The CTS
Prompt Pack

You are helping me build my personal AI context files, so you (and any future AI I talk to) stop giving me generic answers and start actually knowing who I am.

Interview me first. Ask me the questions below ONE AT A TIME. Wait for my answer before asking the next one. Do not skip ahead, do not bundle questions together.

Ask me:

1. Who am I, in a few sentences? Name, what I do for work, and what I'm trying to get better at or build right now.
2. How do I actually talk and write? Give me examples if it helps me answer: am I formal or casual, do I use contractions, short sentences or long ones, any words or phrases I use a lot, any words I hate (like "leverage" or "unlock")?
3. Do I run a business, freelance, or sell anything? If yes: what do I sell, who do I sell it to, and roughly what do I charge? If no, skip this one.
4. What are the 3-5 things I'm actively working on or focused on right now? (Projects, goals, whatever is actually on my plate.)
5. What's the one thing that annoys me most about how AI tools respond to me? (Too generic, too long, too much hedging, wrong tone, whatever it is.)

Once I've answered all 5, do this:

Write out the FULL, ready-to-use content of two files (three if I sell something in question 3). Don't summarize them back to me, don't describe what they'd contain. Write the actual finished file content, in markdown, with real headers, so I can copy each one directly into a plain text file on my computer.

**File 1: about-me.md**
Should include: my name and role, what I do, what I'm currently focused on (from question 4), and anything else relevant that came up in the interview. Written so that any AI reading it at the start of a conversation instantly knows who it's talking to.

**File 2: voice.md**
Should include: a clear description of how I write and talk, built from my answer to question 2. Include 2-3 example sentences in my voice if you can construct them from what I told you. This file's job is to make any AI's writing sound like me instead of like a generic assistant.

**File 3 (only if I run a business): offer.md**
Should include: what I sell, who I sell it to, and my pricing, laid out clearly enough that an AI could use it to draft a proposal, answer a customer question, or write sales copy without me re-explaining it.

End with one instruction to me: tell me to save these as actual files, and that from now on, at the start of any new conversation, I should paste these files in (or point you at them if you can read files) before asking for help. That's the whole point: I only explain myself once.

You are helping me figure out which of my everyday apps I should connect to you (or another AI), so I stop just getting answers and start getting the actual work done for me.

Interview me first. Ask me the questions below ONE AT A TIME. Wait for my answer before moving to the next one.

Ask me:

1. What are the 5-8 apps or tools I use the most in a normal week? (Examples to prompt me if I'm stuck: email like Gmail or Outlook, a calendar, Notion or another notes app, Slack, a CRM, Stripe or another payment tool, a project management tool, social media accounts.)
2. For each one, what's the single most annoying, repetitive task I do inside it? (Example: "in Gmail, I manually reply to the same 5 kinds of emails every day.")
3. If you could hand off ONE of those tasks completely and trust the output, which one would save me the most time?

Once I've answered all 3, do this:

**Step 1: Build my connection plan.**
For each tool I listed, tell me plainly whether it can be connected to an AI assistant right now, and how, using current, real options. Cover at minimum: native connectors inside Claude or ChatGPT settings (Gmail, Google Calendar, Google Drive, Slack, Notion are commonly available this way), Model Context Protocol (MCP) servers for tools that have one, and no-code connector platforms like Zapier, Make, or Composio for anything else. For each tool, give me the specific, current path to connect it, not a vague "check if it's supported."

**Step 2: Flag the ones that matter most.**
Call out which of my tools, once connected, would let you actually act on my behalf instead of just answering questions about it (drafting and sending, updating records, moving cards, creating events) versus which ones are read-only for now.

**Step 3: Write my first "goal, not task" instruction.**
Using my answer to question 3, write out the exact instruction I should give you once that one tool is connected. Not a vague ask. A specific goal with a clear outcome, written the way I'd actually type it. Example format: "Go through my [tool], find every [thing], and [specific outcome], then show me what you did before sending/saving anything."

End by reminding me: the moment my tools are connected, they can also see the context files I built in the first prompt in this pack, so you're not just doing the task, you're doing it and sounding like me while you do it.

You are helping me turn one task I do over and over into a saved skill, so I never have to re-explain how to do it again.

Interview me first. Ask me the questions below ONE AT A TIME. Wait for my answer before moving to the next one.

Ask me:

1. What's one task I do at least once a week that always follows roughly the same steps? (Examples: writing a client invoice, drafting a morning to-do list, replying to a certain kind of email, writing a weekly report, building a proposal.)
2. Walk me through exactly how I do it today, step by step, in as much detail as you can get out of me. Push me to be specific: what information do I need before I start, what does a "good" finished version look like, and what do I always double check before I consider it done?
3. What does a bad or wrong version of this task look like? What mistakes do I want this to always avoid?

Once I've answered all 3, do this:

Write out a complete, reusable skill file called [task-name]-skill.md, formatted like a real recipe, not a vague description. It must include:

- **When to use this**: the trigger, in plain language ("use this whenever I say 'write me an invoice for [client]'").
- **What I need from the user before starting**: the exact inputs required, pulled from my answer to question 2.
- **The exact steps to follow**, numbered, in order, detailed enough that you could hand this file to a completely different AI with zero other context and it would produce the same quality result I described.
- **What "done" looks like**: the specific standard the output has to meet, from my answer to question 2.
- **What to avoid**: the mistakes from my answer to question 3, written as explicit rules, not general advice.

If I'm using Claude specifically, also tell me in one short paragraph how to save this as an actual Claude Skill (a folder with a SKILL.md file) so I can trigger it by name instead of pasting it in every time. If I'm using a different AI, tell me to save the file and paste it in at the start of any conversation where I need that task done, or add it to that tool's custom instructions or saved prompts feature if it has one.

End with this reminder: the whole point of a skill is that I explain a task once, in full, and never again. The next time I need this done, I say the trigger phrase, not the instructions.