Chapter 8: Subagents
What are Subagents?
Subagents are specialized agents in Claude Code, each designed to handle specific types of tasks. Like experts in a team, each has their specialty.
Core Advantages:
- 🎯 Specialized: Optimized for specific tasks
- ⚡ Efficient: Faster execution
- 💰 Economical: Some subagents use cheaper models
- 🔀 Parallel: Can run multiple subagents simultaneously
Available Subagent Types
1. Explore Agent
Purpose: Quickly explore codebase, find files and code
Features:
- Uses Haiku model (fast and cheap)
- Suitable for search and browsing tasks
- No modification operations
Example:
8.2 可用子代理类型
bash
You: explore codebase and find all API endpoints
Claude: I'll use the Explore Agent to find API endpoints.
[Launching Explore Agent...]
Found 23 API endpoints across 5 files:
- GET /api/users (src/routes/users.js:12)
- POST /api/users (src/routes/users.js:45)
...2. Plan Agent
Purpose: Create implementation plans
Features:
- Analyze requirements
- Create steps
- Identify key files
- Consider architectural trade-offs
Example:
bash
You: plan how to add user authentication
Claude: I'll use the Plan Agent to create an implementation plan.
[Launching Plan Agent...]
Implementation Plan:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
1. Create auth middleware (src/middleware/auth.js)
2. Add JWT generation utility (src/utils/jwt.js)
3. Update user routes with authentication
4. Create login/register endpoints
5. Add tests for auth flow
Estimated files to modify: 6
Estimated new files: 3
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Proceed with implementation? (yes/no)3. Bash Agent
Purpose: Handle command-line tasks
Features:
- Execute Git operations
- Run builds and tests
- Manage dependencies
- System commands
Example:
bash
You: install dependencies and run tests
Claude: I'll use the Bash Agent for this.
[Launching Bash Agent...]
$ npm install
✓ Installed 125 packages
$ npm test
✓ 45 tests passed4. Feature Dev Agents
Includes three subagents:
- Code Explorer: Deep analysis of existing features
- Code Architect: Design feature architecture
- Code Reviewer: Review code quality
Example:
bash
You: develop a comment system feature
Claude: I'll use Feature Dev agents for this.
[Code Explorer analyzing existing patterns...]
✓ Analyzed comment patterns in codebase
[Code Architect designing architecture...]
✓ Created architecture blueprint
[Implementation...]
✓ Feature implemented
[Code Reviewer checking quality...]
✓ Code review passedUsing Subagents
Automatic Selection
Claude Code intelligently selects appropriate subagents:
8.3 使用子代理
bash
# This automatically uses Explore Agent
You: find all test files
# This automatically uses Plan Agent
You: plan adding payment feature
# This automatically uses Bash Agent
You: commit code changesManual Specification
Explicitly specify which subagent to use:
bash
# Use Explore Agent
You: /explore search all TODO comments
# Use Plan Agent
You: /plan implement user permission system
# Use Bash Agent
You: /bash run database migrationParallel Execution
Run multiple subagents simultaneously:
bash
You: run in parallel: explore codebase structure and run tests
Claude: I'll run these tasks in parallel.
[Explore Agent] Analyzing codebase structure...
[Bash Agent] Running tests...
✓ Explore Agent completed: Found 45 modules
✓ Bash Agent completed: 120 tests passedPractical Examples
Case 1: Large Codebase Analysis
8.4 实战案例
bash
You: analyze architecture of this large project
Claude: This is a large codebase. I'll use specialized agents.
[Explore Agent: quick scan]
- 234 files found
- 12 main modules identified
- Using: React, Node.js, PostgreSQL
[Analysis complete in 30 seconds]
Summary:
- Frontend: React with Redux (src/client/)
- Backend: Express API (src/server/)
- Database: PostgreSQL with Sequelize
- Tests: Jest (85% coverage)Case 2: Feature Development Workflow
bash
You: develop a complete blog comment feature
Claude: I'll use Feature Dev agents for comprehensive development.
[Step 1: Code Explorer]
Analyzing existing comment patterns...
✓ Found similar implementation in forum module
[Step 2: Code Architect]
Designing comment system architecture...
✓ Created blueprint:
- Comment model
- API endpoints
- Frontend components
- Notification service
[Step 3: Implementation]
Creating files...
✓ models/Comment.js
✓ routes/comments.js
✓ components/CommentList.jsx
✓ services/notification.js
[Step 4: Code Reviewer]
Reviewing implementation...
✓ No security vulnerabilities
✓ Follows project conventions
⚠ Suggestion: Add rate limiting
[Step 5: Tests]
Generating tests...
✓ 18 tests created and passingCase 3: Parallel Task Processing
bash
You: complete simultaneously: 1)run tests 2)generate docs 3)analyze performance
Claude: Running 3 tasks in parallel...
[Bash Agent] Running tests...
[General Agent] Generating documentation...
[Explore Agent] Analyzing performance...
Results:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[1] Tests: 95/95 passed ✓
[2] Documentation: api-docs.md created ✓
[3] Performance: 3 bottlenecks found ⚠
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Performance issues:
1. Database N+1 query in users.js:45
2. Large bundle size in dashboard.js
3. Unoptimized images in /assetsChoosing the Right Subagent
| Task Type | Recommended Subagent | Reason |
|---|---|---|
| Search files/code | Explore Agent | Fast, cheap, focused on search |
| Create plans | Plan Agent | Structured thinking, detailed planning |
| Git operations | Bash Agent | Specialized in command-line |
| Complete feature dev | Feature Dev Agents | Full workflow support |
| Code review | Code Reviewer | Professional review capability |
| Simple tasks | General Agent | General purpose |
Decision Flow:
Does task need codebase exploration?
├─ Yes → Use Explore Agent
└─ No → Does task need planning?
├─ Yes → Use Plan Agent
└─ No → Is task command-line operation?
├─ Yes → Use Bash Agent
└─ No → Use General AgentSummary
In this chapter, we learned:
- ✅ Subagent concepts and advantages
- ✅ Available subagent types and purposes
- ✅ How to use subagents (auto/manual/parallel)
- ✅ Practical case demonstrations
- ✅ How to choose appropriate subagents
Key Takeaways:
- Subagents improve efficiency and reduce costs
- Can execute multiple tasks in parallel
- Claude intelligently selects subagents
- Can manually specify specific subagents
Next Step: Chapter 9 introduces the plugin system!