Agreed. After an implementation session, I often use:

"did you notice any inconsistencies or opportunities to clarify/consolidate parameters or functionality while you were working?"

Another strategy that has worked well for my MCP server is noticing a sequence of failed tool-calls, then returning a response that contains a request for the agent to provide feedback via a skill.

I’d love to give NixOS a try on an old ThinkPad. I’m pretty comfortable with Claude; is it as simple as ask and go, or do you have any suggestions for getting started?

That's exactly what I did, an old ThinkPad t470s.

I installed NixOS, and then I copied /etc/nixos down to my main workstation where I had claude code working, and used it to modify the configs. I'd then rsync them to the other machine and run the "nixos-rebuild" there.

I don't recall exactly how I did the initial copies, because SSH wasn't set up there. Probably would be best to just ask claude what needs to be done to enable SSH, it's only like a 1 line change.

Some of the first things I asked it to do were:

    - Switch to flakes and tracking the latest software.
    - Switch to using Sway for window manager.
    - Various user account setup things: Configure zsh with atuin and zoxide, enable sudo, allow my SSH key for login, set up some aliases and the software I need.
    - Install claude code (I gave it the "installing" URL) once I got tired of copying back and forth.
    - Set up Home Manager and SOPS and put my secrets in SOPS, set up a variety of my scripts and symlinks.
    - Configure vim to my liking (done entirely in English prose).

Those are the top things that come to mind. Honestly, it went pretty smoothly. The only real issue I had to speak of was that I'm running bleeding edge, and the moment I decided to unplug my desktop monitors from the Ubuntu machine and connect them to NixOS I had it do a package update at the same time. This hit a Gnome+Wayland bug which left me at a black screen. A nice thing about NixOS is you can just reboot and at the grub menu select an older system config, so I did that. Claude tracked down an open bug about it, tried a few mitigations, but eventually I told it to just revert the version of gnome until it was fixed upstream.

It has gone extremely smoothly. Zero regrets. My manual attempts at NixOS always ran me into dark corners that I didn't know how to do (like installing gitbutler-cli, which is not packaged for NixOS yet), and Claude had no trouble figuring out.

One addition: Flakes is something that I think Claude recommended at the point I switched over to home manager+SOPS. Flake, Home Manager, and SOPS were all things I was somewhat aware of from my previous attempts at NixOS. If you are coming in brand new, I'd probably recommend asking Claude to start with flakes sooner rather than later. Just specifically ask it to use flakes.

Pretty much everything else I did without asking Claude much in the way of NixOS specifics. I just described what I want in general terms.

Strongly agreed. He played BlizzCon a few years back and it was my favorite part of the weekend.

Is this HN user runako’s comment[1] from 2 days ago turned into an article?

I guess it’s very possible multiple people are coming up with the same idea at the same time but given this was submitted by the author it seems kinda rude not to mention it.

[1] https://news.ycombinator.com/threads?id=runako#48426082

The SpaceX S1 writes this out, all it takes is a blogger to want to write it out further. Who knows if they even saw Runako's post?

At least some, but let me give an example.

Request: “manual step X should not be part of the automated build script”

Fulfilled as: build script is now split in two. X is still done as a manual step in between. Rather than prompting and waiting for it to be done, the documentation and scripts no longer mention X.

Part poorly written requirements, part implementing under pressure, and part lack of engineering discipline.

The main issue is catching stuff like this early enough to course-correct. Differences in time zone, language and cultural norms can make that a challenge, all of which LLMs have the advantage in.

You have said this in a few places throughout the thread. At this point, citation is needed.

I work for a real business and switched from API billing to max+overflow. It saves money. It’s crazy not to. What are you talking about?

Not the op, but it’s fairly easy to hit if you automate a kanban and have some stuff you want to get done. All those little “wouldn’t it be great if” tasks that show up after doing a big task become very doable, it just soaks your tokens.

This is exactly why my preferred method at the moment is simple markdown files with instructions. At worst, a human could do it.

Yes, but at this point I'd almost rather have my CC info exposed than my personal info. There is law and consequence for fraudulent charges that protects me from loss (if not inconvenience) but there is basically no protection for playing fast and loose with my PII--in fact, it's the opposite! They sell it!

I haven’t thought about it this way but yeah I totally agree. Most of the major consequences from my CC info getting leaked can be dealt with without major long term impact to my life. The same cannot be said about PII currently

What kind of PII are you thinking about?

privacy.com let's you generate per-merchant credit card numbers and put limits on them, limiting the damage a leaked CC number can do.

karlgluck.com