More agents are better than fine-tuned model for pen-testing · HackerLangs