#prompt-engineering clear

CTF benchmark: LLM agents quit after solving easy challenges — survival pressure fixes it