Comment by CJefferson
5 hours ago
Yes, I've found tests are the one thing I need to write. I then also need to be sure to keep 'git diff'ing the tests, to make sure claude doesn't decide to 'fix' the tests when it's code doesn't work.
When I am rigourous about the tests, Claude has done an amazing job implementing some tricky algorithms from some difficult academic papers, saving me time overall, but it does require more babysitting than I would like.
Give claude a separate user, make the tests not writable for it. Generally you should limit claude to only have write access to the specific things it needs to edit, this will save you tokens because it will fail faster when it goes off the rails.
Don't even need a separate user if you're on linux (or wsl), just use the sandbox feature, you can specify allowed directories for read and/or write.
The sandbox is powered by bubblewrap (used by Flatpaks) so I trust it.