Insert Edit into File Hook

[jacobdaddario]

Hi Oli!

Thanks for taking the time to work on this plugin. I've recently put together a really fun agentic flow that I'm quite proud of, but it's run me up against a part of the library that I'd like to see if we could modify.

Basically, I'd like CodeCompanion to always show me a file before writing edits to it. It's doing the driving, so I need to see where it's looking. I've tried very very hard to get it to do this with next_edit_suggestion, but it has stubbornly refused to do so correctly.

As a consequence, I've investigated adding the ability to jump to the chosen file myself. I attempted this two ways.

1. I saw you had events (yay!) which are a perfect place for adding a hook. Unfortunately, I couldn't use those because tool args aren't passed to the tool event data.
2. I come from a Ruby background, so I tried monkeypatching things to no avail.

I have two different propositions for you on this matter.

1. We could add args data to allow other developers to add tool hooks which extend functionality based on the args that the LLM returns.
2. We could add an optional jump to the insert_edits_into_file tool.

The first suggestion should be pretty easily to implement and is more robust. Regardless, let me know what you'd be interested in pursuing.

[olimorris]

Hey @jacobdaddario

Some very well made points. Another event, say CodeCompanionToolWaiting, would probably be a sensible addition and we could add plenty of data that could be leveraged.

As I approach agent mode, it does feel like a gap to not navigate a user to the buffer being modified. It also feels like it should be something CodeCompanion does. It's good to have an event, regardless, but really the plugin should be doing that heavy lifting especially in the insert_edit_into_file tool.

It's been a while since I've touched the UI so I'm not sure how challenging it would be to preserve the chat buffer layout and navigate to the buffer being edited.

Let's use this discussion to brainstorm.

[jacobdaddario]

I'm using suggest_next_edit already in my agentic prompt, which itself uses the jump command. The that tool was wired up allowed for very easy customization on the user's part. I actually overrode the tabnew strategy to instead use a command that navigates to the buffer to the left and uses Explore.

All that said, I think the approach in next edit suggestion is pretty decent and could form the basis for such a tool.