How does the judge verification process work?

After a sub-agent completes its assigned step, an independent model reviews the output against the original requirements. If it doesn't meet the quality bar, it provides feedback for a retry.

Which models does this skill use for subtasks?

It dynamically selects between Claude 3 Opus for complex architecture, Sonnet for standard implementation, and Haiku for simple mechanical tasks or documentation.

Is this skill suitable for large refactoring projects?

Yes, it is specifically designed for tasks where changing one component has ripple effects, ensuring that each layer—from the database to the API—is updated and verified in the correct order.

What is the benefit of using sub-agents instead of one prompt?

Using sub-agents prevents context window bloat and reasoning 'hallucinations.' By isolating each task, the agent focuses on specific logic without being distracted by irrelevant file contents.

Sequential Task Orchestrator

Name: Sequential Task Orchestrator
Author: NeoLabHQ

byNeoLabHQ

•

542

•

生産性とワークフロー

Orchestrates complex coding tasks by decomposing them into verified sequential sub-steps using specialized AI agents and model selection.

The Sequential Task Orchestrator implements a sophisticated Supervisor pattern to handle multi-stage development objectives that are too complex for a single prompt. It systematically breaks down tasks into isolated sub-steps, assigns the most appropriate model (Opus, Sonnet, or Haiku) based on complexity, and maintains a clean context window by passing only essential summaries forward. Each step is validated by an independent 'LLM-as-a-judge' with automatic retry logic, ensuring high-quality architectural integrity and preventing the context pollution that often leads to agent failure on large projects.

主な機能

01Isolated sub-agent context management to prevent reasoning degradation

02Automated task decomposition with intelligent dependency mapping

03542 GitHub stars

04Independent LLM-as-a-judge verification for every completed step

05Automatic iteration and feedback loops for failed verification steps

06Dynamic model selection (Opus/Sonnet/Haiku) based on subtask risk and complexity

ユースケース

01Large-scale refactoring of legacy codebases across multiple service layers

02End-to-end feature implementation involving database schemas, logic, and API updates

03Systematic migration of components between frameworks or architectural patterns

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit do-in-steps

For use in Claude.ai and ChatGPT

Download Skill