Usage 0.30.22

你跟 Claude Code 對話到一半想確認還能問幾題，不必再特地跳出視窗開終端機查帳。

只要先後安裝市集與外掛，就能依序在 Claude Code 對話介面輸入以下指令查詢：

/plugin marketplace add aqua5230/usage
/plugin install usage@usage
/usage:quota

執行後會列出 Claude Code 與 Codex 的 5 小時及每週額度（quota，模型使用上限）用了多少與何時重置。外掛只呼叫 usage-cli status --json，未安裝時會透過 uvx 執行。指令使用 usage-cli 是為避免與其他工具同名衝突，而且只有你主動輸入才會執行，Claude 絕不會在背景自行觸發。

打算寫腳本抓取帳號狀態時，串接程式再也不必為了計算剩餘時間而大費周章。

現在執行 usage status --json 時，輸出格式直接替你算好倒數時間。每個統計視窗新增了 resets_in_seconds 欄位回傳額度重置倒數，每個代理人（agent，負責執行特定任務的 AI 程式）也多了 age_seconds 記錄已運作秒數。兩項數值都以報表產生的 generated_at 時間點為準計算完成，接收資料的下游工具不用再自己拿兩個時間戳記相減。

使用精簡模式讓 AI 長話短說時，換班交接的開場訊息不再生硬冰冷。

系統在精簡模式（terse mode，指示 AI 保持回答簡短的設定）中移除了兩行提示詞。首先是刪去「不准淡忘」的嚴格施壓要求，因為每則訊息發送時的提醒機制本來就能穩定簡潔語氣。其次是拿掉「不要為了親切多寫」的禁令，避免它與每位使用者必讀的進度交接摘要互相衝突。

打開瀏覽器或手機查看用量報表時，畫面中的統計數據不再排列得歪歪斜斜。

在 HTML 報表裡，四張關鍵指標卡的高度現在保持完全一致。記錄 token（AI 處理與計費的文字單位）消耗的卡片大字只放約略值，精確數值與漲跌幅度則改以小字分行顯示，解決了過去卡片文字換行不一的問題。手機版各區塊也在兩側補上 16px 的留白，內容不再緊貼著卡片邊界。

When you're mid-conversation with Claude Code and want to check how many questions you have left, you no longer need to jump out of the window and open a terminal to check your account.

Just install the marketplace and the plugin in order, then enter the following commands in the Claude Code conversation interface:

/plugin marketplace add aqua5230/usage
/plugin install usage@usage
/usage:quota

After running, it lists how much of the 5-hour and weekly quotas (the model usage limits) for Claude Code and Codex have been used and when they reset. The plugin only calls usage-cli status --json, and runs it via uvx if it isn't installed. The command uses usage-cli to avoid name clashes with other tools, and it only runs when you type it yourself—Claude will never trigger it on its own in the background.

When you plan to write a script to pull account status, the integrating program no longer has to go to great lengths just to calculate remaining time.

Now when you run usage status --json, the output format calculates the countdown for you. Each statistics window gains a resets_in_seconds field returning the countdown until the quota resets, and each agent (an AI program responsible for a specific task) also gets an age_seconds field recording how many seconds it has been running. Both values are computed against the report's generated_at timestamp, so downstream tools receiving the data no longer need to subtract two timestamps themselves.

When using terse mode to make the AI keep things short, the shift-handoff opening message no longer sounds stiff and cold.

The system removed two lines from the prompt in terse mode (the setting that tells the AI to keep answers short). First, it deleted the strict pressuring demand to "never let it fade," because the reminder mechanism attached to every message already keeps the tone consistently concise. Second, it removed the ban on "writing extra just to be friendly," to avoid conflicting with the progress handoff summary that every user must read.

When you open the usage report in a browser or on your phone, the statistics on screen no longer line up crookedly.

In the HTML report, the four key metric cards now all keep exactly the same height. The card tracking token (the unit of text the AI processes and bills by) consumption shows only an approximate value in large type, while the exact figure and the change move to smaller text on separate lines, fixing the old problem of inconsistent line wrapping across cards. On mobile, each section also gains 16px of padding on both sides, so content no longer hugs the card edges.