---
name: goal-driven-multi-agent
description: A framework for running multi-agent systems that sustain 100+ hours of continuous effort to solve complex coding or math problems with verifiable criteria
triggers:
  - set up a goal-driven multi-agent system
  - run agents for hours on a complex problem
  - create a master agent with subagents
  - solve a hard coding problem with multiple agents
  - long-running multi-agent task
  - use goal-driven to build a compiler
  - automate a complex task with agents until criteria are met
  - keep agents running until a goal is achieved
---

# Goal-Driven Multi-Agent Framework

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

Goal-Driven is a prompt-based framework for running multi-agent systems (Claude Code, Codex, OpenClaw, etc.) that sustain **300+ hours of continuous effort** on extremely complex, verifiable problems. A **master agent** supervises one or more **subagents**, restarting them if they go idle, and halts only when strict success criteria are met.

Ideal use cases: compiler design, theorem proving, database engines, EDA simulation, computational math, and large-scale system architecture.

---

## How It Works

while (criteria not met) { let subagent work on solving the problem toward the Goal }

| Role | Responsibility |
|------|---------------|
| **Master Agent** | Creates subagents, monitors every 5 min, evaluates criteria, restarts idle agents |
| **Subagent** | Breaks down the goal, works continuously, reports status |

The master agent is the **only** entity that decides if the goal is complete. It does **not** stop unless criteria are fully satisfied or the user intervenes manually.

---

## Installation / Setup

Goal-Driven requires no package installation. It is a **prompt template** you paste into your AI agent tool.

### Step 1 — Copy the prompt template

Grab the full prompt from the repository:

https://github.com/lidangzzz/goal-driven

### Step 2 — Fill in Goal and Criteria

Open the template and replace the two placeholder blocks:

Goal: [[[[[YOUR GOAL HERE]]]]]

Criteria for success: [[[[[YOUR CRITERIA HERE]]]]]

### Step 3 — Run in your agent tool

Paste the completed prompt into:
- **Claude Code** — paste into the system prompt or a new conversation
- **Codex CLI** — use as the task description
- **OpenClaw** — configure as the master agent's initial instruction

---

## The Full Prompt Template

Goal-Driven (1 master agent + 1 subagent) System

Here we define a goal-driven multi-agent system for solving any problem.

Goal: [[[[[DEFINE YOUR GOAL HERE]]]]]

Criteria for success: [[[[[DEFINE YOUR CRITERIA FOR SUCCESS HERE]]]]]

Here is the System: The system contains a master agent and a subagent. You are the master agent, and you need to create 1 subagent to help you complete the task.