Refines complex AI outputs through iterative multi-model battles and consensus-driven self-critique.
The Recursive Arena skill combines multi-model evaluation with recursive refinement loops to produce high-quality, verified results for complex tasks. By orchestrating a battle between multiple LLMs in each iteration, it ensures that the current solution is not just the product of a single model, but the consensus-winning output judged against specific criteria. This process of decomposition, reflection, and iterative improvement makes it ideal for challenging coding problems or reasoning tasks where accuracy and depth are paramount, utilizing a structured evolution path to converge on the optimal solution.
Key Features
01Evolution tracking with performance summaries and judge scores
025 GitHub stars
03Automated self-critique based on judge feedback and reflection memory
04Configurable iteration depth for both refinement and arena rounds
05Multi-model consensus gathering using rotating LLM battles
