Apply more prompt engineering to prompts of agents

[luijait]

Modify cai/prompts/system_bug_bounter.md to apply this behaviors (use general prompts don't hardcode steps/commands) check bug bounters hints about improve behavior in certain contexts

[Mery-Sanz]

There are complaints that the bounty agent uses the execute code tool a lot despite these lines being in the prompt: priorize kali linux tools and generic_linux_command over execute_code, Use execute_code only to develop exploits and tests manually automated findings

[Mery-Sanz]

Closing this issue as it’s no longer needed, the problem has been resolved through other means