Add deep research to a initialize a project. This would be used to gather data and give a clear start point for vibe coding. This will help the agent get more familiar with the project overall so it can provide better responses. In addition it can be used as a jumpstart for new chats to gain relevance without having to point it to everything.
Hey, thanks for the feature request, we’ll consider it.
Well, this tool will cost a lot of Agent’s token and context window bro. Right @deanrie ?
I think yes, since this requires higher token usage.
Yes but your essential getting full context of project. This can summarize chunks for project appropriately from the start and create clear guidelines. After the research is completed the context can be reduced as all points of interest are now mapped.
I do this in pseudo manner now. I use opus to review a project and setup agent job profiles or whatever you want to call it. Then use multiple chat panel tabs for each profile. The issue being is the agents can interact with one another. it can get messy controlling them with rules and banning an agent from a section of code makes them go rogue and start trying a “simpler approach”.
Im making an esp32 programmer with crew ai with multiple agents in simliar manner and that seems to work really well.
I assume the same idea could be done using a background runner and integrate with cursor, but im not smart enough to figure that out.
The project start would be expensive for sure, but the idea is to break the goal down so cheaper models could be used for the leg work.