How to Build an AI Advisor Skill From a One-Page Goals Doc

How do you build an AI advisor that checks decisions against your own goals?

Write a one-page plan in a Google Doc: your top goals for the year, what you are not doing, what gives you energy, what drains you, and your non-negotiables. Add a short section telling the AI how blunt to be. Then give Claude the link and ask it to create a reusable skill (a saved set of instructions for one job) that reads the doc before it answers. Keep the doc human-written. The AI can suggest edits, but it never makes them.

That is the whole setup. Product manager turned creator Peter Yang uses this pattern, and a demonstration of a simple version took about 30 seconds to build. The value comes from the doc, not from clever prompting.

What goes in the one-page goals doc?

The doc is what the AI measures every opportunity against, so it needs to be specific. In the setup Yang described, the page covers what he wants this year, what gives him energy and what drains him. It is not only business. Family priorities go on the page too.

A simple version has these sections:

Specific beats broad. "One trip per month" and "March 7 through 10 is blocked" give the AI facts to check. "Maintain work-life balance" gives it nothing.

Why should you be the only one who edits the doc?

If the AI writes your plan, you are measuring decisions against its guess about what you want. That is circular. Yang calls the doc his "slop-free zone," and the example version puts a line at the very top: "I only edit this doc. AI can suggest changes, never make them."

This is a good rule for a practical reason. Plans should change when you decide they change, not because a chat drifted in a new direction. When the AI thinks a goal is out of date, it can say so in its answer. You then decide whether to open the doc and edit it yourself.

Write that rule into the doc itself, and repeat it in the skill instructions, so the AI sees it every time it reads the page.

What should the "how I want advice" section say?

AI models tend to be agreeable. If you ask "should I do this?" and the answer is a polite "it depends," you have learned nothing. The demonstrated doc includes a short section that tells the AI not to sugarcoat things and to say what you need to hear, not what you want to hear.

You can write yours in your own voice. A few lines are enough:

• Give me a clear yes or no first, then the reasoning.
• If a request breaks something I wrote in this doc, name the exact line.
• Do not soften bad news.
• If you are missing information, ask me instead of guessing.

How do you turn the doc into a skill?

A skill is a saved set of instructions for one job, so you do not have to explain yourself each time. In Claude, the setup the creator showed relied on the Google Drive and Google Docs connectors being turned on, so Claude could read the doc from its link.

1. Write the one-page doc and add the "I only edit this doc" line at the top.
2. Copy the doc's URL.
3. Check that the Google Drive and Google Docs connectors are enabled in Claude.
4. Paste the link into a chat with a prompt asking Claude to create the skill.
5. Save the skill and give it a name you will remember, like "advisor."

Here is a template you can adapt. It is a starting point, not the exact prompt from the demonstration.

Create a skill called "advisor."

Every time I use it, first read my goals doc at [paste link].
Treat that doc as the source of truth about what I want this year.

When I describe an opportunity or decision:
1. Check it against my top goals, my "not doing this year" list,
   what gives me energy, what drains me, and my non-negotiables.
2. Follow the "how I want advice" section in the doc.
3. Start with a clear verdict, then quote or name the specific
   lines in the doc that drove it.

Never edit the doc. If you think it should change, suggest the
edit in your reply and let me decide.

In the demonstration, the skill was also connected to the creator's YouTube data, which gave it extra context. That is optional. Start with the doc alone.

How do you test whether the advisor works?

Use a real decision where you already know the constraints. The demonstration used a conference invite: "I've been asked to speak at a conference in March, two days of travel. Should I do it?"

The answer came back in about a minute, and it was clearly drawn from the doc. The short answer was no, with reasons tied to the plan. There was already a trip blocked for March 7 through 10, so a second March trip would break the one-trip-per-month limit. Speaking also meant preparing a talk, and the plan listed keynotes as something that trains you.

That is the result to look for. A good answer cites your own words back to you. A generic answer ("speaking can raise your profile, but consider your workload") means the AI is not using the doc, and you should check the connector and the skill instructions.

After the first test, tighten the doc. Wherever the advisor got something wrong or missed a constraint, add a line. The creator's own advice was to build something, see what it gets wrong, and then tighten it.

How often should you run the advisor?

Less often than you think. Yang runs a separate chief of staff skill every day, which looks at what needs attention right now. The advisor is for direction: are you headed where you said you wanted to go? He runs it about once a month, and says anything more frequent for long-term questions is too much.

You can also use it as a sounding board when you are frustrated or confused about something, which he does too. Treat that as a conversation aid, not a verdict. He said himself he is not sure how healthy it is to lean on it that way.

What are the limits of this approach?

• It only knows what you wrote. If your doc is vague or out of date, the advice will be too.
• It can agree with you in nicer words. Keep the blunt-advice section and test with decisions where you already suspect the answer is no.
• It is not a replacement for people. It checks a decision against your stated plan. It does not know what your manager, partner or friends would say.
• Skills are portable, but connectors vary. Yang says he uses the same skills across ChatGPT, Claude Code and Grok Bot. How each tool reads a Google Doc depends on its own connectors, so check before assuming it works.

What does this mean for you?

If you are a manager, freelancer or anyone who gets a steady flow of requests, invitations and "quick asks," this is a cheap way to make your priorities visible. Write the page first. That exercise alone, deciding what you are not doing this year, is useful even without AI.

Start small:

1. Spend 30 minutes writing the one-page doc. Be specific about limits and dates.
2. Build the skill and test it on a decision you have already made. See whether it agrees.
3. Fix the doc where it misses.
4. Put a monthly reminder on your calendar to run it and review the doc yourself.

You do not need the full system of daily briefings, councils and editors that Yang runs. One recurring task that saves you time is enough to start.

Could you build a decision log to go with your advisor?

The advisor gives you an answer, but nothing records what you decided or why. If you want a monthly review with real history behind it, you could describe a small private tool to Remy: a form where you paste an opportunity, log your yes or no and the reason, and tag which goal it touched. Remy builds a full-stack app from a plain-English description and deploys it, so a decision log, with a page that shows your past decisions by goal, is a reasonable first project. It would sit beside your goals doc, which you would keep writing yourself.