Goal-Driven
A multi-agent master-subagent framework for sustaining 100+ hours of continuous coding or math problem-solving.
Run 100-hour automated loops for complex coding, math, and system design challenges.
Goal-Driven is a multi-agent system prompt template designed to keep autonomous agents working for over 100 hours on highly complex problems. It suits developers and researchers tackling intricate logic like compilers or database architecture who need automated persistence.
What it is
The system assigns a master agent to oversee a subagent tasked with a specific, verifiable goal. The master agent checks subagent activity every five minutes, restarts inactive agents, and evaluates results against strict criteria until the objective is fully achieved.
Who it's for
It suits advanced developers, computer science students, and researchers building complex, verifiable systems like compilers or theorem provers. It does not suit casual users or those looking for quick, single-turn code generation helpers.
Key features
Master and Subagent Hierarchy
Separates task execution from supervision to maintain focus on complex, multi-day objectives.
Five-Minute Activity Checks
The master agent periodically checks if the subagent is active, restarting it automatically if it stalls.
Criteria-Based Evaluation
Stops the infinite loop only when explicit, well-defined success criteria are fully satisfied.
Long-Duration Persistence
Designed to sustain continuous effort for over 100 hours on abstract, time-consuming tasks.
Up and running in 3 steps
Our honest verdict
What's great
- Forces autonomous agents to persist on complex tasks for over 100 hours without manual intervention.
- Built-in supervision loop automatically restarts inactive or stalled subagents.
- Strict success criteria verification prevents premature task completion declarations.
- Proven on difficult projects like writing a TypeScript compiler in C++ or SQLite in Rust.
Where it falls short
- Consumes a significant amount of time and LLM API tokens during multi-day runs.
- Can contaminate your AI agent's context window if added directly to permanent skills or plugins.
- Requires a supported multi-agent tool like Claude Code, Codex, or OpenClaw to operate.
Pricing & license
The prompt template is free to use, but the only cost is running your own multi-agent tool, which may consume a high volume of LLM API tokens over 100-hour runs.
Open remote AI-training roles
874 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
Use it, or have us set it up
FAQ
What tools are required to run this?
You need a multi-agent-supported tool such as Claude Code, Codex, or OpenClaw.
Why shouldn't I add this to my agent's skills?
The author advises against adding it to skills or plugins because it could contaminate your context window.
