Does this change the final output quality?

No. Complex tasks still run on the strongest models, so quality on important work remains the same while routine steps cost far less.

How do I add this to a non-OpenClaw agent?

Paste the model-routing instructions into your project's system prompt or instruction file so the agent follows the same classification logic.

zscole/model-hierarchy-skill

Route agent tasks to the right model tier for major cost savings.

SkillsAgent Skills 339Open source

View on GitHub

Updated 2026-06-16

zscole/model-hierarchy-skill GitHub repository

What is zscole/model-hierarchy-skill?

Model-hierarchy-skill is an installable capability that adds intelligent model routing to AI agent systems. It defines clear tiers for tasks based on difficulty and maps each tier to an appropriate model price range.

Agents first classify incoming work as routine, moderate, or complex, then execute on the matching tier. Sub-agents spawned during runs default to cheaper models unless the parent task signals higher needs.

The skill targets developers and teams operating agents on platforms such as OpenClaw, Claude Code, or Codex who want to control growing inference bills without rewriting their core logic.

Capabilities

classify task complexity

route tasks to model tiers

reduce agent token costs

integrate routing rules into agents

What you can build with zscole/model-hierarchy-skill

High-volume routine operations

Handle file reads, status checks, and formatting on the lowest-cost models to keep daily token usage economical.

Mixed development workflows

Send code generation and summarization to mid-tier models while reserving premium models for debugging sessions.

Budget-conscious scaling

Grow agent usage across many users or projects without proportional increases in monthly model spend.

Install zscole/model-hierarchy-skill

Quick start

model-hierarchy-skill/
├── SKILL.md          # The skill (install this)
├── README.md         # You're here
├── tests/
│   ├── test_classification.py
│   └── scenarios.json
└── examples/
    ├── openclaw.md
    └── claude-code.md

1Copy SKILL.md into your agent's skills folder for OpenClaw or add the routing rules to CLAUDE.md.
2Restart the gateway service so the new skill is loaded.
3Run the included classification tests to verify tier assignments work as expected.
4Process a few sample tasks and confirm the correct model tier is selected each time.
5Review token logs after a day of operation to measure actual cost reduction.

Works with

OpenClawClaude CodePython