[BUG] Contradiction between comments and implementation in plan updating logic

[William-Wang-Dev]

Description

In the _generate_updated_plan method of the MultiStepAgent class. There's a direct contradiction between the code comments and the actual implementation, which likely causes the planning functionality to behave differently than intended.

The Issue

In agents.py,

def _generate_updated_plan(self, task: str, step: int) -> Tuple[ChatMessage, ChatMessage]:
    # Do not take the system prompt message from the memory
    # summary_mode=False: Do not take previous plan steps to avoid influencing the new plan
    memory_messages = self.write_memory_to_messages()[1:]

However, when examining how write_memory_to_messages() and PlanningStep.to_messages() work together, I discovered that setting summary_mode=False (the default when not specified) actually includes previous plans rather than excluding them.
Looking at memory.py, the PlanningStep.to_messages() method shows:

def to_messages(self, summary_mode: bool, **kwargs) -> List[Message]:
    messages = []
    messages.append(
        Message(
            role=MessageRole.ASSISTANT, content=[{"type": "text", "text": f"[FACTS LIST]:\n{self.facts.strip()}"}]
        )
    )

    if not summary_mode:  # This step is not shown to a model writing a plan to avoid influencing the new plan
        messages.append(
            Message(
                role=MessageRole.ASSISTANT, content=[{"type": "text", "text": f"[PLAN]:\n{self.plan.strip()}"}]
            )
        )
    return messages

This means the condition is the exact opposite of what the comment suggests: when summary_mode=False, plans ARE included, not excluded

[sysradium]

Check out #934