What is the mandatory self-critique loop?

It is a quality gate where the sub-agent must generate five verification questions, answer them with evidence from the work produced, and fix any gaps before final submission.

What is a sub-agent in Claude Code?

A sub-agent is a focused AI instance dispatched to handle a specific task within its own isolated context, preventing the main conversation from becoming cluttered and maintaining high precision.

How does the skill select the right model?

It uses a logic-based decision tree: Opus is chosen for complex reasoning and architecture, Sonnet for long-form technical writing, and Haiku for simple, repeatable tasks.

Why is context isolation important?

Context isolation ensures the AI focuses only on relevant files and requirements, which reduces hallucinations and prevents the model from being distracted by unrelated history.

Sub-Agent Launcher

Name: Sub-Agent Launcher
Author: NeoLabHQ

byNeoLabHQ

•

542

•

생산성 및 워크플로

Orchestrates specialized sub-agents with intelligent model selection and mandatory self-critique for high-quality task execution.

The Sub-Agent Launcher implements the Supervisor/Orchestrator pattern to help developers manage complex workflows within Claude Code. By dispatching focused sub-agents for specific tasks, it ensures context isolation and prevents 'context pollution' in the main session. The skill automatically analyzes task complexity to select the optimal model—Opus for high-level reasoning, Sonnet for long-form output, or Haiku for speed—and applies specialized personas like Developer, Architect, or Researcher. Every task is governed by a Zero-shot Chain-of-Thought reasoning process and a mandatory self-critique verification loop, ensuring that deliverables are accurate, verified, and complete.

주요 기능

01Specialized agent matching for roles including Developer, Architect, and Researcher

02Mandatory self-critique verification loop with evidence-based quality checks

03542 GitHub stars

04Enforced Zero-shot Chain-of-Thought (CoT) reasoning for systematic problem-solving

05Context isolation to maintain high accuracy and prevent performance degradation

06Intelligent model selection (Opus, Sonnet, Haiku) based on task complexity

사용 사례

01Implementing specific features or API endpoints in isolated, clean contexts

02Designing complex system architectures using high-reasoning sub-agents

03Conducting deep technical research and comparative analysis without session drift

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add neolabhq/context-engineering-kit launch-sub-agent

For use in Claude.ai and ChatGPT

Download Skill