The AI Red Team Analyst skill for Claude

The problem the skill solves is a double role almost everyone plays. Whoever makes a plan also judges that same plan, and that combination produces predictable blind spots. You look for information that supports your plan and read criticism as noise. Your team is enthusiastic, so pushing back costs social capital and the doubter stays quiet. And you see the paths to success sharply in front of you, while the paths around them stay out of view. What a red team analysis does is organise the pushback your environment does not give you: someone who does not think along but thinks against, with a method instead of a mood.

It is written for anyone about to do something that costs money, time or reputation and is hard to undo. A business owner launching next month who does not know what he is missing. A team where everyone is enthusiastic, which is exactly the reason to get nervous. An investor reading a business case that the maker judged himself. The file's frontmatter lists those situations word for word as moments when Claude should pick up the skill on its own.

One thing you should know beforehand: the skill delivers dangers, not a verdict on you. It attacks the plan and leaves the person alone, and it never ends with objections only. Every main danger gets a defence you can carry out this week, with no budget you do not have. Attacking here is not a goal but a method for building your defence in time.

Confirmation drive. You have spent weeks inside the plan, so every signal that it holds feels like information and every signal that it does not feels like a detail for later. Anyone who wants to see his own assumptions wobble needs someone with no stake in keeping them standing. The Cognitive Bias Detector skill is the wider toolkit for this kind of thinking error; the red team analyst is the applied version for one concrete plan.

Group pressure. The moment a team is enthusiastic, criticism turns from a contribution into an attack on the mood. The doubter pays for that socially and stays quiet. A red team solves that by giving pushback a formal role: it is not your colleague being difficult, it is the method having its say today. Anyone who wants to run that division of roles in a team can also look at the Six Thinking Hats skill, where the black hat has the same function.

Planning optimism. You picture the route to success and see it clearly in front of you. The paths around it, where a customer gets angry, a supplier drops out or you yourself fall ill, do not appear in the estimate because you are not looking at them. That is why the file forces that look with a fixed adversary map and an assumption list of mandatory size: eight to twelve, even when the first five already hurt.

The difference between an ordinary review and a red team analysis is quickest to see as a diff, with lines drawn from the example in the skill:

Anyone who says something like that to Claude while the skill is loaded gets it automatically. The frontmatter also explicitly rules out: inventing plans that do not exist yet, attacking people, and technical security testing.

Next comes the role switch, and it is stated bluntly: once the skill is active, Claude stops defending the plan and attacks it with the means a real adversary would have. It also says this out loud to you, so you know exactly what stage the conversation is at. It then carries out seven fixed actions: naming the role switch, mapping at least three adversaries, exposing the assumptions under the plan including the unspoken ones, scoring eight to twelve assumptions on likelihood times impact, describing concrete attack vectors per adversary, writing a pre-mortem of a few paragraphs in which the plan has failed, and delivering one defensive action per main danger that you can carry out this week.

Before anything gets attacked, the skill runs through a mandatory input checklist: the plan in the form it exists in now, what it has to deliver and by when, what it costs in money, time and attention and what is irreversible, who is affected if this succeeds, what is already fixed and what can still change, the hard deadline, and what you personally find most nerve-racking. That last question is the sharpest of the set, because what you find most nerve-racking is usually the point you do not dare look at. If anything is missing, it asks in a single round and then carries on with marked assumptions.

The output always follows the same six phases: goal and scope of the attack, the adversary map, the assumption list, the attack vectors, the damage score and finally the pre-mortem with the defence. And there is a separate review mode: feed it an existing plan, business case or risk paragraph, and you first get an assumption density with quotes, a list of what is missing from it, an optimism check on timeline, conversion and cost, and a marker per section of whether a mistake there can be undone. Irreversible steps belong late in the plan, not early. The full analysis then still follows.

Anyone who wants to train probing questions as a standalone skill can turn to the Socratic Method skill NL for that; anyone who wants to learn to build a file like this themselves finds the full explanation with a template at writing your own SKILL.md NL.

The file closes with writing rules that guard the tone, and one of them sums up the whole skill: say the unpleasant part out loud, because a red team that softens things is useless. But attack the plan, never the person who made it.

The pre-mortem comes from Gary Klein. He described the technique in Harvard Business Review in 2007: move yourself to a moment in the future where the plan has failed and write the history of that failure. The difference with an ordinary risk analysis lies in the certainty. You do not ask whether it could go wrong, you establish that it has gone wrong and work out how, and that takes the brake of politeness off.

In this skill the pre-mortem is part of phase six; anyone who wants only that technique, in its full form with failure categories, early signals and kill criteria, uses the separate Pre-Mortem Analyst skill NL. If it is not the attack you are after but the assumptions themselves, the first principles thinker strips your plan down to what is demonstrably true.

The assumptions check comes from Richards Heuer. His 1999 Psychology of Intelligence Analysis describes how analysts mistake their own assumptions for facts, and the key assumptions check he later worked out with Randolph Pherson in Structured Analytic Techniques is the practical answer to it: write down every assumption under the plan, including the ones that seem so obvious nobody says them out loud, and test how solid each one is. Phase three of this skill is that technique in working form, including the requirement to mark unspoken assumptions separately. Anyone who wants to see how such an assumption forms in your head, from observation to conclusion, finds that step model in the Ladder of Inference skill.

The variants come from the same field. The file names four you can bring in alongside it: devil's advocate, where one person systematically takes the opposing position; Team A versus Team B, where two groups get the same information and reach a conclusion separately; analysis of competing hypotheses, where you set several explanations side by side and look for evidence that rules one out; and the what-if analysis, where you take an event as settled and reason backwards to what had to be true beforehand. Which form fits when is covered further on, under the limits.

Even without installing the skill you can test your own plan with these principles: name who stands to gain if it fails, write your assumptions down as claims that can be true or false, and treat unanimous enthusiasm as a signal to probe further rather than as proof.

After that, using it is simple: paste your plan and ask for an attack. The more complete your input, the sharper the vectors; whatever is missing, the skill asks for in a single round. New to skills? Read what Claude skills are NL first, and the installation guide for Claude, Codex and Cursor NL. The wider explanation of working with AI is in the knowledge base.

If time is short or the plan is still early, it runs only the pre-mortem: that takes a quarter of an hour and delivers the most per minute. If the plan is technical or financial and adversaries barely matter, such as with an internal system migration, the key assumptions check is enough. If there is a team with two camps, Team A versus Team B is the form: both camps get the same information and reach a conclusion separately. And if the question is not what could go wrong but what is already going on, for example with falling sales that could have several causes, it switches to analysis of competing hypotheses.

There are also situations where you are better off leaving it alone. It does not invent plans: something has to exist to attack, otherwise there is only a blank page and an opinion. It does not attack people, so a conflict with a business partner does not belong here. And it is explicitly not a technical security test: anyone who wants to know whether their systems are safe needs a different discipline. For a personal decision with no external parties, it skips the adversary map and replaces it with you yourself at different points in time, because even without a competitor you can be your own most dangerous adversary.

And the most honest limit: a red team analysis does not make your plan good, it makes it tested. If the attack turns up little, that says something about your preparation, and if it turns up a lot you still have the choice of what to do with it. It protects you against blind spots, not against decisions you make with your eyes open. If, after the attack, you still need to work out what kind of decision this actually is and which approach fits it, the Cynefin Decider skill is the logical next step.

A red team analysis works on any plan, so also on the assumption that customers will find you anyway. If you would rather test that assumption with real data than a gut feeling, the free SEO scan shows in a few seconds where your site stands. And if you want to talk further about what else AI can do for your business, from standalone skills to full automation, we are happy to do that in a conversation.