It would be great to have a “save state” feature. The idea is to few-shot the AI into the exact mindset you need and then capture that snapshot for future sessions.
With the snapshot saved, you avoid the repetitive priming that can chew through up to half of Claude’s context on large projects. This lowers token costs and lets you resume work in seconds instead of re-typing or copy-pasting long context.
How it would work
- Prep the AI into the desired state with one to three question/answer exchanges.
- Snapshot that state.
- Clone the snapshot so every new chat starts from that saved moment.
This maybe something that pairs nicely with Cursor Modes.
NOTE: To all those complaining about AI scope creep - check out RIPER5 Sigma for Cursor - game changer.