The AI Pre-Mortem Analyst skill for Claude

The problem this skill solves is uncomfortable but familiar: the brain that came up with a plan is the worst possible brain to judge that plan. You have put time into it, you want it to go ahead, and every doubt feels like a betrayal of your own work. Dutifully ask yourself what could go wrong, and out comes a shortlist of things like competition, busyness and a disappointing market. Vague enough that you never have to act on it. The pre-mortem turns the question around: the plan has already failed, reconstruct why. That forces stories with causes, data and names, and those you can actually use.

It is meant for anyone facing a decision with real stakes attached: a launch, a pivot, a hire, an investment, a large expense or the step into a new market. The frontmatter names those situations by name as the moments where Claude should pick the skill up on its own. It is part of the skill library we make available for free through our AI and automation service, with no account and no sales email afterwards.

One thing you should know beforehand: the skill only protects you if you give it real input. Anyone who answers the six flags from step 1 vaguely gets a vague pre-mortem back, and the file itself says exactly that: a vague pre-mortem is a worthless pre-mortem.

A pre-mortem asks the question looking backward. It is six months after the deadline, the plan has failed, and you write the report for a friend considering the same thing. According to the file, that reversal bypasses the optimistic filter and activates episodic memory and narrative reasoning: you stop inventing probabilities and start reconstructing a story. And stories need causes, not categories.

You see the difference most sharply in the wording. A risk analysis accepts a line like the market could disappoint. That kind of sentence does not survive the pre-mortem, because in a reconstruction you have to say how the market disappointed, when you could have seen it and why you did not. A hard language rule in the file covers exactly that: no vague risks, always specific, with the example that a competitor might do something is banned and competitor X launches a comparable feature in Q3 because they are already investing in it is the minimum.

In diff form, with the launch plan from FIG.01 as the case: what the vague version says and what the pre-mortem makes of it.

That is exactly why this technique also works so well as a skill. The reversal is simple, but keeping it up consistently is not: after three failure modes your brain wants to go back to defending the plan. Claude does not have that problem. It works through the seven categories, including the uncomfortable ones, and keeps going until the list is complete.

After that the file defines the role in one sentence: tear a plan apart before it gets carried out. The core is a fixed eight-step way of working. Step 1 maps the plan across six flags: the decision in one sentence, the timeline, the resources going into it, the definition of success, the definition of failure and the stakeholders. If a flag is vague, Claude has to push for specificity. Step 2 is the time travel: a mandatory sentence spelt out in full, with the date and a concrete consequence built on your own stake in it. The file calls this the critical emotional trigger: you are meant to read it and pause for a moment.

Step 3 generates at least 15 failure modes across the seven categories from FIG.02, with at least two per category that fit your plan specifically. Step 4 scores each failure mode on likelihood and impact, both from 1 to 5, and sorts by the product of the two.

Step 5 is, according to the file, more important than the failure points themselves: for each critical risk an observable early signal, a metric to track and a threshold at which you step in. Step 6 sets prevention alongside mitigation for each risk, with the honest note that external shocks can only be dampened, not prevented. Step 7 sets kill criteria: the points agreed in advance at which you stop. And step 8 closes with a report in a fixed shape: the plan in one sentence, the five biggest risks, the five early signals, three recommended adjustments, two kill criteria and an honest verdict: go ahead, adjust or stop.

What stands out is how many rules are about tone. The instruction prescribes: cold and concrete, no pep talk, no this is a great plan. The role is not to motivate the user but to protect the user against their own optimism, and if the plan is bad that has to be stated. The language rule is hard too: short sentences, no vague risks, always specific enough to act on. The file closes with five things the skill explicitly does not do, more on that in section 04, and with the example you saw worked out in FIG.01.

If you want to learn to build a file like this yourself, the full explanation with template is in writing your own SKILL.md NL.

The second name in the file is Daniel Kahneman, who cites the pre-mortem in Thinking, Fast and Slow. His point, as the skill summarises it: it is one of the few techniques that genuinely counters overconfidence while staying practical enough to run within half an hour. Overconfidence here is not a character flaw but a systemic one: planners structurally underestimate how long things take and what they cost, and more experience barely fixes that. A procedure that forces you to look back from the failure does, at least partly.

Why the reversal works, Klein described using what researchers call prospective hindsight: acting as if a future outcome is already certain makes it easier to find explanations for it. Ask what could go wrong, and your brain has to weigh probabilities, and it does that badly the moment it believes in something. Ask why it went wrong, and your brain gets to do what it is good at: reconstructing a story. What the file automates is exactly that shift, including the mandatory sentence that forces it.

Anyone who wants to put this kind of counter-thinking to broader use finds three neighbours in the library. The Red Team Analyst skill actively attacks your plan instead of reconstructing it. The Inversion Thinker skill starts from the question of how you would guarantee failure and works back to what you therefore need to avoid. And the SWOT analysis skill sets the risks in the wider context of strengths, weaknesses and opportunities. Same family, different angle of attack.

Even without installing the skill, you can run the ritual with your team tonight: write the time-travel sentence on a board, give everyone ten minutes to write down failure reasons in silence, and only collect them afterwards. With the skill it becomes more thorough and more repeatable, but the technique is and remains everyone's.

After that, using it is simple: paste your plan and ask for a pre-mortem. The more honest the input, the harder the analysis. If you are new to skills, read what Claude skills are NL first and the installation guide for Claude, Codex and Cursor NL; if you want to build your own variants after that, writing your own SKILL.md NL will help you on your way. The wider explanation of working with AI is in the knowledge base.

There are also situations where you are better off leaving it alone. Small, reversible everyday choices are one: working through eight steps for a decision you can undo tomorrow for free is overhead with no payoff. A project that has already failed is a second: that is post-mortem work, and the file explicitly rules it out. And the skill is no substitute for due diligence, legal advice or an accountant: it finds the weak spots in your reasoning, not the mistakes in a contract.

One more honest limit: a pre-mortem does not make your plan better, it makes the weak spots visible. What you do with that is up to you. Anyone who waves away every risk found has lost half an hour; anyone who reads every session as advice to stop never launches anything again. The tool works for people willing to adjust their plan and then still go ahead. The go/no-go structure is deliberately built on that: adjust is the middle answer, and in practice also the most often useful one.

A common failure mode in marketing plans, incidentally, is one you can already measure today: the plan leans on a website nobody can find. The free SEO scan shows in a few seconds where your site stands, before you build a launch on top of it. In the guide eight decision models compared NL you can see how this model relates to the others. And if you want to talk further about what else AI can do for your business, from stand-alone skills to full automation, we are happy to do that in a conversation. Feel free to give that conversation a pre-mortem first too; the Quote Writer skill stands ready for what hopefully comes after.