Dear authors,
Great work, and thanks for releasing the code.
I noticed that there are several places where the implementation is inconsistent with what's described in the paper, for example:
The agent graph is a fixed linear chain that runs all three tools every pass on the current plan step — RouteQuery/question_router is defined, but is dead code.
In method grade_plan_completion, the note will always return "incomplete" until all tasks are completed, but the paper claims the use of an early exit condition.
Could you help check whether the code is outdated?
Thank you!
Dear authors,
Great work, and thanks for releasing the code.
I noticed that there are several places where the implementation is inconsistent with what's described in the paper, for example:
The agent graph is a fixed linear chain that runs all three tools every pass on the current plan step — RouteQuery/question_router is defined, but is dead code.
In method grade_plan_completion, the note will always return "incomplete" until all tasks are completed, but the paper claims the use of an early exit condition.
Could you help check whether the code is outdated?
Thank you!