用 GPT-6 Astra 做多智能体共享状态中枢

Lummox @Lummox_eth原推发布于 2026.09.29

在 X 上打开原推

GPT-6 ASTRA STOPS 4 AGENTS FROM FORGETTING THE TASK EVERY TIME THEY HAND IT OFF most multi-agent systems still lose efficiency at the handoff request → planner → shared Astra core → research / execution / verification → updated state → next action the planner picks the next move research adds evidence execution acts verification checks the result then everything flows back into the same shared state the next role inherits the latest context, output, and correction instead of reconstructing the task from zero the agents can stay specialized Astra keeps the task intact between them

这条是引用着另一条写的：x.com/lummox_eth/status/2100288151897981255