Collection of minor bugs and annoyances

[JeanMertz]

Did you encounter a minor bug or annoyance not worthy of its own issue with a full title and description? Feel free to post them here. Don't worry about being wrong or not sharing enough details, any report is relevant and can help us improve JP's UX.

If at all possible, share your JP_DEBUG=1 logs (or the relevant parts) by e.g. posting them on https://gist.github.com.

[rxgrant]

• delete file permission asks for permission to use tool without first displaying the file being deleted, then shows what file was deleted.
• printer confused by showing a preformated list in markdown: LLM typically starts with five fence quotes "`````", but often ends with two lines of three. Some lists show as a wrapped line, but it doesn't seem related to the fence choice. jp c print renders some of these as they were in on-the-fly response, but sometimes a wrapped list is unwrapped by print.

[JeanMertz]

• delete file permission asks for permission to use tool without first displaying the file being deleted, then shows what file was deleted.

This was a regression introduced in 2bb0602 and e8f4a64.

This is resolved in #656.

• printer confused by showing a preformated list in markdown: LLM typically starts with five fence quotes "`````", but often ends with two lines of three.

I've seen this as well, but have yet to capture a good tracing snapshot to properly debug. If you have JP_DEBUG=1 set, do send me the logs of the offending conversation(s) so that I can delve into them.

Some lists show as a wrapped line, but it doesn't seem related to the fence choice. jp c print renders some of these as they were in on-the-fly response, but sometimes a wrapped list is unwrapped by print.

Same for this, I've seen it before, but don't have a snapshot yet.

The tricky thing about the markdown bugs is that the input is (as expected) very non-deterministic and often what I thought was a markdown rendering bug turned out to be the output the LLM provided and there are plenty of cases with markdown where certain not-well-formatted output can be interpreted in multiple ways, and we have to pick one that is "correct" 90% of the time, but might not always work best.

I'll do some debug runs myself to see if I can reproduce, but if you have JP_DEBUG logs, those would be highly useful.

[rxgrant]

Probably want an option to show failed command output and default it to on.