Model Router — AI skill for Claude Code
Claude Code skill: route each task to the cheapest Claude model that can do it well.
How to install Model Router
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open strikeout/model-router and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Model Router does
Claude Code skill: route each task to the cheapest Claude model that can do it well.
Alternatives in AI
- Agentrq — AgentRQ: Human-in-loop realtime conversational task manager for AI Agents 1.1k ★
- Claude Chat Exporter — Claude Chat Exporter is a JavaScript tool that allows you to export your conversations with Claude AI into a w 407 ★
- Create Module — Scaffold a fully-typed module with controller, service, router, schema, and model files 325 ★
README
model-router
model-router is a Claude Code skill. It sends each task to the cheapest Claude model that can do the task well.
The skill does not run every task on one model. It reads two properties of a task:
- Judgment load. This is the quantity of ambiguity and reasoning that the task needs.
- Blast radius. This is the cost of a wrong answer.
The governing rule is "route down, escalate on evidence". The skill starts at a low tier. A failure or a complexity that you can see moves the task up by one tier.
The tiers
| Tier | Claude Code model |
Use it for |
|---|---|---|
| Haiku | haiku |
Mechanical work with a clear specification and a cheap check: search, log scans, format conversion, simple extraction |
| Sonnet (default) | sonnet |
Most work: features, refactors in one component, tests, drafts, standard analysis |
| Opus | opus |
Hard reasoning: architecture, debugging across many files, plans with no clear specification, decisions that you cannot reverse. Opus also runs the orchestrator. |
| Fable | fable |
All code reviews and security reviews, the hardest tasks, and work in biology, cybersecurity and LLM research |
Each alias resolves to the newest model in its family. The `MODELS` table in `scripts/route.py` holds the API model ID of each tier.
What the repository contains
| Path | Contents |
|---|---|
SKILL.md |
The routing policy. Claude Code reads this file when the skill starts. |
references/routing-rubric.md |
The full rubric, with worked examples and anti-patterns |
references/constraints-and-budgets.md |
Data sensitivity, context budgets, local models and the outage procedure |
references/claude-code-setup.md |
The setup of the worker subagents in Claude Code |
references/api-orchestration.md |
The integration with a custom API harness, and the Fable redirect |
assets/agents/ |
Four worker subagents, one for each tier |
scripts/route.py |
A deterministic tier classifier with no dependencies |
Related Skills
Jev Router
Route to the cheapest model in claude code for your task using jev-router
Claude Auto Router
Auto-route Claude Code prompts to the cheapest model that can handle them. Save 3-5x on tokens. One-command in
Minga Router
Stop paying Opus prices for typo fixes. A Claude Code skill that routes each subagent task to the cheapest mod
Route Feedback
Tell decision-router which model the last routed task should have used
Route Recommend
Ask decision-router which model fits a task, without running it
Flux OS
Send every LLM request to the cheapest model that can handle it. Drop-in OpenAI + Anthropic compatible router
Related Agents
Gemini Runner
Delegates one scoped implementation chunk to Gemini CLI via ~/.claude/gemini/run.sh, choosing the cheapest mod
Router Normal
Tier N (standard, well-specified) tasks from smart-task-router.
Model Router
Use when dispatching work and unsure which model tier to use — returns a recommended model and effort level ba