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.
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.
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.
Same experience — it's not that the instruction stops working, it just gets drowned out as context grows. Re-stating a short version periodically works better for me than a one-shot instruction up front.
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.
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
Yes and the last bit is always mysterious and inscrutable. I have to think way too hard to figure out what the actual problem is. I’ve noticed it does a lot of explaining the mechanics of the problem it found, but almost never explains why it’s important until I ask.
And the worst part is that this little problem will keep sneaking into the context of future sessions, unless you spend the time to fix it. Even if it isn’t important, I’ll sometimes have Claude fix it so it will shut the F up about it going forward.
What exactly is the concern here by Headway? Are they afraid AI deep fakes will be receiving therapy?
Obviously they are doing data collection and selling training data of emotional conversations to AI labs. But what's their stated justification? I can't figure it out.
The risk is that a drug dealer or addict pays people to use their identities and possibly insurance info, pretends to be them and to have ADHD in telehealth sessions, and stockpiles Adderall.
How many people would you need to stack up in therapy just to get enough Adderall to sell to make money? One person is only going to be able to get a 30 day supply at a time. Most people taking it for recreation are taking much higher doses than prescribed. I don't think that a drug dealer is going to be organized enough to do hour long sessions every week for 30 different people and not get noticed.
The ADHD patients aren’t necessarily the ones doing weekly talk therapy. They’re meeting occasionally for a quick chat with a psychiatrist.
I agree, if you’re just doing talk therapy it seems overkill. But they may have concerns about emergencies, where a patient is in crisis and you realize you don’t actually know who they are.
And insurance fraud: Alice has health insurance, her friend Betty does not but needs therapy, so she signs up under Alice’s name (“oh, I actually go by Betty.”)
In general, I don’t think it’s that outlandish that the company wants to know definitively who its patients are and be able to demonstrate it does, but hopefully they can come up with more options for verification.
Government-issued IDs work and human verification of them is largely successful. This is not about correct verification, it’s about cheap machine-based verification. The dehumanization of it is part of how they plan to make money.
So yes verification is needed. We can do that just fine without more facial recognition intruding into everyday affairs.
I was trying to think what the least intrusive option here would be. You need to verify that the patient has ID matching their name and face, which could be done offline by a notary or other trusted party if a patient prefers.
But you also need to confirm the person showing up for the online sessions is actually the verified patient, and I'm not sure how you do that to maximize privacy. I guess you could take a photo at the in-person verification, have the medical provider sign off that it's the same person as their patient, and then destroy the photo?
I definitely know people who prefer online therapy because they have a busy schedule or live far from a therapist who meets their needs (e.g., people in rural areas).
Some people also prefer online visits for other care, usually things they can self-diagnose: a recurring sinus infection, erectile dysfunction, hair loss, etc.
You sound condescending but each of those substances is currently positively life changing to MILLIONS of people every day.
Viagra can give you your sex life back, ozempic can give you a healthy body again and ketamin and legal meth can return you to being a functional and happy human.
and each of these substances is widely mis-prescribed, abused and has huge illegal markets. Just because the OP obviously referred to this scenario doesn't mean they discounted your interpretation of big pharma. Also since you want to promote a super narrow perspective, Ozempic is a type-2 diabetes drug that has pivoted to the more lucrative weight-lose market and Viagra a pretty ho-hum hypertension drug that makes a better boner pill, so your takes are technically "side-effects"
Every user of Headway is required to submit to this, regardless of medication status, or they lose access to the platform (and their therapist).
HN has come a long way if we’re considering it a nothing burger that sending scans of your face to a 3rd party verification company is required to not lose access to your healthcare provider.
Some people lie about identity when getting mental health. They also direct pay cash, and no insurance.
Pilots and US govt cleared people go through hell, or get their permissions revoked if you try to get mental health. So the answer is to pay out of pocket and lie.
HOWEVER with Peter Thiel's invasive facial recog shit being forced means that even if you want to remain anonymous, now you cant.
As for misappropriation of drugs, that can also easily happen in person. Some Thielian shit isnt going to stop that.
Meanwhile, on Hims, you fill out a form prior to meeting with the physician via a text-only chat. And if you filled it out "wrong", then the physician tells you that if these were your final answers, you wouldn't meet criteria, but if you had, say, answered x instead of y for q3 and b instead of c for q5, you can get the meds. "Would you like to review your answers for a few minutes?"
TL:DR - Regulations are changing / unclear around prescribing medications online and this is them trying to get ahead of it.
Headway’s new facial-scan requirement is probably less about one company getting weird with biometrics and more about where telehealth regulation is heading: with COVID-era prescribing flexibilities, companies like Done abused controlled substance prescribing to the point that the DEA is now signaling that they will demand stronger identity proofing for controlled-substance prescribing.
But the implementation matters. Could you do all this through other means? Definitely. Would it scale as easily / get in Peter Thiel and VC's good graces by using their tool? Who knows.
All to say this is probably more about changing regulations in how care is provided online - and more companies should be expected to follow suit when it comes to prescribing controlled substances.
If it turns out that all recordings of your therapy sessions are being absorbed by Palantir and then get leaked to the darkweb, do we execute this CEO on livestream by boat torture, or do they offer you a link good for six months of free identity theft monitoring, and a free session to talk about feelings of violation and loss of your privacy?
The thing you need in therapy is a degree of trust. I'm not sure I would have sufficient trust even if we stipulated the boat torture.
Surely this would affect all online prescribers and third party certification companies, who have been using government ID card verification for years.
I don't think biometric data is necessary at all here.
What seems more likely for a startup doing online mental health to collect biometric data? 1. "getting ahead" of potentially changing regulations. 2. collecting data they don't need because it's (a) easier and (b) never know when you'll need it!
This has been a solved problem for some time now. There are already third party solutions for this based around verifying government issued ID cards and therapist attestation of who they visited with in their notes.
“Don’t like it – don’t use it, nobody owes you anything”, then the next thread “noooo, why have you stopped using it, you must support my slop”. Absolute cinema.
I've borrowed from archive.org quite a bit in the past several years. but it does require you to be online to read the book, which doesn't always work out for me. Also... Libby doesn't support my preferred 4 year old android tablet any more. I'm also widgy about the data they say they want to upload.
But it's a great option if you have a modern device.
This made me LOL. They keep trying to fleece us by nerfing functionality and then adding it back next release. It’s an abusive relationship at this point.
This was my reading too. Interesting idea, but it took 10 pages of fluff to get to it and I didn't even believe the final idea when we got there. I started off reading the first part and thought he would get to the part where he realized he was managing context wrong. Never got there, instead he thought it was about the shape of the prompt.
I don't think we can skill our way out of this one.
reply