Add the ability to reopen chat stream after successful response

[robconery]

I have been working with @isidorn on a chat extension and I'm finding that I need to re-open the chat stream so I can update the user on the results of their message.

For instance: the user asks Copilot for help on a topic. I'm augmenting the prompt and sending. When I receive the reply, I'm extracting code that I then execute, but only after the user clicks a button that I output to the chat window, which explicitly says "Run This". I want to report back the results of execution, but currently the only way I can do that is via popups, which is out of the flow of the chat.

Ideally, I could re-open the chat stream so I can send the results and also add a stream.progress when the execution goes off (or show an error if there was one). The follow up provider doesn't quite fit the scenario as the process is a bit more interactive.

[isidorn]

So the ask is to be able to report progress in the chat even when not in the process of responding to the prompt. Do we have some alternative suggestion how extension authors can achieve this?

fyi @roblourens

[jrieken]

Understood and supportive. I believe re-opening isn't the best way to achieve this because it being closed marks the end of this reply. What we have been considering is some kind of "continuation button" that doesn't run a vscode.Command-command but appends and submits another message to the current chat. That would allow a participant to reply to "itself"

[roblourens]

Agreed, and we'd like to have real API/UX for this but you could get a similar effect today by calling back to yourself with executeCommand('workbench.action.chat.open', { query: ... })

[cjlapao]

I would also like something similar to this. In our example, we might have a long-running execution, or we might need to ask if the user wants to execute something else in the same context. Currently, we are achieving the same thing with dialogs being sent to the user, but that seems to disrupt the flow of the chat.

[axefrog]

Just want to voice my support for this feature request. I'm designing an extension utilising AI agents that remain active even after the chat stream has been closed. Just like a chat with a coworker, I'd like the AI to be able to proactively add new messages to the chat in order to keep me apprised of things it wants to communicate to me. Currently these need to be sent using other modalities such as notifications, which splits the context of communication. Proactive messages could of course go in a custom view docked in the sidebar underneath the chat, but this needs to be implemented manually and adds code and UX overhead that could be avoided if the agent could simply participate in the chat proactively. If spammy activity is a concern, this could be resolved via a participant-specific permission, right?
@jrieken @isidorn