Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

main (latest)

Python Version

3.10+

Operating System

All

Installation Method

pip

Steps to Reproduce

1. Create a Graph with a nested MultiAgentBase executor (e.g., a nested Graph)
2. Either:
   • Enable reset_on_revisit and create a cycle that revisits the nested graph node, OR
   • Use session persistence and deserialize a completed graph run (no next_nodes_to_execute)
3. The nested executor's GraphState gets overwritten with an AgentState

Expected Behavior

reset_executor_state() should handle MultiAgentBase executors correctly — either by resetting their state to the appropriate type or by skipping the reset for executor types that don't use AgentState.

Actual Behavior

reset_executor_state() checks hasattr(self.executor, "state") without distinguishing executor type. Since Graph (a MultiAgentBase) has a .state attribute (GraphState), the method overwrites it with an AgentState instance:

def reset_executor_state(self) -> None:
    if hasattr(self.executor, "messages"):
        self.executor.messages = copy.deepcopy(self._initial_messages)

    if hasattr(self.executor, "state"):
        self.executor.state = AgentState(self._initial_state.get())  # Wrong type for MultiAgentBase

    self.execution_status = Status.PENDING
    self.result = None

Note: __post_init__ has a guard (hasattr(self.executor.state, "get")) that prevents capturing the initial state for MultiAgentBase, so _initial_state stays as an empty AgentState(). But reset_executor_state lacks this guard.

Two call sites are affected:

• _execute_node: guarded by reset_on_revisit (off by default), so only triggers if user opts in with a nested graph in a cycle
• deserialize_state: unguarded — iterates all nodes unconditionally when restoring a completed run

Possible Solution

Add type-aware handling in reset_executor_state(). For example, only reset AgentState-based executors, and for MultiAgentBase executors either call their own reset mechanism or skip the state reset.