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