Do In Parallel

This is a parallel task orchestrator that splits work across multiple Claude sub-agents instead of making you wait for sequential execution. You give it a task and a list of files or targets, and it intelligently groups related work, spins up meta-judges to define quality criteria, then launches implementors in parallel with proper verification loops. The grouping analysis is clever: repeatable tasks share one meta-judge spec, interdependent tasks get evaluated together, and truly independent work runs fully isolated. Most useful when you're applying the same refactoring across a dozen files or running analysis on multiple modules. The orchestrator never touches your code directly, it just coordinates the sub-agents and makes sure nothing ships until the LLM-as-a-judge verification passes. Maximum three retries per target before it kicks the decision back to you.