Optimizes Claude Code performance by delegating complex tasks to specialized, context-isolated sub-agents.
This skill provides a comprehensive framework for implementing sub-agent patterns within Claude Code to maintain context hygiene and improve session efficiency. It guides users through creating custom agents, configuring tool permissions to avoid command approval fatigue, and utilizing built-in specialists like Explore and Plan. By isolating verbose operations—such as deployments, audits, or parallel research—into separate context windows, users can prevent token bloat and maintain high-quality reasoning in long-running development sessions.
Key Features
01Configuration of custom agents with specific tools and system prompts
02Expert utilization of built-in Explore, Plan, and General agents
03169 GitHub stars
04Implementation of context hygiene patterns to prevent token bloat
05Strategies for avoiding Bash approval spam via optimized tool selection
06Techniques for resuming sub-agent sessions to maintain state across tasks
Use Cases
01Performing deep codebase research without cluttering the main conversation window
02Delegating verbose CI/CD deployment logs to an isolated sub-agent context
03Creating specialized read-only reviewers for security or style audits
