Antigravity 1.3.2：外掛市集管理員上線，支援直接安裝與自訂擴充套件

繁體中文

新功能

外掛市集管理員上線，支援直接安裝與自訂擴充套件

想讓 AI 擁有更多新技能，現在不必手動拼湊設定檔。

輸入 /plugin 可逛官方市集、安裝本機外掛，啟用前填表設定。外掛可打包技能、規則、子代理、MCP（讓 AI 連接外部工具的協定）伺服器與掛鉤，在特定事件觸發小程式。外掛清單秒開，Windows 安裝與移除更可靠。資訊清單可放在 .antigravity-plugin/ 或 .gemini-plugin/，技能列在 plugin.json。透過 PLUGIN_DATA 存取各外掛的專屬資料夾，也能把命令列工具放進 bin/ 資料夾，讓代理直接按工具名稱執行。skills/ 內的技能全部載入；遠端 MCP 伺服器填妥設定就能啟動，子代理也能用外掛自己的 MCP 工具。

# 終端機直接安裝官方外掛
agy plugin install <plugin-name>@antigravity-plugins-official
# 本機安裝：在管理員選 + Install from local directory...
# 管理已安裝外掛：安裝、移除、啟用、停用、列出清單
/plugin install
/plugin uninstall
/plugin enable
/plugin disable
/plugin list

新功能

跳過授權旗標不再自動核准計畫，保留人工審查確認步驟

你開啟跳過授權旗標讓代理自主執行任務時，重要的規劃步驟不再被直接略過。

以前帶有 --dangerously-skip-permissions 參數時，代理在 /plan 計畫模式下會自動放行提出的實作計畫。現在該旗標只會跳過工具呼叫的授權提示，計畫生成後會再次停下來等待你的審查。如果你希望計畫與其他產物皆自動推進，可在 /config 中調整 Artifact Review 設定。

問題修復

長指令輸出減少 CPU 負擔，修復計時器空轉

讓代理跑耗時又大量輸出的終端機指令時，現在可減少 CPU（處理器）的負擔。

以前代理執行輸出龐大的命令列指令時，每次有新文字進來都會重新掃描整段內容。現在只要文字已是合法的 UTF-8 編碼就不再全盤重掃，減少處理器負擔。過去代理設定帶有喚醒條件的計時器時，若條件已無法滿足，會導致單顆 CPU 核心持續維持在 100% 滿載空轉直到超時。現在計時器會在背景閒置等待，並準時在排程時間觸發。

問題修復

/diff 與 /resume 介面修復，修正按鍵切換與窄視窗文字裁切

在終端機中檢視檔案差異或切換歷史對話時，鍵盤操作與畫面排版現在更加穩定順暢。

以前在 /diff 的提交樹狀檢視中按下 Tab、向右鍵或向左鍵時，游標會跳到其他行，而沒有如提示說明切換檢視。現在按鍵行為已修復，且在所有 /diff 分頁中都能使用 Shift+Tab 切換回上一個檢視。過去在較窄的終端機視窗中使用 /resume 對話選擇器時，從「Other」分頁按下 Enter 匯入對話曾會導致當機，含有中日韓字元的對話標題也會在字元中途被切斷。這些介面操作與顯示問題均已修復。

問題修復

多圖對話自動縮小圖片尺寸，避免 Claude 模型請求失敗

你在對話中傳送多張截圖或讓代理讀取大量圖片時，不再因為圖片體積過大而引發錯誤。以前如果對話中累積了過多圖片，呼叫 Claude 模型時常會發生錯誤。現在無論是你附加或代理讀取的圖片，單邊上限皆由原本的 4096 像素縮小至最多 2576 像素。這樣能確保不論對話中累積多少圖片，都能穩定維持在模型的限制範圍內。

English

FEATURE

Plugin Marketplace Manager Launches, with Direct Install and Custom Extensions

Giving your AI new skills no longer means hand-assembling config files.

Type /plugin to browse the official marketplace and install local plugins, filling in a setup form before enabling. A plugin can bundle skills, rules, subagents, MCP (the protocol that lets AI connect to external tools) servers, and hooks that trigger small programs on specific events. The plugin list opens instantly, and installing or removing on Windows is more reliable. The manifest can live in .antigravity-plugin/ or .gemini-plugin/, with skills listed in plugin.json. Access each plugin's dedicated folder through PLUGIN_DATA, or drop command-line tools into the bin/ folder so agents can run them directly by tool name. All skills under skills/ are loaded; remote MCP servers start once their settings are filled in, and subagents can use the plugin's own MCP tools.

# Install an official plugin directly from the terminal
agy plugin install <plugin-name>@antigravity-plugins-official
# Local install: choose + Install from local directory... in the manager
# Manage installed plugins: install, remove, enable, disable, list
/plugin install
/plugin uninstall
/plugin enable
/plugin disable
/plugin list

FEATURE

Skip-Permissions Flag No Longer Auto-Approves Plans, Keeping the Human Review Step

When you turn on the skip-permissions flag to let the agent run tasks autonomously, the important planning step is no longer bypassed.

Previously, with the --dangerously-skip-permissions flag, the agent would automatically approve proposed implementation plans in /plan mode. Now the flag only skips authorization prompts for tool calls, and the agent pauses again for your review after a plan is generated. If you want plans and other artifacts to proceed automatically, adjust the Artifact Review setting in /config.

BUGFIX

Long Command Output Reduces CPU Load, Fixing a Spinning Timer

Running time-consuming terminal commands with heavy output through the agent now puts less strain on the CPU (processor).

Previously, when the agent ran a command-line instruction with massive output, it rescanned the entire content every time new text arrived. Now, as long as the text is already valid UTF-8, it no longer rescans everything, reducing processor load. In the past, when the agent set a timer with a wake-up condition that could no longer be satisfied, a single CPU core would spin at 100% until the timeout. Now the timer idles in the background and fires exactly at its scheduled time.

BUGFIX

/diff and /resume Interface Fixes, Correcting Key Switching and Text Clipping in Narrow Windows

Keyboard controls and layout are now more stable when viewing file diffs or switching between past conversations in the terminal.

Previously, pressing Tab, the right arrow, or the left arrow in the /diff commit tree view moved the cursor to another line instead of switching views as the hints described. Key behavior is now fixed, and Shift+Tab switches back to the previous view across all /diff tabs. Also, in narrower terminal windows, pressing Enter to import a conversation from the "Other" tab of the /resume conversation picker used to crash, and conversation titles containing CJK characters were cut off mid-character. These interface and display issues have all been fixed.

BUGFIX

Multi-Image Conversations Automatically Shrink Image Size, Preventing Claude Model Request Failures

Sending multiple screenshots in a conversation or having the agent read many images no longer triggers errors from oversized images. Previously, when too many images accumulated in a conversation, calls to the Claude model often failed. Now, whether images are attached by you or read by the agent, the per-side limit has been reduced from 4096 pixels to at most 2576 pixels. This ensures that no matter how many images pile up in a conversation, they reliably stay within the model's limits.

原始 CHANGELOG

Original changelog外掛市集管理員上線，支援直接安裝與自訂擴充套件

Added `/plugin`, an interactive manager for the official Antigravity plugin marketplace: browse and install plugins that bundle skills, MCP servers, subagents, rules, and hooks, or install them from your shell without opening the CLI with `agy plugin install <plugin-name>@antigravity-plugins-official`; install one straight from a folder on your computer with `+ Install from local directory...`; fill in any settings a plugin needs in a simple form before it turns on; and manage installed plugins with `/plugin install`, `uninstall`, `enable`, `disable`, and `list`.
Added more options for plugin authors: keep the manifest in a `.antigravity-plugin/` or `.gemini-plugin/` folder, list skills in `plugin.json`, store data in a per-plugin folder available as `PLUGIN_DATA`, and ship command-line tools in a `bin/` folder that the agent can run by name.
Improved plugin installs: the plugin list opens instantly, an interrupted install never leaves a half-installed plugin behind, and installing or removing plugins on Windows is more reliable.
Improved plugin compatibility: plugins load every skill in their `skills/` folder, remote MCP servers start correctly once their settings are filled in, and a plugin's subagents can use the plugin's own MCP servers.

Original changelog跳過授權旗標不再自動核准計畫，保留人工審查確認步驟

Changed `--dangerously-skip-permissions` to stop auto-approving the agent's implementation plan in `/plan` mode: the flag now only skips tool permission prompts, so plans wait for your review again. Set `Artifact Review` in `/config` if you want plans and other artifacts to proceed automatically.

Original changelog長指令輸出減少 CPU 負擔，修復計時器空轉

Improved CPU usage while the agent runs long shell commands with a lot of output, since output that is already valid UTF-8 is no longer rescanned in full on every update.
Fixed a timer the agent scheduled with a wake-up condition keeping a CPU core at 100% until it expired when the condition could no longer be met; the timer now waits idly and fires on schedule.

Original changelog/diff 與 /resume 介面修復，修正按鍵切換與窄視窗文字裁切

Fixed `Tab`, `→`, and `←` in the `/diff` Commit Tree view moving the cursor to other lines instead of switching views as the `←/→ switch view` hint says, and added `Shift+Tab` to switch to the previous view in every `/diff` tab.
Fixed the `/resume` conversation picker crashing when you pressed `Enter` to import a conversation from the `Other` tab on a narrow terminal, and conversation titles with CJK characters being cut off mid-character.

Original changelog多圖對話自動縮小圖片尺寸，避免 Claude 模型請求失敗

Fixed conversations that collect many images failing on Claude models: images the agent views or you attach are now scaled down to at most 2576 pixels per side (previously 4096), keeping them within the model's limits however many images a conversation contains.