Devil's Advocate attacks your plan from nine sides before you commit, puts the biggest risk first, and hands you the cheapest test to settle each risk.
Ask an AI to review your plan and it finds something to praise, adds a soft caution, and ends with encouragement. That is fine when you want a draft. It fails when you want a decision.
The cost arrives later. A weak assumption survives review because nobody attacked it. The plan proceeds. Six months on, the review after the failure names a risk anyone could have called on day one.
Nine views
One reviewer sees one kind of risk
Each view asks a single question. None of them is trying to help you.
Attacker
How do I exploit this for gain or damage?
Competitor
How do I beat this, or make it irrelevant?
Skeptic
Which claim here has no evidence?
Pessimist
What is the worst outcome that could still happen?
Contrarian
What if the accepted wisdom here is wrong?
Auditor
Which step has no owner, no record, or no control?
Customer
Why would I not adopt this, or stop using it?
Operator
What breaks on day 30, when the person who built it has gone?
Regulator
Where does this break a law, a rule, or a contract?
The Regulator view names risks to raise with your own lawyer. Nothing here is legal, financial, or professional advice.
The order matters
What happens when you run it
01
It states your plan at its strongest
Attack a weak version and you dismiss the critique. So it builds the best case first.
02
It runs all nine views
That is the default. Each view writes its findings independently.
03
It merges and orders the findings
Duplicates collapse into one. The biggest risk goes first.
04
It cuts anything without a test
A risk you cannot check is just worry.
05
It answers in three parts
Every full review prints all three, in order. They are named below.
06
It asks what would make you stop
If you cannot answer, that becomes the top finding.
Three parts
What comes back
A full review prints all three, under these headings, in this order.
01What could go wrong
The findings, biggest first. Each one names the assumption, what happens if it is false, the evidence, and the cheapest test that settles it.
02How someone could abuse this
The routes in, if this is something a person could misuse for their own gain. When nobody could, it says so in one line rather than inventing an attacker.
03If this had already failed
Written from a date in the future, in past tense, as though the plan has already failed. Each cause in that story becomes a finding.
Free · No signup · Needs a Claude account
Get the skill
One skill. It works in Claude on the web, in Claude Code, and in Cowork. You need a Claude account with code execution switched on.
devils-advocate.zip · 8 KB Install takes about two minutes.
Prefer to read it first? Open SKILL.md. It is the whole skill, and it is plain English.
Download the ZIP.
Open claude.ai and click your initials, bottom left.
Click Customize, then Skills.
Click +, then Create skill, then Upload a skill. Choose the ZIP.
Turn the toggle on.
Skills need code execution switched on first, under Settings, then Capabilities. On a Team or Enterprise plan an owner turns it on for everyone. Anthropic moves these menus from time to time, so if the steps above do not match what you see, their own install guide is the current one.
Paste this into your terminal. It takes the file straight from here, so there is nothing to download.
Claude Code picks it up in the session you already have open, unless the command above had to create the skills folder. Restart it in that one case. On Windows, run the command in Git Bash or WSL.
Install it on claude.ai using the steps in the first tab.
Start a new Cowork session. The skill is already there.
Cowork loads the skills your account has enabled, and it reads them when a session starts. A session you already had open will not see it.
Then paste a plan and type poke holes in this.
Not using Claude?
The skill switches itself on inside Claude. Everywhere else, paste it in. This version works in ChatGPT, Gemini, Copilot, and anything with a chat box. Copy it, paste it at the top of a new chat, then paste your plan under it.
