Dear Cursor devs,
we need to have at least some shallow idea what kind of agents/actors/roles are working behind the scene. Without this understanding it is hardly possible to write good .cursor/rules/*.mdc rules.
Without this info our .mdc-rules is rather kind of praying for unknown kinds of gods – we don’t even know whether we are in monotheistic or in polytheistic religion with our requests! 
Cursor uses a single agent approach - the rules you write get attached to the main prompt and sent to whichever model you’re using (like claude or gpt4). The AI can pick which rules to use based on your descriptions, but after that the rules just become part of the context that gets sent to the model
So you don’t need to worry about different agents or roles when writing rules - just focus on writing clear descriptions of when/how each rule should be used, and the AI will handle picking the right ones
@danperks do you think it would be a realistic/doable feature request to add a field to the rule .mdc header that advises which model to use for this .mdc rule file?
This has already been requested actually - no eta for now, but I can see how it would be useful to have different models for different rules. For now you’ll need to stick with the model dropdown in settings