You should have an agent that grades your agents

Try Warp Factories: warp.dev/factories/request-access Let's measure the quality of coding agents (Claude Code, Codex, etc) with scoring agents. These use LLM-as-a-judge to measure output quality along metrics you care about, like code quality, efficiency, verbosity, and task compliance. These feed self-improvement and benchmarking workflows to improve your software factory automatically.