產物檢視器支援快速翻半頁,Mermaid 流程圖終端圖表更清晰
閱讀 AI 產出的長篇架構文件或分析報告時,畫面瀏覽與圖表呈現變得更加流暢。
在產物檢視器(artifact viewer,檢視 AI 產生之獨立檔案的瀏覽視窗)中,現在支援按下 Ctrl+D 向下翻動半頁與 Ctrl+U 向上翻動半頁。同時,在對話紀錄與產物檢視器裡繪製 Mermaid 流程圖時,系統改進了終端機 ASCII 純文字圖表的渲染排版,讓方塊節點與連線更整齊易讀。
發布日期:2026-09-04
閱讀 AI 產出的長篇架構文件或分析報告時,畫面瀏覽與圖表呈現變得更加流暢。
在產物檢視器(artifact viewer,檢視 AI 產生之獨立檔案的瀏覽視窗)中,現在支援按下 Ctrl+D 向下翻動半頁與 Ctrl+U 向上翻動半頁。同時,在對話紀錄與產物檢視器裡繪製 Mermaid 流程圖時,系統改進了終端機 ASCII 純文字圖表的渲染排版,讓方塊節點與連線更整齊易讀。
每次使用 /resume 挑選歷史對話時,現在可以依照個人偏好鎖定預設的呈現清單。
系統在 /config 指令與 settings.json 設定檔中新增了 pickerGrouping 設定項目。你可以將其設定為依工作區分組或是平鋪清單(flat),讓對話挑選介面一開啟就呈現你習慣的版面結構:
{
"pickerGrouping": "workspace"
}可在設定檔中指定為 workspace 或 flat。
在終端機使用 /model 挑選模型時,未特別手動指定思考深度的模型家族現在會主動投入更多推理思考。過去在互動式選單中切換模型時,未經額外設定的模型預設會採用 low(低推理努力程度)。現在預設檔位全面調整為 medium(中等推理努力程度),讓模型在處理程式碼或分析架構時進行更周全的推導。
放手讓子代理獨立執行任務時,查看即時進度不再意外打斷自動化流程。在啟用 always-proceed(一律放行,不逐一詢問工具授權的自動執行模式)的狀態下,過去使用者只要切換開啟子代理詳細資訊面板,原本在背景順暢運作的子代理任務會突然跳出工具執行確認視窗,中斷原本的連續作業。這項問題已經修復,現在開啟或瀏覽詳細面板時,子代理會維持全自動推進。
結束多代理協作或清理舊對話時,硬碟不再默默累積暫存資料夾。
系統在呼叫子代理進行平行作業時,會於 .system_generated/worktrees 目錄下建立 worktree(Git 讓同一專案平行開多份副本的功能)。過去在強制終止子代理或刪除歷史對話時,這些臨時工作樹常未被完全移除而變成孤立殘留檔案。現在清理機制已修正,會在結束時確實移除並釋放磁碟空間。
在登出帳號、結束指令與載入自訂設定時,底層的狀態處理與儲存機制進行了一批強化。使用 /logout 登出時,若系統金鑰圈被繞過或無法連線,系統會直接捷徑刪除檔案中的 token(授權令牌),縮短登出耗費的時間。在離開命令列介面時,系統修復了 SQLite 資料庫的預寫日誌檢查點機制,確保關閉前將最後的工作階段中繼資料完整寫入硬碟。此外,自訂功能探索機制在存取暫存目錄或工作區以外的路徑時,也不會再印出多餘的虛假錯誤日誌。
Reading long architecture documents or analysis reports generated by AI is now smoother with clearer diagrams.
The artifact viewer (a browsing window for inspecting standalone files generated by AI) now supports pressing Ctrl+D to scroll down half a page and Ctrl+U to scroll up half a page. Meanwhile, when rendering Mermaid flowcharts in conversation history and the artifact viewer, the system improves terminal ASCII plain-text diagram layouts, making box nodes and connecting lines cleaner and easier to read.
When using /resume to pick past conversations, you can now lock in a default list view based on personal preference.
The system adds the pickerGrouping setting to the /config command and settings.json configuration file. You can configure it to group by workspace or display a flat list, ensuring the conversation picker opens in your preferred layout:
{
"pickerGrouping": "workspace"
}It can be set to workspace or flat in the configuration file.
When selecting models via /model in the terminal, model families without an explicitly specified thinking depth now proactively invest more reasoning effort. Previously, when switching models in the interactive picker, models without extra configuration defaulted to low reasoning effort. The default level has now been updated to medium reasoning effort, enabling models to perform more thorough reasoning when writing code or analyzing architecture.
Checking live progress while letting subagents work autonomously no longer unexpectedly interrupts automated workflows. When running in always-proceed mode (an automatic execution mode that approves tool usage without prompting for each one), opening the subagent details panel previously caused subagent tasks running smoothly in the background to suddenly pop up tool approval prompts, halting continuous operations. This issue has now been resolved, so subagents continue running fully automatically when opening or viewing the details panel.
Ending multi-agent collaboration or cleaning up old conversations no longer silently accumulates temporary folders on disk.
When calling subagents for parallel execution, the system creates worktrees (a Git feature that creates multiple parallel copies of the same project) under the .system_generated/worktrees directory. Previously, when forcefully terminating subagents or deleting conversation history, these temporary worktrees were often not fully removed, leaving behind orphaned lingering files. The cleanup mechanism is now fixed to reliably remove them upon completion and reclaim disk space.
Underlying state handling and persistence mechanisms have received a batch of enhancements during account logout, command exits, and customization loading. When logging out via /logout, if the system keyring is bypassed or unreachable, the system now short-circuits token removal directly to file storage, speeding up logout execution. When exiting the command-line interface, the system fixes SQLite database WAL checkpointing to ensure trailing session metadata is fully flushed to disk before shutdown. Furthermore, customization discovery no longer logs spurious errors when accessing temporary directories or paths outside the active workspace.
Added half-page scrolling with `Ctrl+D` and `Ctrl+U` to the artifact viewer.
Improved terminal ASCII diagram rendering for Mermaid flowcharts in the artifact viewer and conversation.Added the `pickerGrouping` setting to `/config` and `settings.json` to configure the default conversation list view in `/resume` (flat or grouped by workspace).Changed unselected model families in the interactive `/model` picker to default to medium reasoning effort instead of low.Fixed subagent tasks unexpectedly prompting for tool approval in always-proceed mode while viewing the subagent details panel.Fixed orphaned Git worktrees accumulating on disk under `.system_generated/worktrees` when killing subagents or deleting conversations.Improved `/logout` execution time by short-circuiting token removal directly to file storage when keyring storage is bypassed or unreachable.
Fixed SQLite database WAL checkpointing on CLI exit so trailing session metadata updates are flushed to disk before shutdown.
Fixed customization discovery logging spurious errors when accessing temporary directories or paths outside the active workspace.