Long-Running Harness

Orchestrate a multi-agent harness for building complex, long-running applications. Based on Anthropic's harness design for long-running apps.

When to Use

• Building full applications that require hours of sustained agent work
• Projects where a single agent session loses coherence or quality
• Tasks needing iterative generation + evaluation cycles (UI design, full-stack features, games)
• When you need quality beyond what a single-shot agent produces

Architecture: Three-Agent System