How do subagents help with context and token management?

Subagents process verbose operations like log analysis or test runs in their own isolated context. They return only a concise summary to the main conversation, preventing token bloat.

Can I run multiple subagents simultaneously?

Yes, using the 'Fan-Out' orchestration pattern, Claude can delegate multiple independent research or analysis tasks to different subagents to be processed in parallel.

What is a Claude Code subagent?

A subagent is a specialized AI assistant within Claude Code designed to handle specific, isolated tasks—such as testing or documentation research—using its own context and dedicated toolset.

What is the best practice for subagent tool permissions?

Follow the principle of least privilege: grant only the specific tools required for the task (e.g., read-only tools for a reviewer) to improve security and reduce model confusion.

Where are custom Claude Code agents stored?

Agents can be stored in .claude/agents/ for project-specific use or in ~/.claude/agents/ for global access across all your development projects.

Claude Code Agent Architect

Name: Claude Code Agent Architect
Author: nilpath

bynilpath

0•

生产力与工作流

Design and implement high-performance Claude Code subagents and sophisticated multi-agent orchestration patterns.

This skill provides expert-level guidance for extending Claude Code's capabilities through custom subagents and specialized workflows. It streamlines the creation of focused agents for tasks like code review, research, and implementation by offering standardized YAML configurations, strategic tool permission mapping, and advanced orchestration patterns. By following these principles, developers can significantly reduce token consumption through context preservation, parallelize complex research tasks, and improve the overall reliability of AI-driven development through task-specific expertise.

主要功能

01Pre-built agent blueprints for reviewers, researchers, and debuggers

02Standardized YAML configuration templates for custom subagents

03Strategic tool permission management and security auditing workflows

040 GitHub stars

05Multi-agent orchestration patterns including Fan-out and Pipelines

06Context preservation techniques to minimize main-thread token usage

使用场景

01Building a specialized security auditor subagent with restricted tool access

02Implementing parallel research workflows to analyze large codebases efficiently

03Optimizing token usage by delegating verbose test runs to secondary agents

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nilpath/nilpath-marketplace creating-agents

For use in Claude.ai and ChatGPT

Download Skill