improve-agent
Overview
Systematic agent improvement loop for simulation competitions. Each iteration is one scientific cycle: diagnose → hypothesize → implement → verify → decide.
Iron rule: No new version without a written hypothesis FIRST (log-experiment). No submission without local eval FIRST (run-local-eval).
When to Use
- User says "improve", "make it better", "next version", "iterate", "vN", "what should we try"
- Current agent has a known failure mode or rating ceili
[Description truncada. Veja o README completo no GitHub.]