clairekart··on Show HN: 20+ Claude Code agents coordinating on real work (open source)What’s the failure mode you see with single-agent Claude Code on complex tasks? (looping, context drift, plan collapse, tool misuse?)