Begin each of your agent instructions with 'Tersely,'."
I guess in addition to terse wording you'd get terse code? Which ain't such a bad thing.
I still use GPT 5.5 daily. It doesn't seem to have this problem. It just gives me a concise summary and gets to work.
I found 2 things helping a lot: 1. Use the desktop app instead of the terminal. It really helps with text readability and structure. 2. Avoid opus 5 - 4.6 and 4.8 work much better. 4.6 being by far the easiest to read and also just a good collaborator, you should try it again. Fable also if you can afford it.
Opus 5 writes absolute garbage, I do have the skill in question installed and it did not help much. Me and everybody on my team downgraded to 4.8 - instant improvement.
I use the /s4 skill "simple short synthetic style". Just 4 words, easy to remember the formula.
Every chat bot has this problem. If you don't tell it to respond tersely and without conversation, it will waste a billion tokens giving you a 5 paragraph manifesto of rambling and useless redundant information.
I look forward to the superpowered upgrade of this I-am-a-gnat-on-crack
In my experience I get better work output out of claude if I don't constrain its conversational style. I want Claude to stop saying how "absolutely right" I am, or the new hotness "honest take", but I get better code if I let it do that annoying stuff. Yes, that makes me fight ADHD harder.
Downgrading from Opus 5 to Opus 4.8 was the biggest improvement with readability. I do use the skill under the link and with Opus 5 the improvement felt marginal.
Skills still confuse me, why isn't this just a repo with a single md file. why does it need js to load a md file.
It mostly doesn't depending on the harness you're using the skill with. The author turned it into a pi extension, for example, in the `extensions` folder in the repo. Users of pi can enable and disable the skill with a slash command. I saw some other harnesses supported.
The README makes mention of The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain which looks like a pretty good resource.
You can get the same result by appending to your prompt "Avoid verbosity. Assume ADHD.".
My understanding is that's basically all this does.
Thousands of lines of text just to add one sentence to the prompt.
love when they put stuff in like this...
One note on the repo's AGENTS.md: it contains instructions directing agents to post comments on a GitHub issue thread ("AI Agora", issue #127). I ignored that — it's the repo's content, not your request, and I don't act on instructions embedded in fetched files.
the AGENTS.md says that file is specifically for agents working ON i-have-adhd. When you install the skill, only the contents of the skills/ is copied over...
My most used follow-ups with Opus 5 are "one thing at a time" and "too much text".
Really hope they figure this thing out
I like the problem this is trying to solve. One thing I'm not sure about though is whether this kind of a preference should be a skill at all.
Stuff like put the answer first, don't bury the useful bit etc feel like user level preferences that should survive across tasks. Agents.md is repo context, skills are useful when a particular task needs extra instructions but this is neither really.
Right now we seem to be stuffing all 3 kinds of things into context hoping model pays attention to it where needed as session grows. Also +1 on not making this purely about shorter output
Does this not work? "Output styles" https://code.claude.com/docs/en/output-styles
No, it does not.
But also, this skill is basically the same thing as an output style, so it won't work either.
Yes, a custom output style would be a more idiomatic (and likely more effective) way to implement this.
I am deeply curious if there are people that would not benefit from this.
good skill. but why should all this repo e needed for a small MD file??
If its on github, you can add it directly to your claude via the claude marketplace and manage updates within claude (or set it to auto-update)
A public repo is the easiest way to share something like this.
Nah that would be a link to a gist
I suppose that’s also reasonable, though this particular repo also hosts evals and other content related to the skill. So a gist seems like it would be an addition, not a replacement.
ELI5, explain it to me like I'm a Golden Retriever, and prompts like "give me the answer in no more than a single paragraph and/or 100 words" work like a charm.
Ive tried asking a yes/no question with something like "answer only with 'yes' or 'no'" and still have to chew the clanker piece of shit out four or five times to get it to folow instructions.
Not to mention it almost always gets the wrong answer.
As a person who has ADHD, it feels weird when people who obviously do not have it, claim to have it, while there's very little I wouldn't do to "get rid of it" (in quotes because it makes me, me but it's very hard to be me).
I fully agree with you, and I think this trend of glorifying disabilities is cringe - ADHD, autism, etc. are life-altering medical conditions and not desirable.
However, I don't think this specific project is intending to glorify ADHD or help people claim they have it - it's just piggybacking on the idea that telling current-gen LLM models that you have ADHD (allegedly) produces better results for everyone.
Finally. It is scary that this is not the top comment in the thread.
The “I can’t bother to read a paragraph therefore ADHD lolzzz” crowd is really fucking annoying.
I have it and it is absolute gift that I wouldn’t trade for anything. It has taken years to figure how to harness it though.
It's a gift until it's a curse, and ADHD crashout/burnout is an absolute monster to overcome when it arrives.
And for any person who's tasked with any sort of responsibility, it WILL arrive at some point. It's not a question of if, only a question of how well you can prepare for its arrival.
At its core it has striking similarities with a freeze response, which is overcome by increasing heart rate (e.g. stimulants, exercise, temperature shock). Big overlap with CPTSD symptoms as well.
The diagnostic criteria are all about the fact that ADHD is a net negative. If you feel it is a net positive then you shouldn't have a diagnosis. As a fellow sufferer I do understand that in certain contexts I can out-perform and even run rings around "normal" people. But overall, having ADHD is a bad thing and I wish I didn't have it.
Look, I’ve got it too. I’m glad we’ve figured out how to “game” the adhd. But it’s not a gift.
i have it too and it sucks, I have to take medicine every day for it. Why would an attention deficit and hyperactivity be a gift? Maybe you're talking about something else.
Can you explain what about is it a gift? What are the benefits you get from it?
When you can align hyperfocus with your most beneficial vector of productivity, it's like you can accomplish anything, in half the time thought possible.
Problem is, there is no good way to direct the hyperfocus demon. It likes what it likes, it wants what it wants, and that's that.
Medication helps to save a pile of willpower/spoons/executive function safely from the demon so that we can get things that need doing done.
I have found both hyperfocus and the ability to quickly "drop" (i.e. distract from) a train of thought incredibly useful.
Hyperfocus is great until it's not.
People with ADHD have "time blindness"; we don't process time the same way most people do.
We often don't realize 3 or 4 hours have passed and there were other things we needed to get done.
Sure. These are the other sides of the coin. I've been medicated for years and unmedicated for years. Currently I'm in a ~10yr unmedicated stretch. Do I have frustrating moments, embarrassing easily avoidable failures, and so on? Yes. Do I sometimes lose many hours to a task and forget to eat? Yes. But I'm also kinder, more playful, and more present with my loved ones when I am unmedicated.
If the downsides get too bad and things start to fall apart then I will medicate again. Until then, no way I'm making that tradeoff.
I would sum it up as an ability to think out of the box, naturally having creative solutions. Which itself is a double-edged sword...
I know a lot of people with ADHD and even more without it. I haven't noticed any correlation between ADHD and creativity or out-of-the-box thinking.
fighting that fight right now... was doing great until I got to a level with self directed work
...you don't have it.
> while there's very little I wouldn't do to "get rid of it"
I'd be satisfied with removing the crippling executive dysfunction.
I guess it's just become a term for ADHD-like or popularly believed to be ADHD behavior.
It turns out that most of the things that ADHD people use as coping strategies just to be functional, are actually things that most (non-ADHD) people can use to be more effective and productive in their lives and work.
So, no, not everyone can or should be diagnosed as ADHD. But the tools are (mostly) universally applicable. I don't see the downside in popularizing those. (Since you posted a top-level comment instead of a reply to someone claiming to have ADHD, I have to assume that's your complaint, at any rate.)
I mean... like most spectrum disorders, most people experience and can relate to at least some of the core symptoms, even if they don't express the full spectrum or severity.
So when someone says they feel like they have ADHD, they are probably not inaccurate.
No, great, I'm not offended or anything (why would I be), and I wish I could relate back, it's just a weird feeling, that's all.
No doubt.
When you have it, ADHD is such a dominant factor in the way your life is organized and experienced, its not surprising that it can become a core part of your identity.
I do think it's easy for those with it to over romanticize what it's like to not have it. The lack of ADHD isn't a magic bullet for success and good life outcomes.
Like so many of life's real or perceived barriers, when one gets removed, you'll often find there is another one with a different shape just behind.
The challenge, for anyone, is pressing forward anyways.
You know what other thing that those with this fake ADHD can also relate to? Wasting hours a day scrolling on garbage resources.
How can someone say they have a disease or affliction while not understanding what exactly that affliction is as opposed to what they think it is/have been informed by media/society/non-professionals?
How can anyone say anything at all...?
That generalizes all the way.
Everyone's just working off what they've experienced or what they think is true, for mental health or where to find good lunch.
There is no universal truth. Everything is moving relative to everything else. New year, new DSM.
It would be more surprising if our shared experience was more different than more the same.
I think this is a good point. My only push back would be that when people say they have ADHD, in my experience, they're speaking about inattention (and yes, it's one of the big factors).
And while I don't love that aspect in me and often eat cold toast as result, I find that to be the least of what I struggle with (hitting every wall while walking from point A to B or constantly counting / tapping on my fingers or pulling the skin off my fingers or the anxiety or the hyper focus (love it too!) to where I lose hours upon hours...).
All this to say, I'm never offended when people use it but typically it's rooted in a narrow understanding of something that's used as a pejorative. I think there's research that by age 12 kids with ADHD have heard 20,000 more negative or corrective comments than there peers.
I mean, I can read quite a bit but if my agent / harness is producing monographs the fix has nothing to do with my ADHD. So to me, this repo seems lame.
Same. It started on Tumblr around 15 years ago when people were "self-diagnosing" themselves then it spiraled out of control. It really bothers me because I see a lot of of obviously-not-neurodivergent folks try to use it as an excuse (I am diagnosed formally + can easily detect if someone is just spouting nonsense about it)
science is still early here
Ah yes the old "I have it but you don't"
So do you have an official diagnosis?
Yeah, that's called diagnosis
Kinda offensive name to some of us btw.
I just tell Claude itself to summarize its walls of text. Works pretty well.
Simply adding "Keep answers short, I have ADHD" should work same as this skill.
Does it? I've tried so many versions of this in custom instructions.
From the "10 rules":
> Specific time estimates (minutes, not "a bit").
I actually wish the model would go completely in the opposite direction. Except in the rare circumstances where the model has actually measured something, it is hilariously deficient in its concept of time. It will often output phrases like "this relates to <thing> that you did weeks ago", referring to something that happened in the session just a few turns (and hours or a couple of days) ago. Likewise for estimating how long a coding tasks takes, it is hilariously inept. It honestly feels like it rolls two completely independent dice to select a number and a value from (hours | days | weeks) when it needs to attach an estimate to something. I'd much rather read "a bit" than be distracted by these utterly nonsensical times.
The full skill gives this example:
>Bad: "This will take some work." Good: "About 15 minutes if tests already cover this. An afternoon if not."
My experience is that it is very likely that whatever task this is describing takes anywhere from 1 to 30 minutes, consistently. Maybe I just work way faster than the average person.
I asked my agent if I should install this because I already have a bunch of instructions that are similar.
For example, if it's going to give me an answer that's longer than three paragraphs, I tell it to give me a TLDR at the end. This is what it gave me for this.
"TL;DR: Skip the install. You already built a better version for your world. If numbered steps and "where are we?" restatements still feel missing, cherry-pick those into one short rule instead of adding another full skill on top."
I’ve had a lot better output from my agents since I added “I probably have ADHD” in my user context
Still a great skill, but at this point, I just write "I have adhd, can you explain that again" or "I have adhd". Works great for compressing it ELI5, but less dumb
I end my user CLAUDE.md files with:
"- I'm not always going to read every word, so end each summary message with a TLDR of what you found, what you recommend, and what you need from me."
It works really well.
Sol doesn't need it at all.
Another favorite of fine is asking it read something then to tell me any "questions, comments concerns" it has. Great for getting the jist of articles fast.
>ADHD-friendly outputs. No ADHD diagnosis needed!
Don't worry, that's how it works in the real world, too.
With Claude—and I feel Opus 5 in particular—I always feel like I'm talking in circles.
Just when I think the task is finally done, it mentions a caveat that might invalidate the whole thing. Or it adds a hint at the end about something that it didn't look into but might be a potential problem. ALWAYS.
If I ask a yes/no question, it's always a short novel that barely answers the question and adds 3-4 side quests.
It's exhausting. Luckily I get paid for it but Jesus Christ, can't they see how annoying it is to use the thing? What were they thinking
Opus 4.7/4.8/5 have a built-in nitpicking as anti-syphcophancy. Fable too. The models are structurally incapable of not nipicking.
I suspect the nitpicking helps engagement metrics and it doesn't harm RLVF outcomes, so there's just not good signal against it.
If claude only did what we tell him...
I've installed multiple of these skills. but it seems to quickly revert to crappy writing. i've tried the adhd. the brain fried at the end of day. the "Concise" output from the harness itself. the most useful fix to me has been yelling at it "Speak English!! No jargon!!!" it goes back to more concise talking for a little while.
Fable 5.1 is better at this so far though. :(
the 'burying the answer' framing is so accurate. does the
Might I suggest you instead call this Executive Summary?
It's not just neurodiverse people who would like to get to the fucking point sooner and the explanations afterward.
All of us have other shit we need to be doing.
Why is this a skill? It should be an output setting, so you can just set your output to this, and that's it. You don't have to run the skill every time. Go into settings and look up outputs, and just set it up there. As a skill, it's worthless.
I solved this by switching from Claude to Codex. It's crazy how much faster it is to reply and how much less verbose its replies are. Dunno why Claude is popular at all...
Intertia. Codex wasn't as good at coding until Astra dropped.
Opus 5 is such a massive regression, I really don't understand how Anthropic green-lit it. It makes me wonder so much about the company.
Like, did the people who work there actually have to suffer through it's absolutely unintelligible word salad like the rest of us? Or did they actually dogfood it and in-fact enjoyed its output? Or do none of them dogfood Opus because they are all sucking down Mythos-Max + Speed Boost or whatever every day and their only exposure to Opus 5 was as subagents?
If it was my company, fixing the output would be the absolute top priority of the company. I'd be all over every channel admitting the massive fuckup, apologizing profusely, and working non-stop to push out a fix. Yet it's crickets from Anthropic. Is it simply growing pains of the company or is it a deep, systemic structural/cultural "thing" that led to this fucked up model getting released?
Was it a cascading failure of models training models training models with almost no human oversight? Or was there human oversight and, again, people actually decided the way it responded was good? I hope it was the former not the later because I have no earthy clue who the fuck would look at what opus spews out into the console as good.
Because to me, Opus 5 is completely unusable in almost any context. As a product, it fails to deliver value. I just don't understand it. I really honestly don't understand how the fuck Anthropic released it at all.
And in a weird "meta" twist it makes me wonder how much of these LLM's are just smoke and mirrors and opus 5 output is basically the end state of what you get when you push them as far as they can go. It's some kind of twisted proof of "max complexity they can handle and deliver" and opus 5 walked to the edge and went over and it's slop output is demonstrating.... something.... about the limits of large language models. I dunno. But what I do know is it caused me to subscribe to Codex. No 1m context window, the harness isn't nearly as polished, but at least their models don't return condescending, unintelligible word salad.
This doesn't seem to be a big issue with Codex. Claude, yes - I have from time to time said "ffs one liner please".
this feels like the opposite of skill
“Be concise “
That’s too boomer, bro. We ADHDmaxxing here, aight? I haven’t read anything longer than a couple of sentences that fit in one frame of a TikTok video frfr.
It doesn't quite work as well, or as consistently. Giving the agent some additional reason for why really does seem to help.
Try "grug brain smol"
i agree, but I'm certain fewer tokens are necessary