thank @Wendong-Fan ,i will take it

To ensure we have a solid design before implementation, an RFC is needed for this feature (Including Problem, Goal, Solution etc). @zjrwtx , would you be able to help prepare this?

yeah sure

Existing problems ：
Lack of task replanning mechanism
If a task fails repeatedly, it may indicate that there are flaws in its decomposition method or execution strategy. Directly creating a new agent is just a "remedy" and does not optimize the task structure at the root.

Failure to utilize existing resources
The system may already have nodes capable of handling the task, but they are not used due to misjudgment by the coordinator.

Lack of failure cause analysis
The specific reasons for failure (such as model limitations, lack of tools, logical errors, etc.) are not recorded, making it impossible to develop a more targeted recovery strategy.

The task nesting depth threshold is fixed get_depth() >= 3 is a hard - coded judgment condition and cannot adapt to all task types.

Task Replanning Module that:

Analyzes the reason for failure
Adjusts the task description or goal
Makes smarter decisions on whether to rewrite,retry, decompose, or create a new worker