Models Have Blind Spots: Debugging Unfamiliar Code with a Multi-LLM Loop | Dark Hacker News