Free Prompt

Onboard Your AI to Your Job

You'd never hand a new hire a laptop and zero context, but that's how most people use AI: every conversation starts from a blank slate. This is one prompt that fixes it. Paste it into Claude Code once and it interviews you about your role, your team, your projects, and how you like to work. Then it writes that down in files it reads at the start of every future session, so it remembers who you are without you re-explaining anything.

By Preston Chin · ~3 min read · one prompt, copy-paste ready

What this is

Claude Code reads a file called CLAUDE.md at the start of every session. Whatever's in it, the assistant treats as standing instructions. Most people leave it empty, which is why their AI feels generic no matter how good the model is.

The prompt below runs a structured onboarding interview, the same way a good manager would onboard a new employee, except you're the manager and the AI is the hire. It asks about one area at a time, checks its understanding after each section, then writes everything into three context files. From that point on, every conversation starts with your AI already knowing your role, your stakeholders, your deadlines, and your writing style.

The prompt

Open Claude Code in the folder where you want your work assistant to live (a folder like ~/work-assistant works fine), paste this in, and answer its questions. The interview takes 10 to 15 minutes and you only do it once.

The setup prompt 
1. Interview me about my job. Ask me questions one section at a time — don't dump everything at once. Cover these areas:
   - My role, company, and what I actually do day-to-day
   - My team, who I report to, key people I work with
   - What I'm working on right now — top projects, deadlines, challenges
   - How I like to communicate — email style, meeting prep, pet peeves
   - My goals — this quarter and longer term
   - What tools I use and what tasks eat the most time

2. After each section, summarize what you heard and ask if I want to add anything before moving on.

3. Once we're done, create these files:
   - CLAUDE.md — my main instruction file. Include a session start protocol that loads my context automatically, my communication preferences, a correction log where you track mistakes so you don't repeat them, and a section for my active projects.
   - .claude/rules/my-role.md — detailed role context: org chart, stakeholders, team dynamics, current initiatives with deadlines.
   - .claude/rules/my-style.md — how I write and communicate: email tone, deck conventions, phrases I use, phrases I hate.

4. Make the system self-improving. Add this rule to CLAUDE.md: after every conversation, if I correct you or share new context about my job, ask me if you should save it to my profile. Update the right file automatically.

5. Start the interview now.

What it builds

When the interview ends, you'll have three files. They're just plain markdown, so you can open and edit any of them by hand whenever you want.

CLAUDE.md is the brain. It holds your communication preferences, your active projects, and a correction log. That last one matters more than it sounds: every time the AI gets something wrong and you correct it, the mistake goes in the log as a rule, so it doesn't happen twice.

.claude/rules/my-role.md holds the org context: who you report to, which stakeholders matter, what's shipping this quarter and when. This is what lets it draft an email to your VP differently than a Slack message to your team.

.claude/rules/my-style.md captures how you actually write. Email tone, deck conventions, the phrases you use, the phrases you hate. Without this file, everything the AI drafts sounds like AI. With it, drafts start sounding like you on a good day.

What changes after setup

The next time you open Claude Code in that folder, ask it to draft a status update or prep you for a meeting. Notice what you didn't have to do: explain who the meeting is with, what the project is, or how formal to be. It already knows.

Step 4 of the prompt is the part people underestimate. It makes the whole setup self-improving. Whenever you correct the assistant or mention something new about your job mid-conversation, it asks if it should save that to your profile and updates the right file. So the assistant you have in month three is meaningfully sharper than the one you set up on day one. I've been running a version of this system for my own company for months, and the correction log alone has killed whole categories of repeat mistakes.

One tip: be honest in the interview, especially on the pet peeves and the tasks that eat your time. The more specific your answers, the less generic every future draft becomes.