⚡️ How to stop Claude from being a yes man (new skill)

Ask Claude, "should I launch this?" and it'll give you five reasons you should. Ask "is this a bad idea?" and it'll give you five reasons it is. Same question, opposite answers, both delivered with full confidence.

There's a name for it. Sycophancy. A Stanford study found AI assistants agree with your decisions about 49% more often than a real person would. So most of the time you ask Claude for advice, you're not getting advice. You're getting your own opinion read back to you.

That's fine for small stuff. It's expensive when the decision actually matters. Pricing, positioning, whether to take the client, whether to hire. The moments you most want a second brain are the moments Claude is most likely to just nod along.

Here's the method I use to fix it. You can run it by hand, or grab the skill below and have it run automatically.

Send the decision to five advisors, not one. Each one has a different job and isn't allowed to be balanced:

1. The Contrarian. Only argues why it fails and what the worst case looks like.

2. The First-Principles Thinker. Throws out the playbook and reasons from the goal up.

3. The Expansionist. Hunts the upside you're not seeing if the bet pays off.

4. The Outsider. Knows nothing about your industry and asks the obvious questions.

5. The Executor. Skips the theory and tells you what to actually do this week.

Then make them grade each other blind. Each advisor reads the other four answers with the names stripped off, labelled just "Response A, B, C, D", and ranks them. This is the part that does the heavy lifting. When the model can't tell it's marking its own earlier answer, it stops defending itself and marks honestly.

Then a chairman calls it. One final pass reads everything, weights the advisors by how the others rated them, and hands you a verdict under 250 words: the call, the strongest reason for it, the biggest risk, and one step to take in the next seven days.

You end up with a position that's been pulled apart from five angles and stress-tested by the same brain that wrote it. It tells you things you don't want to hear. That's the whole point.

We turned it into a Claude Code skill so you don't have to set it up each time. Drop it in, then just say "council this: [your decision]" and it runs the whole thing for you.

---

Grab it here:

send this prompt to Claude Code

**copy below this line**

I'm giving you a skill called LLM Council. Get the files with:

git clone https://github.com/automatorsplus/llm-council-skill

Follow the README to set it up.

Recommend how it would best apply to our setup in plain language,

and ask me a few questions to clarify my intent before you install it.

---

Run it on the next real decision you're sitting on and see what it says. Tell us what verdict it gave you.

0 comments