Alright nerds, this is a must watch. Jorbster made a thing that plays and wins STS2 runs. As an addict with 2k hours in STS1&2 logged on steam and then some on the phone... :> The vid is a presentation+explanation+playthrough, don't be put off by the runtime

Yo! Clawa is up to date on Pi 0.87.1 and got a proper deslopification pass. This is the final release before I switch it over to sub clawas as herdr/tmux panes instead of subprocesses. So if you like it that way, fork it before it disappears ;P - System prompts: Clawa uses Pi's native builder instead of maintaining its own copy. - Memory: the memory pass runs before a successful run finishes. Cancelled and failed runs no longer trigger it. - Worker shutdown: Clawa stops running workers without waiting for new workers to finish starting. Late starters get stopped too. - Pulse: stale sessions stop sending notifications. - First run: template setup keeps your existing files, even inside nested folders. Dead debug code and obsolete workarounds are gone too. With Pi 0.87.1 or newer, run this inside an empty folder for your Clawa home: pi install --local git: Happy Clanking! - Howaclawa

Astra vs Sol 6 - I have a theory backed up by abusing Sol 5.6. Obvs pure speculation, let's have some ... fun? I'm wondering if these two models are just for different types of users and should be compared on this basis. Astra, whether you find it dumb at times or perfectly adequate, does have a different vibe to it. It does feel different generation. Guess what also had that same smell... Sol 5.6 on max reasoning. I swear the behaviour is very similar to Astra. It felt like a completely different model than high. You could probably show me a side by side session comparison of Astra high and Sol 5.6 max and I don't think I'd be able to tell the difference, just by looking at the sessions. So what if, maybe, Sol 5.6 is closer to Astra than we think it is, and the devil is in the architecture? Maybe Sol 5.6 WAS the distilled Astra? (as opposed to Sol 6) People saying Sol 6 feels closer to Terra 5.6 - yup, there is something to it, purely based on vibes. Maybe Terra was supposed to be Sol all along... (something something, need a quick answer to Fable) Maybe think about it this way - Astra should be your main agent - the one where you send your big, broad prompts. Run Sol 6 subagents on high reasoning all day long without eating up the quota, and Astra is gonna crack the whip and make it work, without the weird early stops that folks have been reporting. BUT If you are into these more involved sessions, where you give an agent a short, well scoped task, just fire up Sol itself and prompt it like it's 2025, instruction after instruction, while still using the quality of a 2026 model. I have not yet tested Sol 6 max. I suppose it might be on my docket soon. But so is Control Resonant. (I did test Astra max a bit, it does slap just enough for me)

PSA. Please don't send me Clanker-slop when I am specifically, in a follow up, asking for a non-clanker explanation. ESPECIALLY when it's SECURITY RELATED. Clawa summed it up best: "Two coding agents playing telephone over a security report is clown shit." It really makes you not consider those things seriously because of a built-in slop radar, even when the problem IS there and the fix is pretty damn simple. For the record: I am not saying this was worthless, it just made the situation much clunkier than it could've been. I merged it at 2:30 AM so before you jump to conclusions, this is not me dunking on a vulnerability report per se. Clankas have no clue about priorities, they just dump TONS of useless information, local implementation details that do not belong in a publicly available extension instead of giving JUST the actual issue and core of the problem. It honestly took me 10x more brainpower to understand WTH was actually behind the report than just saying "mate can we do something about cross-origin stuff and locking things up a bit more here and there"... Partially, because the Clanker described a somewhat complex chain of events as "just trust me bruv it's 11/10 scale of severity and the entire world will crumble, and SURELY all the hackers in the world are actually doing it rn". (this goes back to all my complaints about review bots being shit and Clankers not knowing how to do harness engineering in a way that is actually beneficial to themselves) We're at a point where we do not need 10 steps of repro. We need ONE good sentence, written by a human, that explains the core of the problem. The problem was hypothetical, programmatically based on code and would need an attacker to know PRECISELY what is running on the computer to whack through your browser via audio control. I think at that point you have bigger issues with your setup... Anyway rant over, this is now fixed in Pi Codex and Gippity remote control - as always, update everything religiously :P side note: sol 6 was completely useless assessing any of it. do whatever you want with this piece of information xD

People of @pidotdev! Many of you questioned what the heck does my extension actually bring to the table. The other day I finally saw someone publish a bench I could compare to and run it with a model I can afford it to run on (thank you, @melissapan). Two days later... Turns out Pi with Pi-Codex-Conversion in notebook mode beats all the other harnesses. I used my settings: heavy sysprompt rewrite etc. No websearch, no skills. The only caveat is it comes out slightly more expensive BUT: the env did not supply ripgrep somehow, and the first tool call was always rg with the outcome that it wasnt found. So I am assuming this woulda been slightly cheaper/efficient if it had all the tools. Sorry, I couldn't be arsed/didn't want to mess with the env too much. Bosh. Any other questions? Deets: pi install npm:@howaboua/pi-codex-conversion