Consider making search invocations case-insensitive

[ntoxeg]

What feature would you like to see?

I've noticed that searches that code performs (at least when using GPT-5) prefer to use ripgrep and they are case-sensitive, which sometimes creates problems, especially when looking through documentation as the agent will usually search for something like test and the search won't find instances of Test.

Either the agent should do at least two searches in this case or just use case-insensitive search; the latter solution is simpler and guarantees no false negatives, at the cost of lower search specificity.

[zemaj]

Thanks for raising this. At the moment we’re leaving ripgrep’s default behavior in place so the model can decide when case sensitivity matters. For many languages the capitalization carries meaning (e.g. exported vs. private symbols), and we’ve seen the agent explicitly reach for rg -i … or rg -S … when it needs a broader match. Flipping every search to be case-insensitive would add a lot of noise in those scenarios, so we’re going to keep the current behavior. If you run into concrete cases where the model fails to add -i when it should, feel free to share the transcript and we can take another look.