Agent Creator Skill

Comprehensive guide for creating professional Claude Code subagents.

Quick Start: Create an Agent

---
name: my-agent
description: What this agent does. Use PROACTIVELY for [tasks].
tools: Read, Write, Edit  # Optional
model: sonnet             # Optional
skills: skill1, skill2    # Optional
---

You are an expert in [domain].

## Core Expertise
- Skill 1
- Skill 2

## Workflow

When invoked:
1. Step 1
2. Step 2
3. Deliver results

## Quality Checklist
- ✅ Check 1
- ✅ Check 2

Agent Structure

Required Fields

• •name: Unique identifier (lowercase-with-hyphens)
• •description: When to use this agent (natural language)

Optional Fields

• •tools: Specific tools (or omit to inherit all)
• •model: sonnet/opus/haiku/inherit (default: sonnet)
• •skills: Auto-load reference documentation
• •permissionMode: Permission handling strategy

Agent Archetypes

1. Analyzer/Reviewer

Reviews and provides feedback

tools: Read, Grep, Glob, Bash
model: sonnet

2. Generator/Creator

Creates new code/content

tools: Read, Write, Glob, Bash
model: sonnet

3. Debugger/Fixer

Fixes problems

tools: Read, Edit, Bash
model: sonnet

4. Researcher/Explorer

Searches and gathers info

tools: Read, Glob, Grep, Bash
model: haiku  # Fast

5. Tester/QA

Tests and verifies

tools: Read, Write, Edit, Bash
model: sonnet

System Prompt Sections

Essential sections for agent system prompts:

1. •Core Expertise: What agent knows
2. •Workflow: Step-by-step process
3. •Patterns: Common code patterns
4. •Best Practices: Do's and don'ts
5. •Quality Checklist: Verification steps

Tool Selection

Model Selection

Activation Strategies

Proactive (Automatic)

description: Use PROACTIVELY after code changes
description: MUST BE USED for all UI tasks

Domain-Specific

description: Expert React developer
# Auto-invoked for React tasks

File Locations

# Project-level (priority 1)
.claude/agents/agent-name.md

# User-level (priority 2)
~/.claude/agents/agent-name.md

Best Practices

✅ Focused purpose - Single responsibility
✅ Clear triggers - When to activate
✅ Concrete examples - Code patterns
✅ Quality checks - Verification steps
✅ Right tools - Minimal necessary

❌ Too broad scope
❌ Vague description
❌ No examples
❌ Missing quality checks

Quick Templates

Code Reviewer

---
name: code-reviewer
description: Reviews code. Use after changes.
tools: Read, Grep, Glob, Bash
---

You are a senior code reviewer.
When invoked: check quality, security, best practices.

Component Creator

---
name: component-creator
description: Creates components. Use for new features.
tools: Read, Write, Glob, Bash
skills: architecture-patterns
---

You are a component expert.
When invoked: generate following project patterns.

References

See references/agent-structure.md for complete documentation on:

• •YAML fields reference
• •All agent archetypes
• •Tool selection guide
• •Skills integration
• •Permission modes
• •Multi-agent workflows
• •Testing strategies
• •Domain examples

Workflow

1. •Define purpose and trigger
2. •Choose archetype
3. •Select tools and model
4. •Write system prompt
5. •Add skills if needed
6. •Test agent
7. •Iterate based on usage