메뉴
Decision framework for selecting the right model for each task
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
End-to-end pipeline to augment any repository for AI-assisted development. Installs opinionated engineering workflows, generates documentation, and configures Claude Code or Cursor.
Destructive command safety guardrails. Warns before dangerous operations in production and shared environments.
Transform a knowledge graph into a human-readable /docs folder with markdown documentation that both humans and agentic workflows can reference.
Systematic root-cause debugging and incident investigation. No fixes without understanding the problem first.
Architecture review before implementation. Walk through design, failure modes, scope, and test strategy before writing code.
Engineering retrospective and velocity analytics from git history. Generates weekly insights on team productivity, code quality, and contribution patterns.
| name | model-routing |
| version | 1.0.0 |
| description | Decision framework for selecting the right model for each task |
| author | iscmga |
| tags | ["models","routing","cost","optimization"] |
| triggers | {"keywords":["model","routing","cost","haiku","sonnet","opus","cheap","expensive","token"]} |
Use the cheapest model that meets quality requirements. Not every task needs the most powerful model. Route by task complexity to extend budget and reduce latency.
| Task Complexity | Model Tier | Examples |
|---|---|---|
| Simple | Haiku | Formatting, renaming, simple lookups, generating boilerplate, summarizing short text |
| Standard | Sonnet | Code implementation, bug fixes, test writing, code review, documentation |
| Complex | Opus | Architecture design, multi-file refactoring, security analysis, debugging subtle issues, ambiguous requirements |
Ask these questions in order:
Claude Code model can be set per-session:
Subagents spawned via the Agent tool can specify subagent_type:
Explore — research tasks, can use SonnetCursor allows model selection per chat. Apply the same framework:
When building apps that call the Claude API, route at the application level:
def select_model(task_complexity: str) -> str:
routing = {
"simple": "claude-haiku-4-5-20251001",
"standard": "claude-sonnet-4-6",
"complex": "claude-opus-4-6",
}
return routing.get(task_complexity, "claude-sonnet-4-6")
Approximate relative cost per million tokens (input):
| Model | Relative Cost | Best For |
|---|---|---|
| Haiku | 1x | High-volume, simple tasks |
| Sonnet | 3-5x | Balanced quality/cost |
| Opus | 15-25x | Highest quality, complex reasoning |
Routing 60% of tasks to Sonnet and 30% to Haiku instead of sending everything to Opus can reduce costs by 70-80%.