Google's open-source framework that treats the agent harness as the trainable surface: contrastive Propose–Probe–Compose refines the harness per task without ground-truth verifiers, while a per-domain experience library turns adopted harnesses into lessons for the next task. Weights stay frozen; a single step lifts GLM-5 and Gemini 3.5 Flash by up to 11.5 points across three professional benchmarks.
As an Amazon Associate, we earn from qualifying purchases.