Do In Steps

功能

• Sequential sub-agent orchestration
• Automatic task decomposition
• Intelligent model and agent selection
• Parallel execution of meta-judge and implementer agents
• LLM-as-a-judge verification with scoring and iteration
• Context passing between steps

使用场景

• Refactoring large codebases across multiple files
• Implementing complex features with distinct stages
• Automating software development workflows with quality gates
• Handling tasks requiring specialized agents for different sub-problems

非目标

• Directly performing the task implementation itself
• Skipping task decomposition or verification steps
• Overriding the orchestrated sub-agent workflow
• Performing tasks that do not require sequential execution or sub-agent involvement

工作流

1. Analyze and decompose the task into sequential subtasks
2. Select optimal models and agents for each subtask
3. For each step, dispatch meta-judge and implementation agent in parallel
4. Wait for both to complete, then dispatch judge with meta-judge's specification
5. Iterate if judge fails the step (max 3 retries)
6. Collect outputs and pass context forward
7. Report final results

实践

• Orchestration
• Agent Workflow
• Verification
• Task Decomposition