Cross-task context continuity

[Liuhanlin23]

Currently, every time execute(task) is called, this.history is completely cleared, leaving no memory carried over between tasks.
For example, if a user says "continue the previous task," the agent has no idea what just happened.
My suggestion: when a task finishes, the done text output and key memories should be preserved as a task_summary, which is then injected as additional context when the next task starts. Alternatively, we could take inspiration from browser-use's context management mechanism and keep multiple tasks in a single history without clearing it.
If this proposal is worth pursuing, feel free to assign it to me — I'd love to give it a try.