当你在tg同一个上下文聊了很久的时候，token超过大模型上下文的上限的时候，同时你配置了自动记忆刷新 
agents.defaults.compaction.memoryFlush 
{
  agents: {
    defaults: {
      compaction: {
        reserveTokensFloor: 20000,
        memoryFlush: {
          enabled: true,
          softThresholdTokens: 4000,
          systemPrompt: "Session nearing compaction. Store durable memories now.",
          prompt: "Write any lasting notes to memory/Y...