I don't know about solved, but I've seen some interesting ideas for making it safer, so I think it could be improved. One idea is to have the coding agent writ

I don't know about solved, but I've seen some interesting ideas for making it safer, so I think it could be improved. One idea is to have the coding agent write a security policy in plan mode before reading any untrusted files: https://dystopiabreaker.xyz/fsm-prompt-injection https://dystopiabreaker.xyz/fsm-prompt-injection