Column 1
Column 2
Subagents are specialized AI agents configured in your project's .claude/ directory. Here's how to create them:
Basic Configuration
Create or edit .claude/subagents.json:
Key Configuration Fields
- name: Unique identifier for the subagent
- description: What the subagent does
- model: Claude model to use (e.g., claude-opus-4-5-20251101 for advanced tasks)
- instructions: System prompt defining the subagent's behavior and expertise
- tools: Array of tools the subagent can access (bash, file_editor, web_search, etc.)
- permissions: Control what the subagent can access
Common Subagent Patterns for Software Development
- Code Reviewer Subagent
- Documentation Generator Subagent
- Test Writer Subagent
- Security Auditor Subagent
- Performance Optimizer Subagent
Best Practices for Defining Subagents
- Narrow Focus: Each subagent should have a specific, well-defined purpose
- Clear Instructions: Write detailed system prompts that define expertise and behavior
- Minimal Permissions: Grant only the tools and permissions each subagent needs
- Model Selection: Use appropriate models (Opus for complex analysis, Haiku for simple tasks)
- Tool Selection: Only include tools the subagent actually needs
- Naming Convention: Use descriptive, kebab-case names (e.g., code-reviewer, test-writer)
Invoking Subagents
You can invoke subagents through:
- Custom slash commands (like your /review and /security-review commands)
- Direct CLI calls: kiro-cli subagent <name> <task>
- Within agent workflows
Your Current Setup
I notice you already have custom slash commands configured:
- /review: Likely uses a code review subagent
- /security-review: Uses a security audit subagent
- /pr-comments: Integrates with GitHub
