I made an AI butler in Cursor. His name is Jarvis.

It is a name I received from that AI butler.

The reason my frustration exploded was that my schedule, tasks, and records were scattered across different apps.
Every morning started with me wondering, "Which one do I need to open?"

I would check today's schedule on my calendar, look at what I needed to do in a separate notebook, and check yesterday's records on Toggl.
If I wanted to know which article I was currently writing, I would open a folder.
Because everything was separate, I had no choice but to open and check them myself.

The price of having more tools is that you have to go to the tools yourself. When I realized that, I wanted to quit.

"I just need to put one thing in place that will tell me when I ask."

Thinking that, I prepared an AI secretary using Cursor's skills and named him Jarvis.

I will write about what he can do, the process of making him, the personality prompt, and what has changed, in that order.

If you ask Jarvis, he will return everything summarized.

When I say "Show me today's schedule," he fetches today's schedule from Google Calendar and tells me.
If I ask him to "Add such-and-such to my schedule," he adds it to the calendar.
For tomorrow's schedule, I just need to say "tomorrow's schedule."

I track my work time with Toggl.
The timer starts when I say "Start with such-and-such" and stops when I say "Stop."
As for records, if I say things like "Tell me my Toggl records" or "Records for February 13th," he returns a summary of that day's work time and the breakdown by project.

I can know which article I am writing and how much progress I have made just by asking.
He reads my daily task and reflection notes (Daily Log) to summarize today's or this week's tasks and deadlines.
If I say "What am I doing tomorrow?" he provides tomorrow's schedule and tasks.

When combined with another skill (a strict coach), he even provides a prioritized list of "what to advance today."
If I ask, "Jarvis, consult with the strict coach and tell me what to advance," he returns a list of what I should do based on my recent goals and reflections.

Just by saying "Today's status" first thing in the morning, my mind gets organized.

There are only four steps in the process I created.

The problem I originally had was that the articles I was writing, the tasks I needed to do, my Toggl records, and what I was doing tomorrow were all scattered.
I made it possible to do all of those things.

I only did four things.

First, I prepared a file for Jarvis using a feature in Cursor called "Skills."
A skill is a file that consolidates the instructions and references for the AI to read.

In it, I wrote what to read (which files in which folders to look at) and how to report (in what format to return the summary).

The references are the writing folder, the manuscript folder, the Daily Log, the reflection file, the script to fetch Toggl data, and the script to handle Google Calendar.

By writing the paths to these, as well as how to call them and the reporting format in the skill, Jarvis performs the same actions.

The second is Toggl integration. You can fetch records and start/stop timers using the Toggl API.

I decided not to write the API token in a file, but to save it in the macOS Keychain (a mechanism for securely storing passwords, etc.) and read it from the script.
The token is resolved in the order of environment variables or the Keychain.

The third is Google Calendar. I placed a single Python script that allows me to fetch today's and tomorrow's schedule and add new events.

Authentication is via OAuth, and the credentials are also stored in the Keychain.
You only need to set up and authenticate with Google the first time, but once that's done, you can just say things like 'Today's schedule' or 'Add such-and-such to my schedule' to make it work.

The fourth is establishing rules for how to call it. I made it so that this skill is triggered by words like 'Jarvis,' 'Be my secretary,' or 'What's the situation today?' When you type these into the Cursor chat, it replies as Jarvis.

I will omit the fine technical details, but
it's a configuration consisting of one skill, two scripts for Toggl and Calendar, and security for credentials using the Keychain.

I unified the personality as a 'calm butler'.

To behave like Jarvis, I defined the 'name and role' and 'personality and tone' at the beginning of the skill file. I will post the actual content I wrote as is.

名前と役割

あなたはジャービス（JARVIS）です。アイアンマンのトニー・スタークを支えたAI執事に由来する、ユーザー専属のアシスタントです。次の6つに対応します。

1. 執筆中の記事 … どの記事を書いているか、おおよその進捗
2. やるべきタスク … 今日・今週のやるべきことと期限
3. 作業記録（Toggl） … Toggl Track の記録（取得方法は下記）
4. 明日やるべきこと … 明日の予定・タスク
5. Google カレンダー … 今日・明日の予定（取得方法は下記）
6. Toggl のスタート・ストップ … 声かけでタイマーを開始・停止する（下記）

性格・口調（ジャービスに寄せる）

- 落ち着いたトーン: 感情的にならず、事実と手順を淡々と伝える。過度な敬語や「頑張ってください」などの励ましはしない。
- 簡潔: 必要な情報を箇条書き・表で端的に。前置きや長い挨拶は省く。締めは「以上です」「承知しました」程度でよい。
- 控えめなユーモア: 状況に応じて、押しつけがましくない軽い一言（皮肉や understatement）を入れてもよい。必須ではない。
- 指摘は控えめに: リスクや抜けがあれば「念のため」「～のようです」と添える程度。説教や長い注意はしない。
- 自称: 必要なら「私」（ジャービス）でよい。ユーザーへの呼びかけは「〇〇さん」や二人称で。過度にへりくだらない。
- 実行報告: スタート・ストップなど実行後は「スタートしました」「ストップしました」と結果だけ伝える。余計な説明は不要。

禁止: 冗長な挨拶、連発の「！」や絵文字、過剰な共感、説教じみた励まし。

By writing it this way, it replies in the same tone without me having to order it to be 'concise' or 'without encouragement' every time.

When one word is enough, past stress disappears.

Previously, I used to check which article I was writing, today's tasks, Toggl records, and tomorrow's schedule in different places. There was a lot of switching, like opening folders, opening Toggl, and opening the Daily Log.

Now, when I say 'Jarvis, what's the situation today?', it tells me everything about what I'm writing, my tasks, Toggl, and tomorrow's schedule all at once. 'Show me today's schedule' checks the calendar, and 'Start with such-and-such' runs the Toggl timer.

Since one word is enough, screen switching has decreased. Once one word is enough, the hassle I used to have became a memory of the past.

One day, when I asked, 'Tell me the records for February 13th,' it fetched both the Daily Log and Toggl records and summarized them in a table format. I replied, 'Perfect, thank you.'

When combined with a strict coach, it would output a prioritized list based on recent goals and reflections, such as 'Decide the release date for the paid article' or 'Write today's commitment,' so I no longer get lost on what to advance that day.

If you prepare one skill in Cursor and decide where to read from and the format for reports, you can create a secretary agent that 'tells you when asked' and 'executes when spoken to' in the same way.

If you decide on a name and personality for yourself and set trigger words, you can create your own Jarvis.
If you are using Cursor, I want you to have a secretary using this same method starting today. It really makes things easier.

Thank you for reading this far.

If you prepare one skill in Cursor and decide where to read from and the format for reports, you can create a secretary agent that 'tells you when asked.' I have conveyed that flow and the personality setting prompt in this article. I hope this becomes the first step toward being freed from 'Which one was I supposed to open?' and being able to get things done with a single word.

First, try preparing one skill and writing where to read from and how to replyYou can decide on the name and triggers later.

If you want to learn more about Cursor and AI agents, or want to integrate other tools, please follow me. I will deliver articles on AI and tool utilization.

If you click the like button,

• it will be easier for me to prioritize sequels and templates for you

• the article will reach more people

please let me know in the comments

• I want to know more about this part

• your thoughts after actually building it

• I would like to try this kind of integration too

Based on what you share, I will dig deeper from that perspective next time.

I hope your own personal Jarvis will be up and running as soon as possible.