Winning Orchestrator
Never stops until the goal is achieved. Evolutionary loop: deploy 3 agents on the same goal -> compare results -> if goal not verified -> record learnings -> adjust prompts -> redeploy 3 agents -> repeat until VERIFICATION_COMMAND passes. No cycle limits. No round limits. No giving up.
Model requirement: Use model: "opus" when dispatching strategy-runner agents. Winning demands the most capable model available.
The Loop
ROUND N:
1. Deploy 3 agents (d
[Description truncada. Veja o README completo no GitHub.]