cited_ids extract failed for missing "memory"

[zsweet]

def _extract_cited_ids_from_history(self, history: Any) -> List[str]:
        """
        Extract cited bullet IDs from browser-use agent thoughts.

        Parses only the agent's reasoning (model_thoughts), filtering out
        tool calls, action results, and other noise.

        Args:
            history: Browser-use AgentHistoryList

        Returns:
            List of cited bullet IDs found in agent thoughts
        """
        if not history or not hasattr(history, "model_thoughts"):
            return []

        try:
            thoughts = history.model_thoughts()
            # Extract only the thinking/reasoning text (filter noise)
            thoughts_text = "\n".join(
                t.thinking for t in thoughts if hasattr(t, "thinking") and t.thinking
            )

            # Use public utility to extract IDs
            from ..roles import extract_cited_bullet_ids

            return extract_cited_bullet_ids(thoughts_text)
        except Exception:
            # Graceful degradation if extraction fails
            return []

I try run example in browser_use, but always failed to extract the cite_ids. Finally i found that in _extract_cited_ids_from_history, ids only extract from "thinking" field while this field almost MISSING but "momery" is not empty. Is this normal?

AgentOutput(thinking=None, evaluation_previous_goal=None, memory='The user wants me to compare the top 5 best-selling smartphones and analyze the price difference between the 1st and 2nd ranked phones.\nThe first step is to search for the ranking of the best-selling smartphones and their prices.\nI will use a search query in Chinese, matching the task language, following [search_strategies-00004].\nI will use the query "前5款最畅销智能手机价格" (Price of top 5 best-selling smartphones).', next_goal=None, action=[ActionModel(root=SearchActionModel(search=SearchAction(query='前5款最畅销智能手机价格', engine='duckduckgo')))])