First - its an AI development tool and these sorts of “ideal workarounds” should be handled by the bot.
Second - It would be lovely for those paying for an API key from [model] – Cursors load should shed to that key and know that they may be out of context.
It woulld be interesting to take my previous comment about a second session which has a MIRROR of context, but Sandbox (NO YOLO) which can inquire @codebase and submit prompts via personal API - yet output is chat like and must be manually integrated.
Mirroring context to any session for a given codebase is HOly Grail [Shippable_Context, Embedable_Context, etc] as would be [Shippable_Persona and Embedable_Role]