程序员小灰

self-improving-agent keeps corrections, preferences, successful approaches and failed attempts in layered memory, so the agent carries what it learned into later conversations and repeats fewer mistakes.