Autonomous supervisor → worker → critic loop for Claude Code — delegate one goal, an independent critic grades each round against a rubric, re-dispatch until done. 主Agent监督子Agent+独立评委，定验收标准反复迭代到完美，全程不问你。

Multi-agent content pipeline that loops generator, evaluator, and refiner agents until output passes your voice rules. Built with Claude Agent SDK. Forkable for any voice.

10 hands-on demos covering every major agentic workflow design pattern — built with the OpenAI Agents SDK