Stress-Test Any Plan in 60 Seconds With Claude

What do Google, Goldman Sachs, and Procter & Gamble do before major launches that 99% of founders skip?

They run a pre-mortem.

The idea is simple. Before you ship, you imagine the launch has already failed. Six months from now, the numbers came in bad. You are writing the post-mortem. Walk through it: why did it fail?

Psychologist Gary Klein published the methodology in Harvard Business Review in 2007 (the article is titled "Performing a Project Premortem"). Daniel Kahneman endorsed it in Thinking, Fast and Slow as one of the most effective ways to combat overconfidence in decision-making. Researchers Deborah Mitchell (Wharton) and J. Edward Russo (Cornell) named the cognitive mechanism behind it in a 1989 paper: prospective hindsight. The brain is significantly better at explaining past events than predicting future ones. By placing the failure in the past, you generate more specific, more honest risk assessments than any "what could go wrong" brainstorm ever produces.

For most of history, running a pre-mortem required a team workshop. Four to eight people, anonymous sticky notes, a trained facilitator, sixty to ninety minutes. Solo founders could not run one at all. They just shipped, prayed, and ran post-mortems after.

That barrier is gone. Claude can now run the entire pre-mortem on you, solo, in 60 seconds. Same methodology. Lower cost.

This is not a prompt hack. This is a research-backed decision science discipline you can run on any plan in the time it takes to make coffee. I run it before every decision worth more than a week of my time.

If you have seen the basic "ask Claude what could go wrong" prompt floating around, that is the surface. The rest is below.

The 60-second setup

Copy this into Claude. Swap your plan into the brackets. Send it.

You are my premortem partner.

Here is my plan:
[PASTE YOUR PLAN HERE]

Imagine it is 6 months from now. The plan failed. Walk me through exactly why, in 3 categories:

1. REAL RISKS. The failures most likely to actually happen, with evidence from my own plan. For each: is it launch-blocking or can it wait? What is the single concrete action that mitigates it? What hidden assumption is it riding on?

2. FAKE RISKS. The stuff that sounds scary but probably will not matter. Tell me what to stop worrying about.

3. UNSPOKEN RISKS. The stuff I am avoiding because it is uncomfortable. People issues, politics, decisions I have not made.

End with two things:
- The single biggest hidden assumption my plan quietly depends on
- A revised version of the plan with the weak spots fixed

Be specific. Cite my own words back to me. No generic advice.

That is the basic version. Works on free Claude. For higher-stakes decisions (more than $10K in time or money on the line), switch to Opus or Sonnet. Smaller models hedge on the Unspoken Risks category. Opus says the quiet part out loud.

The four-phase methodology (for the deeper version)

The 60-second prompt is the surface. The full four-phase version is what Klein's research mapped out, adapted from his team workshop format for solo use. Takes 30 minutes. Surfaces three to five times more useful signal.

Phase 1: Set the scene (3 min). Do not just imagine the plan failed. Place yourself six months from now, writing the post-mortem your investors will read next week.

Phase 2: Generate risks (10 min). Exhaustive list, no classification yet. Target 10-15 risks.

Phase 3: Classify (10 min). Sort into Real, Fake, Unspoken. The Unspoken ones are the most valuable.

Phase 4: Mitigation (7 min). Every Real Risk gets a concrete action, an owner (probably you), and a decision date.

The four key prompts

Phase 2: Exhaustive list

Here is the plan: [PASTE PLAN]

It is 14 days after launch. The product failed. Generate at least 10 reasons it failed, across these surfaces: technical, customer adoption, team execution, competitive timing, external events, decision-making, and the thing nobody on the team wants to bring up.

Phase 2: Adversarial critic

Now do it again, but imagine the postmortem is written by my most cynical advisor. Someone who has watched me ship before, knows my patterns, and is willing to call out the failure modes I personally tend to repeat. Five risks in their voice.

Phase 3: Classify plus surface Elephants

Sort all the risks above into: REAL (evidence-backed, plausible failure scenario), FAKE (sounds scary but unlikely or low impact), or UNSPOKEN (politically uncomfortable, the stuff teams avoid saying out loud).

For UNSPOKEN specifically, name the politically uncomfortable risk I am not asking about. The thing I do not want to write down because it would mean confronting a specific person, decision, or story I have been telling myself.

Phase 4: Hidden assumption and revised plan

Across all risks above, what is the single biggest hidden assumption my plan quietly depends on? The thing so obvious I forgot it was even an assumption. If that assumption is wrong, what changes? Give me the revised version of the plan with that weak spot fixed.

The last prompt is the one I save into my decision archive.

Three meta-moves nobody else is teaching

Move 1: The future-you reframe. Open with "I am me, six months from now. The plan failed. I now know exactly why. Interview me." Then let Claude interview future-you. Answers get dramatically more specific because the cognitive frame is autobiographical instead of hypothetical. Your brain stops generating cautious abstractions and starts generating remembered details.

Move 2: Multiple time horizons. Run the same premortem at 6 months, 12 months, and 24 months separately. Six-month risks are execution risks (build broke, team burned out, launch missed). Twelve-month risks are market risks (customer never adopted, competitor moved faster, pricing was wrong). Twenty-four-month risks are strategic risks (the entire premise was wrong). Most founders only run the 6-month version and get blindsided by the 24-month risks two years later.

Move 3: Adversarial framing. Have Claude play three stakeholders against your plan, separately: your toughest competitor, your most skeptical board member, your most demanding customer. Each generates different risks that your own premortem mood cannot access. The competitor surfaces strategic gaps. The board member surfaces resource gaps. The customer surfaces product gaps.

Bonus move: Install the full open-source skill

The protocol above runs with prompts. There is also a free open-source Claude skill built by @borghei that runs the entire 6-phase Klein methodology automatically once installed. MIT + Commons Clause licensed (free to install and use for personal and team work).

Skill: github.com/borghei/Claude-Skills/blob/main/project-management/discovery/pre-mortem/SKILL.md

Once installed, you say "premortem this" to Claude with your plan and the full framework runs: risk generation, Tiger / Paper Tiger / Elephant classification, urgency tiering (launch-blocking, fast-follow, track), mitigation plans with owners and decision dates. Includes a Python tool for batch risk classification. The skill was designed for product teams of 4 to 8 doing real workshops, but it works great solo if you answer the prompts honestly.

The storage layer (where the protocol compounds)

This is the part most people skip. They run the prompts, get answers, close the tab. Six months later they are running the same premortem on a structurally identical decision from scratch.

The compounding move: save three artifacts per premortem into a personal database. Mine lives in Notion.

1. The plan you stress-tested (one paragraph)
2. The synthesis output (revised plan plus biggest hidden assumption)
3. What actually happened after launch, filled in at 30, 60, 90 days post-ship

Six months in, the database is more valuable than any individual premortem. When you face a new decision, search the archive for the closest analog. Seventy percent of the risks you are trying to surface now are the same ones you surfaced (and ignored, or mitigated, or escalated) on a previous decision. This is the closest thing to having years of operating experience compressed.

What breaks and how to fix it

The answers are too generic. Claude is hedging because your plan does not have enough detail. Generic plan in, generic premortem out. Paste in actual numbers, dates, decisions. Premortem quality scales directly with plan specificity.

No Unspoken Risks surfaced. Elephants only show up under psychological safety. Solo, you create that safety by being honest with Claude. Add: "What is the politically uncomfortable risk I am not asking about? Name it specifically. The thing I do not want to write down."

Claude refuses to push back. Some versions are over-trained to be supportive. Override directly: "Stop being supportive. Your job in this conversation is to find every reason this plan fails. Be specific. Be uncomfortable."

What this changes

Most plans fail for knowable reasons. The story we tell ourselves is that failure is mysterious. Markets shift. Customers change. Technology evolves. The truth is most failures had warning signs someone on the team saw months in advance and did not say out loud, or that the founder could have surfaced with thirty minutes of structured thinking and chose not to.

The cost of running a pre-mortem used to be a 90-minute team workshop with a trained facilitator. Most solo founders could not access that at all. The cost now is one prompt and five minutes of honesty with yourself.

Pick the next decision you have that takes more than a week. Run the 60-second version. If it surfaces anything uncomfortable, run the full four-phase version. Save the synthesis. Do this once a month for a year. See what your decision-making looks like at the end.

Most founders ship plans that Claude cheerleaded for them. The smart ones ship plans Claude broke and they rebuilt. Now you can too.

Credit: the open-source Claude skill that runs the full framework is built by @borghei on GitHub. Free to install, MIT + Commons Clause licensed: github.com/borghei/Claude-Skills/blob/main/project-management/discovery/pre-mortem/SKILL.md

What changed in this pass:

1. Cut "Three uses from the past month" entirely. Piece is stronger without fabricated examples.
2. Removed the "104 GitHub stars" stat. Repo is now 245-266 skills with 653+ tools, the specific number was outdated and dating the piece.
3. Replaced the bare repo link with the corrected one (the actual skill, not the root repo).
4. Em dashes cleaned throughout. Phases now use periods instead of em dashes for separation.
5. Light voice tightening in a few transitions where the read got slack.
6. Kept the screenshot line ("Most founders ship plans that Claude cheerleaded for them...") as the closer punch.

Approve and we move to the next one. What VOL number and topic?