發布日期：2026-10-05
同時開多個對話讓 AI 處理不同任務時，現在不必切換視窗就能掌握進度。
側邊面板會在其他 Claude 對話跑完或卡住等你時，跳出小通知。這項通知只在下一輪檢查仍是相同狀態時才跳，短暫切換狀態不會吵。工作階段清單中每個項目的第二行會顯示 AI 最後一句回覆，並自動去除 Markdown 符號。若對話正在等待，會清楚寫出在等什麼，例如要你回答或允許沙盒（限制程式執行範圍的防護機制）權限。工作階段狀態標記也改用圖示：執行中轉圈、等待時顯示黃色問號、跑完十分鐘內顯示綠勾，之後變灰點，並拿掉了「忙」與「閒置」文字。
開啟側邊面板查看各項 AI 工具的使用進度時，新的版面配置讓關鍵資訊更醒目。
面板現在會自動將最緊的額度排在第一位。每個工具改為佔用兩行，名字統一放在左欄，並拿掉了模型名稱。用量低於百分之五十的視窗會顯示為灰色，只有值得注意的項目才會上色。Antigravity（agy，Google 開發的 AI 工具）的 Claude 與 GPT 額度集不再出現在面板中。
當某些 AI 代理程式無法順利讀取額度時，現在能直接查明具體原因。
執行 usage status --json 指令時，讀不到額度的工具會帶上 reason 欄位，說明是 not_signed_in 未登入、no_data 無資料還是 error 錯誤。在面板上，Claude 與 Codex 讀取不到額度時會改成顯示具體原因，不再直接消失。Antigravity 與 Grok 則比照選單列的邏輯保持隱藏，只有在讀取出錯時才會顯示。
When you run multiple conversations letting AI handle different tasks, you no longer need to switch windows to track progress.
The side panel now pops up a small notification when another Claude conversation finishes or gets stuck waiting for you. This notification only appears if the status is still the same at the next check, so a brief status flicker won't bother you. The second line of each item in the session list shows the AI's last reply, with Markdown symbols automatically stripped. If a conversation is waiting, it clearly states what it's waiting for, such as your answer or permission for the sandbox (a protective mechanism that limits what programs can do). Session status indicators now use icons too: a spinning circle while running, a yellow question mark while waiting, a green checkmark within ten minutes of finishing, then a gray dot, and the "busy" and "idle" text labels have been removed.
When you open the side panel to check usage progress across your AI tools, the new layout makes key information stand out.
The panel now automatically places the tightest quota first. Each tool now takes up two lines, with names aligned in the left column, and the model name has been removed. Windows with usage below fifty percent appear in gray, and only items worth noticing get colored. The Claude and GPT quota sets for Antigravity (agy, an AI tool developed by Google) no longer appear in the panel.
When certain AI agents fail to read their quotas, you can now find out the specific reason directly.
When you run the usage status --json command, tools whose quotas can't be read now carry a reason field, indicating whether it's not_signed_in, no_data, or error. In the panel, when Claude and Codex quotas can't be read, they now show the specific reason instead of simply disappearing. Antigravity and Grok follow the menu bar logic and stay hidden, appearing only when a read error occurs.
**Claude Code side pane notifications:** A toast appears when another Claude conversation finishes or starts waiting for you. It fires only when the next refresh still shows the same state, so brief flips don't ping.
**Session rows show the latest reply:** Each conversation's second row shows the last assistant reply (Markdown stripped); a waiting conversation says what it waits for, such as needs your answer or sandbox permission.
**Session marks follow state:** A spinner while working, a yellow `?` while waiting for you, a green check for ten minutes after finishing, then a gray dot. The Busy and Idle labels are gone.**Side pane quota block is easier to scan:** The tightest quota is listed first, each tool takes two rows with its name in a left column, model names are gone, and windows under 50% are gray so only the ones worth watching are colored. Antigravity's Claude / GPT pool is no longer shown in the pane.**`usage status --json` explains missing quotas:** Unavailable agents carry `reason` (`not_signed_in`, `no_data` or `error`). The pane shows Claude and Codex with that reason instead of dropping them; Antigravity and Grok stay hidden like the menu bar unless reading fails.