Delegate Command - Direct Agent Invocation

Purpose

Directly invoke a specific specialized agent for a targeted task. Use this when you know exactly which agent you need and want direct control over the delegation.

When to Use

• You need a specific agent's capabilities
• Quick, focused task for one agent
• Bypassing orchestration for simple tasks
• Testing agent behavior
• When workflow commands are overkill

Syntax

/delegate [agent-name] [task description]

Examples:
/delegate planner Design the API structure
/delegate reviewer Check this file for issues
/delegate builder Create a validation utility
/delegate security-auditor Audit the auth module

Available Agents

Agent	Name	Model	Capabilities
🎯 orchestrator	orchestrator	opus	Multi-agent coordination
📋 planner	planner	opus	Planning, architecture, research
🔨 builder	builder	sonnet	Code implementation
🔍 reviewer	reviewer	opus	Code review, quality analysis
🔧 fixer	fixer	sonnet	Bug fixes, issue resolution
🧪 test-writer	test-writer	sonnet	Test creation
📚 doc-fetcher	doc-fetcher	haiku	Documentation research
🔒 security-auditor	security-auditor	opus	Security analysis
♻️ refactorer	refactorer	sonnet	Code improvement

Instructions

When this command is invoked:

Step 1: Parse Request

Extract:

1. Agent name
2. Task description
3. Any context from conversation

Step 2: Validate Agent

Check that:

• Agent name is valid
• Agent is enabled in settings
• Task matches agent capabilities

If invalid:

Unknown agent: "[name]"

Available agents:
- planner: Planning and architecture
- builder: Code implementation
- reviewer: Code review
- fixer: Issue resolution
- test-writer: Test creation
- doc-fetcher: Documentation
- security-auditor: Security analysis
- refactorer: Code improvement

Usage: /delegate [agent] [task]

Step 3: Prepare Context

Gather relevant context for the agent:

Agent: [agent-name]
Task: [task description]

Project Context:
- Working directory: [path]
- Relevant files: [from conversation or inferred]
- Previous work: [if any]

Constraints:
- Follow project conventions (CLAUDE.md)
- Save outputs to appropriate locations
- Report completion status

Step 4: Delegate

Invoke the agent with the Task tool:

Task({
  subagent_type: "[agent-name]",
  prompt: [prepared context + task]
})

Step 5: Report Results

Show agent output and status:

## Agent: [agent-name]

### Task
[Original task]

### Result
[Agent output]

### Files Created/Modified
- [file list if any]

### Status
✓ Complete / ⚠ Partial / ✗ Failed

### Next Steps
[Suggested follow-up if any]

Agent Capabilities Detail

planner

Good for:

• Creating implementation plans
• Architecture design
• Requirement analysis
• Research and comparison

Example tasks:

/delegate planner Design the user authentication system
/delegate planner Compare REST vs GraphQL for our API
/delegate planner Create a migration plan for the database

builder

Good for:

• Writing new code
• Modifying existing code
• Creating utilities
• Implementing features

Example tasks:

/delegate builder Create a date formatting utility
/delegate builder Add validation to the user form
/delegate builder Implement the search functionality

reviewer

Good for:

• Code review
• Quality assessment
• Bug finding
• Best practice check

Example tasks:

/delegate reviewer Review app/services/userService.js
/delegate reviewer Check the authentication module for issues
/delegate reviewer Analyze code quality of recent changes

fixer

Good for:

• Bug fixes
• Addressing review findings
• Error resolution
• Quick patches

Example tasks:

/delegate fixer Fix the null pointer in getUserById
/delegate fixer Address the issues from the last review
/delegate fixer Resolve the failing tests

test-writer

Good for:

• Unit tests
• Integration tests
• Edge case coverage
• Test utilities

Example tasks:

/delegate test-writer Write tests for the validation module
/delegate test-writer Add edge case tests for date parsing
/delegate test-writer Create integration tests for the API

doc-fetcher

Good for:

• API documentation
• Library research
• Best practices
• External references

Example tasks:

/delegate doc-fetcher Get JWT authentication best practices
/delegate doc-fetcher Document the Express.js middleware API
/delegate doc-fetcher Research caching strategies

security-auditor

Good for:

• Security review
• Vulnerability detection
• OWASP checks
• Compliance review

Example tasks:

/delegate security-auditor Audit the authentication system
/delegate security-auditor Check for SQL injection vulnerabilities
/delegate security-auditor Review data encryption practices

refactorer

Good for:

• Code cleanup
• Structure improvement
• Performance optimization
• Technical debt

Example tasks:

/delegate refactorer Clean up the utility functions
/delegate refactorer Reduce complexity in the order processor
/delegate refactorer Extract common patterns into helpers

Quick Reference

# Planning
/delegate planner [design/analyze/plan something]

# Building
/delegate builder [create/implement/add something]

# Reviewing
/delegate reviewer [review/check/analyze file or feature]

# Fixing
/delegate fixer [fix/resolve/address issue]

# Testing
/delegate test-writer [write/add/create tests for X]

# Research
/delegate doc-fetcher [research/document/find X]

# Security
/delegate security-auditor [audit/check/review security of X]

# Refactoring
/delegate refactorer [improve/clean/refactor X]

Notes

• Delegate is for focused, single-agent tasks
• For complex multi-agent work, use /orchestrate
• Agent output is returned directly to you
• Failed delegations can be retried with more context
• Agents follow CLAUDE.md conventions automatically