Defending OpenClaw against indirect prompt injection · HackerLangs