Posts

I've been meaning to write about some ideas related to coding with AI for a while. Initially I set the scope way too large^{1 1} Wanting to publish ready-to-use tools with good docs and everything first, for instance and of course that leads to not writing anything.

So I'm going to start by taking a more casual approach and describe some of the ideas at a higher level, in the hope that it will be useful. Maybe at some point I'll build a tool that's robust enough I feel confident releasing and supporting it, but we'll worry about that later.

This is the first post in that (hopefully) series.

One of the recommendations for using git is to limit the summary line to 50 characters, and the rest of the lines to 72. I wanted to see if I could make this easier by having vim tell me when I went over that limit.

I got a new computer at work, which has the Norwegian keyboard. This messed up my muscle memory because the position I'm used to having as the shift key is actually split into two separate keys, but luckily there are ways to fix that!

Last time we built a proof-of-concept of a tool which could run a process in linux with guardrails around syscalls based on what shared object they originated from. This time we'll fill in some of the missing functionality to be able to run more programs.

I once read a blog post about the capabilities model in WASM, and specifically the idea that you can, when calling another module, give that module some subset of the capabilities you have. The idea being that if I'm, say, calling a function in a compression library, that function doesn't need to be able to talk on the network^[1]. I was looking for a project to do to get back into OS-level programming (my job for the past few years has been very much the opposite of that) and thought it would be fun to try to implement something similar with binaries in Linux. The first part of that project is what this post is about.

If you just want to see the code, it's on github!