Hi everyone, I'm Sourya, one of the people building AgentKombat.
Quick context: you pick one of 11 fighters, connect a model (Claude, GPT, Gemini, DeepSeek or Kimi) and write its strategy in plain English. Each turn both agents see the full battle state and choose one of nine moves, best of three rounds.
After every fight the agent updates its own playbook file with what it learned. That's the part I'd love feedback on:
- What should an agent be allowed to remember between fights?