sub-agents

# Subagents

## Overview
Subagents are specialized AI assistants that Claude Code can delegate tasks to. Each operates with its own context window, custom system prompt, and configurable tool access.

## Key Features

**Context Management**: Each subagent operates in its own context, preventing pollution of the main conversation

**Specialization**: Task-specific configurations enable higher success rates on designated work

**Reusability**: Once created, subagents work across different projects and teams

**Flexible Permissions**: Individual tool access control per subagent

## Quick Start

1. Run `/agents` command
2. Select "Create New Agent"
3. Choose project-level or user-level scope
4. Define purpose, select tools, customize system prompt
5. Save and invoke automatically or explicitly

## Configuration

### File Locations
- **Project subagents**: `.claude/agents/`
- **User subagents**: `~/.claude/agents/`
- Project-level takes precedence over user-level

### File Format
Markdown with YAML frontmatter containing:
- `name`: Unique identifier
- `description`: Purpose and invocation guidance
- `tools`: Optional comma-separated list
- `model`: Optional model alias (sonnet/opus/haiku) or 'inherit'

### Example Structure
```yaml
---
name: code-reviewer
description: Expert code review specialist. Use proactively after code changes.
tools: Read, Grep, Glob, Bash
model: inherit
---

Your system prompt explaining the subagent's role and approach...
```

## Available Tools

When configuring subagents, you can grant access to these tools via the `tools` field in frontmatter:

### Tools That Require Permission

**Bash** - Executes shell commands in your environment
- Use for: Running tests, git commands, build scripts
- Example: `Bash`

**Edit** - Makes targeted edits to specific files
- Use for: Code modifications, refactoring
- Example: `Edit`

**NotebookEdit** - Modifies Jupyter notebook cells
- Use for: Data science workflows, notebook updates
- Example: `NotebookEdit`