Grand Diomande Research · Full HTML Reader

Model Router — Multi-Model Orchestrator

``` User Goal → OpenAI (gpt-4o) decomposes → Task Plan → builder tasks → Claude (Opus) via Clawdbot → research tasks → OpenAI (gpt-4o) scout → review tasks → Claude (Opus) critic → summaries → Claude (Sonnet) ```

Agents That Account for Themselves proposal experiment writeup candidate score 18 .md

Full Public Reader

---
name: model-router
description: Multi-model orchestrator — OpenAI architects plans, Claude builds, with intelligent task routing
triggers:
- orchestrate
- route
- decompose
- plan task
- multi-model
---

Model Router — Multi-Model Orchestrator

Route tasks between OpenAI (architect/planner) and Claude (builder/implementer) for optimal execution.

Architecture

User Goal → OpenAI (gpt-4o) decomposes → Task Plan
  → builder tasks → Claude (Opus) via Clawdbot
  → research tasks → OpenAI (gpt-4o) scout
  → review tasks → Claude (Opus) critic
  → summaries → Claude (Sonnet)

Agent Profiles

Agent	Role	Provider	Models
architect	Planner/Decomposer	OpenAI	gpt-4o, o1
builder	Implementer	Claude	opus, sonnet
scout	Researcher	OpenAI	gpt-4o, gemini-flash
critic	Code Reviewer	Claude	opus
summary	Synthesizer	Claude	sonnet

Pipelines

Pipeline	Agents	Strategy
build	architect → builder → critic	Sequential
research	scout → summary	Sequential
full	architect → scout → builder → critic → summary	Sequential
quick	builder	Single
plan_only	architect	Single

MCP Tools

`decompose` — Send a goal to OpenAI for task decomposition
`route_task` — Analyze a task and route to optimal agent
`dispatch` — Send a task to Clawdbot for Claude execution
`orchestrate` — Full pipeline: decompose → route → dispatch
`list_agents` — Show all agent profiles and pipelines
`pending_tasks` — List tasks ready for dispatch
`status` — Orchestrator status overview

Usage

Decompose a goal

/orchestrate decompose "Build a REST API with authentication"

Full pipeline

/orchestrate orchestrate "Add dark mode to the app" --pipeline build

Dry run (plan only)

/orchestrate orchestrate "Refactor the database layer" --dry-run

Configuration

OpenAI API Key: Set `OPENAI_API_KEY` env var for full decomposition
Without OpenAI: Falls back to template-based decomposition
State: `[home-path]`
Dispatch: Tasks sent to local Clawdbot sessions via `clawdbot sessions spawn`

Notes

Architect (OpenAI) handles high-level planning — "what to build"
Builder (Claude) handles implementation — "how to build it"
This separation leverages each model's strengths
All dispatched tasks run as isolated Clawdbot sessions
Results can be monitored via `clawdbot sessions list`

Promotion Decision

Attach run IDs, datasets, metrics, and reproduction commands.

Source Anchor

homelab/clawdbot/skills/model-router/SKILL.md

Detected Structure

Method · Evaluation · Architecture