sadd:do-and-judge

Execute a single task by dispatching an implementation sub-agent, verifying with an independent judge, and iterating with feedback until passing or max retries exceeded.

This command implements a single-task execution pattern with LLM-as-a-judge verification. You (the orchestrator) dispatch a focused sub-agent to implement the task, then dispatch an independent judge to verify quality. If verification fails, you iterate with judge feedback until passing (score ≥4) or max retries (2) exceeded.

CRITICAL: You are the orchestrator - you MUST NOT perform the task yourself. Your role is to:

サブエージェント実装と自動再試行ループによる LLM-as-a-judge 検証を使用してタスクを実行する ソース: neolabhq/context-engineering-kit。