Development & Testing Featured

Subagents

Specialized sub-agents with isolated context — a reviewer, researcher or debugger, each with its own tools and instructions.

Claude Code

Overview

Subagents let you define specialized agents inside Claude Code: a simple Markdown file sets the agent name, its task description, the tools it may use, and its system prompt. When a task matches the description, Claude delegates automatically.

The core benefit is context isolation: a reviewer reads the diff away from your main conversation, a researcher sweeps files without flooding your context with intermediate results — and several can run in parallel on large tasks.

Common patterns: a code-reviewer invoked proactively after edits, a debugger that takes a stack trace and returns the root cause, and a researcher that gathers codebase context before planning. They live in .claude/agents/ per project or ~/.claude/agents/ globally.

Who Is This For?

Splitting large tasks across parallel specialized agents
Automatic code review with a clean context after every change
Isolating codebase research from the main conversation context
Restricting a given agent to read-only tools for safety

Installation

Setup for Claude Code

mkdir -p .claude/agents
# Create reviewer.md with frontmatter: name, description, tools — Claude delegates matching tasks automatically.

Configuration

---
name: code-reviewer
description: Reviews diffs for bugs and style. Use proactively after edits.
tools: Read, Grep, Bash
---
You are a strict senior reviewer...

Visit Source (Official docs)

Explore AI Tools

Discover the best AI tools that complement your skills

Read AI & Design Articles

Tips and trends in the world of design and AI