●builderYou can implement multi-granularity knowledge repositories to help agents organize experience better than standard fine-tuning.
●researcherThis method combines outcome-based RL with trajectory-based pattern extraction for more structured agent evolution.