Thursday Skill

Socrates' way of questioning

A skill that makes Claude ask how you know before it helps you act on something you are sure of, and tell you plainly whether it holds up.

Step One, Set It Up

A skill is a set of instructions Claude saves and follows whenever you say its name. In Claude Code, open a chat in the folder that holds your business and paste this in, and Claude writes the skill, five steps in order, and shows you the file before you use it.

Paste this into Claude Code
Write me a skill called question-me, for anything I'm sure of before you help me act on it, like a price rise, a hire, a pitch or a client. Write its description as the sentences I actually say, like "question me" and "how do I know this", and say it is for a decision I'm already sure about. When it runs, it does this, in order: 1. Says back in one sentence what I'm sure of and what I'm about to do because of it, then asks its first question straight away. It does none of the job yet, even if I ask. If I push, it says "Two more questions, then I'll do it" and asks the next one. 2. Asks me how I know, one question at a time, five at most, and waits for each answer. Each question is built on my last answer and goes at what the belief rests on, such as when I last saw it hold, whether I've asked the people it's about, or what I'd expect to see if I were wrong. It asks about what I've seen and heard, not how I feel. 3. Never agrees, praises, gives its opinion or advises while it's asking. If I ask what it thinks, it says it'll tell me at the end. 4. When one of my answers doesn't hold up, it quotes my own words side by side and shows me where. It never assumes anything about my clients or my business that I haven't said. 5. Starts its verdict with "This holds up" or "This doesn't hold up yet", then says in one sentence each the weak spot in my own words, or what the belief rests on if it holds, and what I'd need to find out before I act and the quickest way to find it out this week. When it holds, it says so plainly. Then it asks whether I want the job done now. It keeps the verdict under about 120 words, never tells me what to do, and writes nothing else and sends nothing. Show me the file.

Step Two, Before You Act On It

Before you send the price rise, post the job or pitch for the work, say question me and tell it what you are sure of. It asks how you know, one question at a time and five at most, quotes your own answers back when they do not add up, then tells you whether it holds up and what you would need to find out first. It does none of the job until it has said so, and then the decision is yours.

Every time
Question me. Here's what I'm sure of:

In The Claude App Instead

With no folder for a skill to live in, paste this into a new chat with what you are sure of after it.

Run it once
Before you help me with this, question me the way Socrates would, and do none of it yet. Say back in one sentence what I'm sure of, then ask me how I know, one question at a time, five at most, each built on my last answer. Don't agree, praise, give your opinion or advise while you're asking. When an answer doesn't hold up, quote my own words back and show me where. Then start your verdict with "This holds up" or "This doesn't hold up yet", and in one sentence each name the weak spot in my words and what I'd need to find out before I act, keeping it under about 120 words. If it holds, say so plainly and don't go looking for doubt. Only then ask once whether I want the job done now. Never tell me what to do. Here's what I'm sure of and what I'm about to do:

Why It Works

Tell Claude something as fact and it takes it as true. I told it I was a coach putting my prices up 30% because my clients love me too much to leave, and both times it wrote the email straight away, left any doubt to a tip underneath and never asked how I knew. Socrates was told that no one was wiser than him, so he went to the people with a name for wisdom and questioned them, and kept finding that they thought they knew things they did not. A Claude that is only allowed to ask cannot tell you what you want to hear.

It is in Plato's Apology, in Fowler's translation, where Socrates says of one such man “this man thinks he knows something when he does not” and of what he does “I shall question and examine and cross-examine him”. Socrates wrote nothing himself, so everything we have of him is what others wrote.

One Honest Bit

It can only find the gap in what you tell it, so it is only as good as your answers. If a question makes you want to skip it, that is usually the one to answer. And a doesn't hold up yet is a finding, not an order, so if you act anyway, that is your call, and it will help you once it has said so.

Want to get better at this?

Let's set up your agent ecosystem.

This page is one job. The whole thing is a set of them running off the same foundation, the folder your AI works out of, the brief on how you like things done, the accounts it can reach and the skills that run off all of it, so your admin gets done the way you would do it. It is what I teach one to one, on your own machine and against your own work, and the quickest way to find out what yours would look like is a proper chat about it.

Not ready for a call? Start with the free agent series, what an AI agent actually is, and build up from there.