Alex

Funnily hype of html instead of md output started 2 days after I started to refactor all my skills to make my understanding of LLM output easier lol, but I really thought that it would be too bold to just ask it to write raw html lol

@zhenyaThinks It should go under Siri folder, so it will be easier to throw away everything alltogether

11GB on my macbook & I can't delete it... Does anybody even use it?

I've just dicovered cmux cli, and it's so much better already then other agent's shells and even subagents. I'm currently evaluating that but omg why didn't I see that sooner

and after few times of that loop now you're running 2 research teams and developing 3 new agents before breakfast because you didn't want to copy 2 jsons manually

You know that Sunday morning, where you want to do something real quick, but don't like that you have to do something manually, open claude to make a small tool for that, then you see something that you can do better in claude setup, you press that cmd+d and ...

Also it's so slow up to a point where I just have to skip some steps I usually include in workflow, because I know that it will be taking 15 minutes to read something from that webpage

I just don't get how people get excited about code and used desktop app for generally agentic tasks, it's always huge downgrade for what it can actually do

I might not get something, but Claude Cowork gives me much worse experience then any harness overall with the same Opus 4.7. Like man, it just feels soo much dumber like I talk to 1 year old model.

@ShieldifyMartin Basically what happened on early days of openclaw, if i'm not mistaken that was one of reasons antropic used to start making guardrails for openclaw requests on API

Unfortunately, everyone is using AI skills blindly. I won't be surprised to see an exploit from a malicious AI skill in the near future.

@zhenyaThinks @GergelyOrosz And block llm providers on wifi firewall in gym

@GergelyOrosz that's the way to go. otherwise I will start a chain of "thinking gyms" in Amsterdam in order for humanity not to lose its ability to... think

There is massive irony in how AI coding tools are starting to become TOO expensive for many enterprises - after eg Anthropic removed subsidizing AI subscriptions. We might go from "everyone use AI for everything!" to "you have $300/month AI budget; use your brain for the rest."

@zhenyaThinks @ThePrimeagen mythos-v-1-2-52.zip

@ThePrimeagen I wonder when they all gonna switch to SemVer and start releasing model bugfixes too. Oh look at us, we just released mythosV1.2.53, one day after mythosV1.2.52, we are so cool!

I cannot wait for gpt 5.5 here in ~30 minutes

Soo when Opus 4.5 could one-shot creating an app, Opus 4.7 can one-shot only my limits for Max plan

@zhenyaThinks Lol I never open cc in terminal big enough to fit that pets feature, is it useful ?

I am boring...

Also I've split app because I wanted to have only very little code outside of container, just as much as I need. Also it have auth and other things, so your Things should be save even if you pass it around multiple Openclaws

Well the thing with Things is that it don't really have API, and it's running only on client, and even sync happens through icloud mostly, so not really easy to connect something there. But still, with few tweaks you could make it work.

If anyone like me, stuck with using Things app and don't really want to switch to something else - there is a mcp connector for that github.com/Oleks-Y/things…

@ihor_katkov Yep, compact never hit me as bad as this context window

1M context window of Opus is pure marketing. Opus 4.6 becomes horribly stupid after 150k tokens used. I

Was today years old when I figured that I can just write to modify permission here. That's much more convenient

@trq212 Will use mostly to get notifications

We just released Claude Code channels, which allows you to control your Claude Code session through select MCPs, starting with Telegram and Discord. Use this to message Claude Code directly from your phone.