Grok Build 1.0.49

當你讓 Grok 處理任務時，它何時會停下來請你確認授權，現在有了更明確的規矩。

現在進行網路搜尋、網頁抓取與圖片生成等操作時，在跳出提示詢問你之前，會先遵循 prompt_policy 設定與既有的規則判定。對於掛鉤（特定事件發生時被觸發的小程式），其權限提示現在於每一種核准模式下都會如實出現，確保自動化腳本在各種環境中都能正常跳出確認。

當你在提示畫面上放行指令時，每一次批准能管轄的範圍現在界定得更精確。

現在當你在權限卡片上點選「允許一次」（Allow once），這項放行僅嚴格套用到下一條 shell 指令，後續指令不會被順道放行。而執行 pkill -f 與 pgrep -f 這類搜尋或終止行程的指令時，如果指定的比對條件會殺死包裝 Grok 自身的 shell 外覆程式（shell wrapper），現在會提早被拒絕執行，避免終端機環境意外中斷。

When you have Grok handle tasks, the rules for when it pauses to ask for your authorization are now clearer.

When performing operations such as web searches, web scraping, and image generation, Grok now follows your prompt_policy settings and existing rule evaluations before popping up a prompt to ask you. For hooks (scripts triggered when specific events occur), permission prompts now appear reliably under every approval mode, ensuring that automated scripts prompt for confirmation properly across all environments.

When you approve commands on the prompt screen, the scope governed by each approval is now defined more precisely.

Now, when you select 'Allow once' on a permission card, this release strictly applies only to the next shell command, and subsequent commands are not passed along with it. When running commands like pkill -f and pgrep -f to search for or terminate processes, if the specified matching pattern would kill the shell wrapper enclosing Grok itself, execution is now rejected early to prevent accidental terminal disruptions.