model-selection

# Model Selection Skill

Choose the right model for custom agent tasks based on complexity, cost, and performance requirements.

## Interactive Model Selection

Use AskUserQuestion to understand requirements and recommend the optimal model:

```yaml
# Question 1: Primary Priority (MCP: CLI best practices - tradeoff selection)
question: "What is your primary priority for this agent?"
header: "Priority"
options:
  - label: "Cost Efficiency (Recommended)"
    description: "Minimize API costs, high-volume operations"
  - label: "Balanced Performance"
    description: "Good quality at reasonable cost for most tasks"
  - label: "Maximum Quality"
    description: "Best results regardless of cost, complex reasoning"
  - label: "Lowest Latency"
    description: "Real-time responses, user-facing interactions"

# Question 2: Task Complexity (MCP: Agent SDK model selection)
question: "How complex is the task this agent will perform?"
header: "Complexity"
options:
  - label: "Simple"
    description: "Transformations, extraction, formatting, classification"
  - label: "Moderate"
    description: "Code generation, analysis, planning, most tasks"
  - label: "Complex"
    description: "Architecture decisions, multi-step reasoning, critical code"
  - label: "Variable"
    description: "Mix of simple and complex tasks in one agent"
```

Use these responses to apply the decision tree and recommend the appropriate model.

## Purpose

Guide selection of appropriate Claude model (Haiku, Sonnet, Opus) for custom agent tasks to optimize cost, speed, and quality.

## When to Use

- Designing a new custom agent
- Optimizing existing agent performance
- Balancing cost vs quality
- Meeting specific latency requirements

## Model Overview

| Model | Speed | Cost | Quality | Use Case |
| --- | --- | --- | --- | --- |
| **Haiku** | Fastest | Lowest | Good | Simple tasks, high volume |
| **Sonnet** | Fast | Medium | Very Good | Most tasks, balanced |
| **Opus** | Slowest | Highest | Best | Complex