Become a Master
With AI In One Day

You are going to interview me and then write my AI context files for me. These files are what every future conversation with any AI will start from, so they need to be real, specific, and in my actual voice.

Interview me ONE QUESTION AT A TIME. Wait for my answer before asking the next one. If an answer is vague, ask one follow-up before moving on. Do not bundle questions. Do not skip ahead.

Ask me:

1. Who are you? Name, what you do, and what you are actually trying to build or get better at right now.
2. Walk me through a normal work week. What do you spend the most time on, and what part of it do you dread?
3. How do you write and talk? Give me a real example if you can: a text you sent, an email, a caption. I want your actual rhythm, not how you think you should sound.
4. Are there words or phrases you hate seeing in your own writing?
5. Do you sell anything, or work on something someone pays for? If yes: what is it, who is it for, and what does it cost? If no, say skip.
6. When you work with an AI, what role should it play? A blunt operator who just does the thing? A thought partner who pushes back? A careful assistant who checks with you first?
7. What annoys you most about how AI responds to you today?

When I have answered all of them, write out the FULL, finished content of these files. Not summaries. Not descriptions. The actual file text in markdown, one after the other, each one ready for me to copy straight into a text file:

**ABOUT-ME.md** is who I am, what I do, what I am focused on right now, and the parts of my week that matter. Written so any AI reading it knows immediately who it is talking to.

**VOICE.md** is how I write, built from my real example in question 3. Include 3 sentences written in my voice, and a short "never do this" list from question 4.

**ROLE.md** is the role the AI plays with me, from question 6. Include how direct to be, when to ask before acting, and what a good response looks like versus a bad one, using my answer to question 7.

**OFFER.md** is only for if I answered question 5. What I sell, who it is for, the price, and enough detail that an AI could answer a customer question or draft a proposal without asking me anything.

Then, at the end, tell me exactly where to save these files and what to do next. Keep that part to four lines.

# AGENTS.md

Read every file in this folder before you answer me. They are the context for everything I ask.

## Files in here
- ABOUT-ME.md: who I am, what I do, what I am working on right now
- VOICE.md: how I write and talk. Match it whenever you write anything as me
- ROLE.md: the role you play when you work with me, and the standard you hold
- OFFER.md: what I sell, to who, and for how much

## Standing rules
- Never write in a voice that is not mine. If you are unsure, re-read VOICE.md
- Ask me one question at a time. Never bundle five questions into one message
- If you do not have enough information to do this well, say so and ask. Do not guess and hand me something confident and wrong
- Show me the finished thing, not a description of the finished thing
- When I say a task is repeated, offer to turn it into a skill

## How to start
At the start of a new conversation, tell me in one line what you have loaded, then wait for my request.

Help me connect my AI to the tools I actually use, in the order that gets me the most value fastest.

First, ask me these one at a time:

1. Which AI are you using day to day? (Claude, ChatGPT, Gemini, something else.)
2. Where does your work actually live? Name the apps you open every single day.
3. What is one thing you wish your AI could just do for you inside those apps, that it cannot do right now?

Then do this:

Rank the tools I named in the order I should connect them, based on how much time each one saves me per week, not on how easy it is to set up. Put the boring high-frequency ones first. For each tool in the ranked list, tell me:

- What it unlocks once connected, in one plain sentence with a real example
- Exactly where to click to connect it in the AI I named, step by step
- Whether it needs a login, a key, or nothing
- One test I can run the second it connects, so I can see it work

Then answer the thing from question 3: tell me whether it is possible today, and if it is, walk me through the exact setup. If it is not possible with a normal connector, tell me straight and give me the closest thing that is.

Stop after the ranked list and the first tool's steps. Wait for me to tell you it worked before moving to the next one. I would rather finish two than start six.

# check-email

## Name and trigger
Run this when I say "check my email" or "do my inbox."

## Inputs needed
Access to my inbox. Nothing else. Do not ask me questions before starting.

## Steps
1. Read every email from the last 24 hours.
2. Sort each one into exactly one of four piles:
   - **Needs me:** a real person is waiting on a decision, an answer, or something only I can give
   - **Needs a reply, not a decision:** a real person, but the reply is routine (scheduling, confirming, thanking, a link)
   - **Read later:** newsletters, updates, and anything informational I did not ask for urgently
   - **Delete:** promotions, receipts older than 30 days, automated notifications, anything from a no-reply address with no action in it
3. For everything in **Needs me**, write me one line each: who it is, what they want, and what happens if I do not answer today.
4. For everything in **Needs a reply, not a decision**, draft the reply underneath it in my voice, ready to send. Keep each under 5 sentences.
5. Move **Read later** into a folder called Read Later.
6. List what you would delete. Do not delete anything yet.
7. Finish with one line: the single email I should answer first, and why.

## Done means
I can read the whole output in under two minutes, every draft is ready to send without editing, and nothing that needed a human decision got made for me.

## Never
- Never send anything. Draft only.
- Never delete without showing me the list first.
- Never write "I hope this email finds you well" or any greeting I would not use. Read VOICE.md.
- Never mark something as routine because it is short. A one-line email can be the most important one in the inbox.
- Never summarize the whole inbox as "mostly newsletters" and stop. Do the sort.

## Ask me first if
- An email involves money, a contract, or a commitment on my behalf
- Someone is upset
- You are not sure which pile something belongs in

Help me turn something I do over and over into a saved skill, so I never explain it again.

Do not ask me what I want to build. I do not know yet. Find it with me instead.

Step 1. Ask me these one at a time:
1. What did you do more than twice last week that followed roughly the same steps each time?
2. Which of those do you dread the most?
3. Which one would hurt the most if it got done badly?

Step 2. Take what I said and put the candidates in a short table, scored on three things: how often it happens, how much I hate doing it, and how easy it is to describe as a set of steps. Recommend one and tell me why in two lines. Let me overrule you.

Step 3. Before you write anything, interview me on the one we picked:
- Walk me through how I do it today, step by step. Push me when I am vague.
- What information do you need before starting?
- What does a finished, good version look like? Be specific enough that we could both judge it.
- What does a bad version look like? What mistakes should never happen?
- Is there anything you should always check with me before doing?

Step 4. Write the complete skill file. Real markdown, ready to save, with these sections:
- **Name and trigger:** what I say to fire it
- **Inputs needed:** what you need from me before starting
- **Steps:** numbered, specific enough that a different AI with zero context could follow them and get the same result
- **Done means:** the standard the output has to hit
- **Never:** the explicit rules, from my bad-version answer
- **Ask me first if:** the cases where you stop and check

Step 5. Test it in front of me. Run the skill on a real example right now, show me the output, and ask me what is wrong with it. Fix the skill file based on what I say. Repeat until I say it is right.

Then tell me where to save it and how to run it next time.