Any particular reason that this formatted with backticks instead of just the repo names?

If i understand this right, this will e expanded to something like this:

You have access to the `sourcegraph/sourcegraph, sourcegraph/cody repositories'`. You are able to answer questions about the `sourcegraph/sourcegraph, sourcegraph/cody repositories'`.

Which feels a bit unnatural to me (may be fine though, the LLMs are smarter than I am :D)

There's also a lot of repetition of the codebase var going on. If this holds e.g. 10 repos, this might contribute a lot to the prompt already. Maybe we can simplify this a bit?

Updated the template based on your suggestions.

I see await here. One issue that we repeateldy run into again is that if we start to await anything in the recipe execution path, it would often cause a delay after a message is submitted and before the loading indicator is displayed in the UI. Is there a way to refactor this to not require async code paths in this branch?

This promise, will ideally be already resolved. We fetch repoIDs, whenever the context is changed and cache it. Because the type is still a promise, we need to await it here as well. But by the time the user enters the message, it should already be resolved. There should be only 1 event loop delay.

I’m not really sure what's gping on here. Is this adding the inferred context? If so, should this all be behidn an if(scope.includeInferredFIie)?

Yeah this can be confusing. I'll add comments in the code as well.

So basically, we allow users to disable inference of the current file and repo right? But what if they use the editor widget? The editor recipes still need the context of the current repo & file.

So here, if we are getting options.scope.editor as an argument of executeRecipe, we make sure the repo & file is included irrespective of includeInferredRepo/File.

Added comment

Should this always be created or only if the unified scope is set?

This is only for web and it uses unified scope. If other clients use this is future we can make it dynamic.

I’m a bit out of sync on the latest architecture but is this file used only for web or also for vs code?

only for web.

The indentation in this file is a mix of tabs and spaces

I’m a bit confused by this hook. Does it mean if scope changes, we setScope? 🤔

We are checking for scopeInitialized so it will only happen at load.

typo?

yes

I wonder if this abstraction still makes sense if 80% of the classes that implement this interface have most of the API calls stubbed out 😅

Lol yeah. I think this will improve in-future when we have extensions for other IDEs building in TS. As the code is shared, we need to keep it for now.

Should we instead of constructing a HTML string return JSX here? We can't use .replace, granted, but we can just loop over the regex matches manually and return a fragment maybe?

I’m worried that this code is reused for something else in the future and then the item may contain XSS vulnerabilities

Really good idea. Updated: