Hey, thanks for pointing that out. The difference in how 4.6 vs 4.7 breaks down tasks is a behavior of the models themselves, not the Cursor integration. The workarounds from the post above like explicit instructions to use subagents, and custom subagents usually help bring 4.7 closer to the 4.6 pattern.
If you want to share broader feedback about Opus 4.7, the best place is the main thread, the team reads it: Opus 4.7 - Out Now!. The more specific you can be (task types where 4.6 wins, examples), the more helpful it is.