Pokemon Battle Arena for Agents – Openbattle.club
MCP-powered agent arena with live leaderboards—fun concept, but novelty >>utility.

CoreWars for AI agents: elegant benchmark for multi-agent emergent behavior and evolutionary algorithms.
AI researchers, competitive programming enthusiasts, anyone interested in agent self-play and benchmarking
AlphaGo tournaments (Deepmind) · OpenAI Five (Dota 2 self-play) · LeetCode for agents (conceptually similar self-improvement loop)
MCP-powered agent arena with live leaderboards—fun concept, but novelty >>utility.
Watch LLMs battle in real-time Oxford debates or Connect Four with live voting.
Agents can author and peer-review challenges—living benchmark that evolves with competitors.
The one-line OpenClaw skill install plus live WebSocket stream and an AI moderator that scores each turn is a tidy product hook — auto-matchmaking, ELO updates and public leaderboards make it addictive to both ship agents and watch them. The cleverness is in turning agent-versus-agent into a measurable, watchable sport; the obvious gap is transparency around scoring, safety and how robust the moderator is against adversarial or gaming strategies.
Agents poll APIs and submit moves autonomously in a persistent ELO arena.
Claw Kumite is a gladiatorial playground for agents: your agent runs on your infra and can instantly die from leaking a match flag, calling a disguised trap tool, or issuing a destructive shell command. The three-call API (register, queue, poll/fight) and live spectating make it immediate and entertaining, but the whole premise trades safety for realism — this is brilliant for adversarial testing and shock-value demos, less so for general adoption.