How does this skill handle context window isolation?

It provides patterns for ensuring agents have sufficient domain knowledge to operate independently, as subagents in Claude Code do not share context windows with each other.

Why should I avoid hardcoding workflow phases in agent files?

Hardcoding phases like 'planner' or 'reviewer' limits an agent's reusability. By defining a domain expert instead, the team lead can utilize that same agent in any phase of the development lifecycle.

What is a Claude Code agent file?

A Claude Code agent file is a markdown file stored in the .claude/agents/ directory that defines a reusable role, such as a security specialist or test engineer, which Claude can spawn as a subagent.

Which tools should I grant to my subagents?

The best practice is to grant the minimal necessary set: read-only tools (Read, Grep, Glob) for researchers and reviewers, and write tools (Edit, Write) only for implementation agents.

Agent Design Best Practices

Name: Agent Design Best Practices
Author: CodySwannGT

byCodySwannGT

•

Herramientas de Colaboración

Standardizes the creation and optimization of Claude Code agent files to ensure focused domains and efficient tool usage.

This skill provides a comprehensive framework for architecting Claude Code agents, emphasizing the separation of capabilities from lifecycle management. It leverages Anthropic’s official guidance alongside battle-tested patterns to help developers build specialized subagents that excel in specific domains like security, testing, or implementation. By following these best practices, teams can prevent common pitfalls such as workflow coupling, tool over-provisioning, and context window fragmentation, resulting in a more robust and predictable AI-assisted development environment.

Características Principales

01Domain-specific agent architecture guidance

02Minimalist tool access optimization strategies

03Context window isolation management for multi-agent teams

04Separation of concerns between agent capabilities and execution lifecycle

051 GitHub stars

06Standardized YAML frontmatter and markdown body structures

Casos de Uso

01Designing specialized security subagents with restricted read-only permissions

02Refactoring monolithic agents into modular, domain-focused teammates

03Establishing governance and consistency for AI subagents across large repositories

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add codyswanngt/lisa agent-design-best-practices

For use in Claude.ai and ChatGPT

Características Principales

01Domain-specific agent architecture guidance

02Minimalist tool access optimization strategies

03Context window isolation management for multi-agent teams

04Separation of concerns between agent capabilities and execution lifecycle

051 GitHub stars

06Standardized YAML frontmatter and markdown body structures

Casos de Uso

01Designing specialized security subagents with restricted read-only permissions

02Refactoring monolithic agents into modular, domain-focused teammates

03Establishing governance and consistency for AI subagents across large repositories

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add codyswanngt/lisa agent-design-best-practices

For use in Claude.ai and ChatGPT