self-improving-agent-builder

Run a closed-loop improvement cycle on any goal-seeking agent implementation:

Each iteration measures L1-L12 progressive test scores, identifies failures with erroranalyzer.py, runs a research step with hypothesis/evidence/ counter-arguments, applies targeted fixes, and gates promotion through regression checks.

Run the L1-L12 progressive test suite on the current agent implementation.

为寻求目标的代理编码持续改进循环：EVAL、ANALYZE、 研究（假设+证据+反驳）、改进、重新评估、决定。 自动提交改进（净值+2%，无回归>5%）并恢复故障。 適用於所有 4 個 SDK 實作。 自动激活“改进代理”、“自我改进循环”、“代理评估循环”、 “基準代理”、“運行改進週期”。 來源：rysweet/amplihack。