Self-HPO Framework Recursively Optimizes Coding Agent Policies and Harnesses

Self-HPO implements a recursive framework where a coding agent analyzes its own failures to propose new environment harnesses. It alternates between harness proposal and reinforcement learning policy updates to prevent performance plateaus caused by fixed evaluation environments.