agent-arena

Agent Arena facilitates evidence-first multi-agent debates among AI coding agents such as Claude Code and OpenAI Codex to deliver independent analyses and critiques of code and architecture decisions. It is particularly suited for high-stakes scenarios, including architectural reviews, implementation plan assessments, and bug root-cause analysis, while ensuring that dissenting viewpoints and evidence verification are preserved. The tool supports diverse model backends, allowing cross-collaboration among differing AI models, though it functions as a protocol rather than a direct execution orchestrator.