matltc

Yep! Only (potential) downside to --bare is it runs in print mode; can't kick off an interactive session with it. The flags I listed above are a close approximation.

Anthropic had something about API-only in the description a couple months back, but they've been waking all those back for months since the fable access rollback fallout

You can do this but you have to pass a million long options like --disable-system-prompt --allowed-tools "" etc

I have some wrappers but the only one I really use is a variant of that pinned to haiku for quick questions.

Totally. Ralph wiggum was added to every big shop's stdlib (harness?) just a couple weeks after it dropped. That simple iterative workflow basically underpins all of the notable/novel achievements of agentic coding or whatever.

I'm sure many more examples in the "official marketplaces" for mcp/skills/what have you

Prompt engineering was deprecated around the time Opus-/Codex-class models dropped imo. Replaced by skills up until about opus 4.5/6 dropped. Pinned model to Opus 4.6 slug in ENV.ANTHROPIC_MODEL, disabled the adaptive thinking thing, and saw my skills start causing degradation

Only ones I use today are for very specific quirks (eg wiredtiger/mongodb 8+ incompatible with ext4/Linux 6.19+ specifically causing segfaults. Have a 20 line mongo skill that says as much. Pinned docker container to mongo 7, can prob delete it now)

I spent a few days reading up on the docs for these things, hook lifecycles, tried writing a few, but they never work as documented, or the documentation changes so frequently that whatever you built is deprecated by the time you get it humming.

Now if I have some non-trivial unit of work, I basically iterate on spec in plan mode then put it on auto and let it rip. Way better results with Fable. jury out on Opus 5, but no regression like 4.7/8

Usually it's just echo "do this lil thing then pr closing issue 123" --model sonnet --effort low. Works well enough, sonnet 5 low is a workhorse and quite resourceful in a good way when things go sideways; doesn't cheat its way out IME

Can you explain how you piped recent agent actions/turns/outputs? I was looking into this yesterday and tried playing around with streaming json, including hooks, but i (really Claude) would need to write a parser (prob with jq) to filter the noise. Was thinking we tail -f that and select whatever we care about. Kind of an observer to catch such spiraling.

One of those programs that shoulda been in coreutils

Thanks for the references, especially FMD tip; have the Google version disabled due to lack of trust.

Wow, straight from the source! Heartfelt thanks for maintaining this project; it becomes more pertinent by the day as we slide ever deeper into living into this dystopian surveillance state.

Thanks for the info! Might take the dive.

Curious to hear: how much effort did it take to migrate to GrapheneOS? I own a pixel, grabbed it in anticipation of unlocking, but the effort it seems it would take from reading GrapheneOS docs has stopped me from committing to it on my daily driver. Would you please provide a quick time estimate and any snags you hit?

I am with you but I inevitably think it's a skill issue. Some devs put a lot of work into the harness. I tried doing this, felt like running in place to me. the lack of RFC-tier standardization and ever-shifting api is demotivating as well since you might have written a harness that will be obsolete with next model gen/version bump

There probably is a correct way to do this, but your average "vibe coder", or even a competent solo dev just yoloing a project, are not implementing such harnesses. In the former case, they may not be able to; in the latter, you're probably better off in 1:1 human to agent workflow due to indeterminate/bad outputs and the amount of back-and-forth in prompts/specs/reviews required to get high-quality work

Then again, I saw Matz harness for his compiled ruby project was just like perms.allow find and grep. No skills, no agents.md or anything like that. Would love to see those prompts! I'm sure the fact that it was Matz had a lot to do with that working, though

Yes, but false negatives fall through. I don't pay for the call screening; maybe a paid version would be more successful.

Mostly ignore unknown numbers, but if I'm expecting a call, I will pick up.

A terse, altered "Hello" is all I say. Sometimes I don't say anything. Most humans would wait a few seconds then prompt with "...Hello?", whereas bots tend to hang up after ~2s silence

> YouTube... Ways around that on Android

Have DoH/DoT enabled as you mention, then open brave, go to YouTube.com

Maybe, but that's a good use of Claude IMO. How else would one learn--by googling? Claude does the annoying aggregation of data and distills into a list.

I can tell you where Claude wouldn't have been much use: if it didn't fire right up due to a short or voltage drop that might occur during reset propagating through circuit. You're on your own at that point and gotta break out the multimeter.

Well idk if the author is actually geohot but if it is then look him up. Famous hacker from cmu ppp, I think he was first to jailbreak iPhone 4

Really? Matt is pushing for it now? Dang. Might try Sorbet out.

What IDE/LSP do you use? I was on VSCode/ruby-lsp and disabled sorbet, but after working with Zod, I became quite intrigued with the value of letting the schema do a lot of the guarding. I was under the impression that things like Crystal (statically typed Ruby) were not in vogue, and that the reason no one was moving toward static typing was because Matz did not give his blessing.

(Just checked sorbet landing page, looks like it's mainly/only for fn signatures?)

Just fyi: you can toggle auto-switching to opus if safeguarded.

My ifconfig is simple: if it's made in Shenzhen, throw it out

Looking to do this to get off stock isp leased router. What's your hardware/distro rec?

I got a degree in philosophy. Couldn't be less interested in this kind of job. I hate philosophy now

One of my biggest regrets is not getting into this stuff when I was in school. Didn't know about tech at all when I was going, just picked whatever was easy to major in and somewhat bearable. Had zero interest in school until later adulthood

I thought about writing a markdown variant that is easier to type--every token in grammar should be easily reachable from home row and should not require more than one keypress (e.g. no #). I figured same conclusion as many in comments here: no need to reinvent the wheel. Plus, the aggregate amount of time it'd save me from typing may not exceed the time it would take to implement.

Like another commenter said below, last Opus version to respect adaptive thinking and token budget flags was 4.6

I locked my default model to opus 4.6 around the time of the nerfs. Such better results compared to 4.7+

That's enshittification for ya I guess

> moving more work to the type system

Sounds right up my alley. Thanks to you (and other siblings) for the thoughtful replies.

I've spent a good amount of time with C, nowhere near mastery though. Is it worth still writing C, or better off just learning Rust if my goal is to write embedded/systems code?

For me, the best way to learn a tool is for a quick example or two showing its utility, then practicing with those, reading the man as needed on specific flags. Google or bot ”how do x" ? Repeat : done

Some pages have a nice up-front synopsis of flags, others put them in a wall of text. Browsing the former can supplant Google, /\b-x while paging is helpful for the latter.

Very refreshing to see a utilitarian series such as this. Disappointing that the latest ai drama gets 20x more discussion and visibility on this site

Some of us are unsubscribing, what with the coming face scans/enshittification/downtime/throttling...

Yeah to those of us who are on the know, but maybe it can be spun as "Claude adoption breaks the internet" to the consumer