Remove backticks on replacement

[salah5]

Is it intended behaviour that when using phantom mode for example and selecting the replace or append option, the code that gets inserted still includes markdown syntax such as triple backticks (e.g. python <llm code> ). It doesn't feel really good having to remove the backticks every time when inserting code.

[yaroslavyaroslav]

Fair, it's pretty dumb rn, it's also not that easy to achieve though, since actually phantoms content not proposed to be parsed after have been presented. But I'll look how to improve the UX of replacement/appending after I finish the grand refactoring anytime soon.

[lukebelz]

This and actual word filter, not just prompt engineering to remove the words. With deepseek r1 for example, it will always show the think section, followed by the code, and then more words. So there's still a lot of cleaning up to do. And for whatever reason, it also only shows 1 line for text. It doesn't show everything, so I have to click insert, and then I see everything.

This package is very close to being great. Just needs a push over the finish line. Thanks for building this.

[yaroslavyaroslav]

There's a catch with r1 and all the other reasoning models to come: they uses kinda html tags, which is what Phantoms work with, so after <thinking> arouse it starts to wait the closing tag to render the html. I have still have to think how to work around it within phantom. I even not sure how well it'll present such huge amount of text at once.

However, panel mode works quite reliable with reasoning models, I suggest you to stick with it for some time till it isn't fixed.

And could you elaborate what word filter are you thinking of?

[github-actions]

This issue has been automatically marked as stale due to inactivity. It will be closed if no further activity occurs.

[yaroslavyaroslav]

I believe this was implemented for phantoms recently. So closing it. (Reopen if it's not)