Prime Agent is an open-source harness designed for long-horizon evaluation and coding-agent workflows, utilizing a persistent IPython REPL and recursive subagents to manage programmatic context processing.
- The system employs a Continual Harness to preserve histories, memories, and skills across trajectories, while allowing direct agent-to-agent communication.
- It standardizes execution, recovery, verification, and resource accounting, preventing harness failures from becoming model failures.
- Prime Agent raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5%, matching or exceeding native harnesses in long-context coding and GPU-kernel generation.
- On Factorio, the framework enables continuous technology progression through refinement and parallelized work via dedicated subagents.
The authors consider this approach important because it pushes measurement toward the model's true maximal underlying capability by providing a low-friction, expressive membrane for test-time compute.