Fair. I'm comparing it to the output I see generated in more mainstream, day to day, languages. I'm not going to bother to get into an argument on if the frontier models can generate amazing code in general.

What I've found is I can get get the frontier models to generate bash scripts, perl one liners, etc that do exactly what I need at roughly the same quality as any other code it generates.

“Only language” eh? I only have a passing familiarity or very distant memory so could be wrong but I’d say these are the same:

Perl (needs use strict)

Ancient VB/VBA/VB script

Original PHP (no idea about modern)

PowerShell

Old Windows/DOS batch

> Bash scripts requires "set -eu" at minimum, when written by LLM or human.

No. It does not. Just write good scripts and catch errors and handle them. That's the same more less with every language. And bash can be written in a way that it is sane and readable and maintainable. Just because you have seen a lot of junk in the language does not make the language bad per se. Sure there are a lot of languages "features" which are more then questionable but I want to rise again the point that it has its place and can be used in a good way.

$ Hey, claude, I've figured out the powershell one-liner we need and I've pasted it below. Please insert it into the application.

...

I have analysed your idiomatic one-liner and propose adding the following 30 lines into the application Yes/Yes and always accept such idiocy/No

ESCAPE ESCAPE

/clear

I want to update my powershell skill so that it writes idiomatic pipelines and does not introduce idioms from other languages.

[PASTES overeagerness wording from Claude prompting best practices]

It's frustrating when I catch it doing this because it's wasting tokens and time on what it reports as "one-off" scripts. Might add a requirement to never use shell scripting unless the task is truly a one liner.

> It's reached the point of annoyance I might just go back to bash.

You can use zsh as the default shell and still write your scripts for bash. Actually, it’s an advice I saw more than a couple of times. Otherwise you need to translate the bash-isms when you get bits of code from random places on the Internet.

Just put the right shebang (or ask Claude to do it). What’s the problem?

I find that Claude does this much more than other models. GPT, for instance, knows I'm on Windows running PowerShell, and writes very idiomatic PowerShell.

But maybe that's because I'm the sucker, and since PowerShell is more verbose it costs me more tokens than the terseness of Unix shells. Oh well.

Just run your llm in a container with the tools it needs. It makes your pc more secure by removing the risk of doing something destructive to your computer and the llm can do llm. Also, llms ssh'ing to the outside world? Asking for trouble later on