name: auto-model
description: Automatic model selection based on task type. Routes planning to Opus, coding to Sonnet, simple tasks to Haiku. Optimizes cost and quality automatically.
license: MIT
metadata:
author: thesimpleapp
version: "1.0"

Auto Model Selection

Automatically route tasks to the optimal model based on complexity and type.

Model Hierarchy

┌─────────────────────────────────────────────────────────────┐
│                    MODEL SELECTION                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  OPUS 4.5 (Reasoning)                                       │
│  ├── Planning & Architecture                                │
│  ├── Complex debugging                                      │
│  ├── Multi-file refactoring                                 │
│  └── Critical decision-making                               │
│                                                             │
│  SONNET 4 (Balanced)                                        │
│  ├── Code generation                                        │
│  ├── Code review                                            │
│  ├── Feature implementation                                 │
│  └── Documentation                                          │
│                                                             │
│  HAIKU (Fast)                                               │
│  ├── Simple edits                                           │
│  ├── Typo fixes                                             │
│  ├── Rename operations                                      │
│  └── Quick questions                                        │
│                                                             │
│  GEMINI FLASH (Bulk)                                        │
│  ├── Research tasks                                         │
│  ├── Large file analysis                                    │
│  ├── Bulk operations                                        │
│  └── High-volume queries                                    │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Selection Rules

Use OPUS 4.5 When:

Task contains:
- "plan", "design", "architect", "strategy"
- "decide", "evaluate", "compare approaches"
- "debug complex", "trace issue", "root cause"
- "refactor entire", "restructure", "migrate"
- "security audit", "vulnerability assessment"

Scope:
- Touches 10+ files
- Affects core architecture
- Requires deep reasoning
- High-stakes decisions

Examples:
- "Plan the authentication system architecture"
- "Debug why the app crashes on startup"
- "Design the database schema for multi-tenancy"
- "Evaluate whether to use GraphQL or REST"

Use SONNET 4 When:

Task contains:
- "implement", "build", "create", "add"
- "fix", "update", "modify"
- "review", "analyze code"
- "write tests", "document"

Scope:
- 1-10 files
- Standard feature work
- Clear requirements
- Moderate complexity

Examples:
- "Implement the user profile page"
- "Add form validation to checkout"
- "Review this PR for issues"
- "Write unit tests for the auth service"

Use HAIKU When:

Task contains:
- "fix typo", "rename", "format"
- "simple", "quick", "minor"
- "update import", "change variable"
- "answer question about"

Scope:
- 1-2 files
- < 50 lines changed
- No logic changes
- Mechanical edits

Examples:
- "Fix the typo in README"
- "Rename userId to customerId"
- "Update the import path"
- "What does this function do?"

Use GEMINI FLASH When:

Task contains:
- "research", "explore", "find examples"
- "analyze all", "scan entire", "bulk"
- "summarize", "extract from"
- "compare many", "list all"

Scope:
- Read-only operations
- High token volume
- Research/exploration
- Cost-sensitive bulk work

Examples:
- "Research best practices for rate limiting"
- "Analyze all error handling in the codebase"
- "Find all TODO comments across the project"
- "Summarize the main patterns in this repo"

Detection Algorithm

def select_model(task: str, file_count: int, estimated_complexity: str):

    # Opus triggers
    opus_keywords = ["plan", "design", "architect", "strategy", "debug complex",
                     "refactor entire", "migrate", "security audit", "evaluate"]
    if any(kw in task.lower() for kw in opus_keywords) or file_count > 10:
        return "opus"

    # Haiku triggers
    haiku_keywords = ["typo", "rename", "format", "simple", "quick", "minor",
                      "update import", "change variable"]
    if any(kw in task.lower() for kw in haiku_keywords) and file_count <= 2:
        return "haiku"

    # Gemini triggers
    gemini_keywords = ["research", "explore", "find examples", "analyze all",
                       "scan entire", "bulk", "summarize", "list all"]
    if any(kw in task.lower() for kw in gemini_keywords):
        return "gemini-flash"

    # Default to Sonnet
    return "sonnet"

Claude Code Integration

In Claude Code, use the --model flag or configure defaults:

# Override for specific task
claude --model opus "Plan the new authentication architecture"

# Use default selection
claude "Implement the login form"  # Auto-selects Sonnet

Settings Configuration

// ~/.config/claude-code/settings.json
{
  "defaultModel": "sonnet",
  "modelRules": {
    "planning": "opus",
    "coding": "sonnet",
    "simple": "haiku",
    "research": "gemini-flash"
  },
  "autoSelect": true
}

Cost Optimization

Monthly Estimate (Heavy Usage)

Without auto-selection: ~$500/month (all Opus)
With auto-selection:    ~$150/month (optimized mix)

Override Patterns

Force a specific model when auto-selection isn't optimal:

"Use Opus for this: [task]"       → Forces Opus
"Quick task: [task]"              → Forces Haiku
"Research: [task]"                → Forces Gemini Flash
"[task]"                          → Auto-select

Subagent Model Assignment

When spawning parallel agents:

Main Orchestrator:     Opus (planning, coordination)
Feature Agents:        Sonnet (implementation)
Test Agents:           Sonnet (test writing)
Doc Agents:            Haiku (documentation updates)
Research Agents:       Gemini Flash (codebase scanning)

Monitoring

Track model usage to optimize over time:

Task: "Implement user dashboard"
Selected: Sonnet 4
Reason: "implement" keyword, 5 files, standard complexity
Result: Success
Time: 45 seconds

Integration with Other Skills

/auto-model           → Model selection rules
       ↓
/parallel-swarm       → Assigns models to agents
       ↓
/orchestrator         → Routes tasks with optimal models