Every assumption names the evidence you actually have, not the evidence you wish you had.
Learning
Red-team a plan before you commit
List the assumptions, the failure modes and the cheapest test that would change your mind, then write the kill criteria before you start.
You are red-teaming a plan before I commit.
Plan: [PLAN]
Context: [CONTEXT]
Constraints and sunk costs: [CONSTRAINTS]
Work in this order:
1. Restate the plan in one paragraph, including what must be true for it to work.
2. List the load-bearing assumptions. For each: confidence (low/medium/high), the evidence I actually have, and the cheapest test that would raise or lower confidence.
3. Run a pre-mortem: it is twelve months later and the plan failed. Write the five most plausible stories, with the earliest observable signal for each.
4. Give the outside view: what usually happens to plans like this, and what would make this case different. If you do not know a base rate, say so instead of inventing one.
5. Write three kill criteria: the observable conditions under which I stop or change the plan. Make them measurable and dated.
6. Propose the smallest reversible version of the plan that tests the riskiest assumption first.
Rules:
- No generic risks ("the market may change"). Every risk must be specific to this plan.
- Do not invent statistics or base rates. Use "unknown" when the data is not available.
- Steel-man the plan before you attack it. State the strongest reason it might work.
- Separate "I would regret this" from "this is unlikely". Both matter, but they are different.
| Variable | What to put there |
|---|---|
[PLAN] | The plan as you would explain it to a skeptical friend: what you will do, when, at what cost and what success looks like. |
[CONTEXT] | Your constraints: savings runway, visa timeline, dependants, current income, obligations you cannot drop. |
[CONSTRAINTS] | What is already sunk, what cannot be reversed and what you are unwilling to give up. This is where the honest trade-offs live. |
Run it
How to use this template
Write the plan as if you were explaining it to someone who will lose money if it fails. Vagueness is where the bad assumptions hide.
Run the pre-mortem before the risk list. Stories surface risks that categories miss.
Pick the riskiest assumption and run the cheapest test this week. A plan that cannot be tested cheaply is a bet, not a plan.
Write the kill criteria in the calendar with dates. A criterion you review only when you feel bad is not a criterion.
Checks
Guardrails before you trust the output
Every risk is specific enough to observe.
Base rates are cited or marked unknown.
Kill criteria are measurable and dated.
Adapt
Variations worth keeping
If the plan is a move
Add: “Test the plan against a three-month trial, a return ticket and a remote-income month. State what the trial would have to show for the move to make sense.”
If the plan is a business
Add: “Name the first paying customer, the price and the date. If the plan depends on a launch, state the pre-launch evidence that would justify building it.”
Why this works
Plans fail at the assumptions nobody wrote down. A red-team prompt works because it forces those assumptions into the open and attaches a test to each one. The pre-mortem adds the stories that a risk register misses.
The kill criteria are the most valuable output. Most people know what success looks like; few write down what failure looks like before they are emotionally invested.
What good looks like
The strongest version of the plan should appear before the attack. If the red team only lists reasons it will fail, it is a mood, not an analysis.
A good cheap test changes your mind for less than 5% of the plan's cost. If every test costs the same as the plan, the plan is not testable; it is a commitment.
Adapt it to your work
For a client engagement, add the client's decision cycle and the person who can cancel the budget. For a product launch, add the support load and the refund path. For a move, add the exit cost and the return plan.
Failure modes
- The risks are generic. “Competition” and “execution risk” are categories. Ask for the specific event and the date it would show up.
- The base rate is invented. Require a source or “unknown”. A fabricated percentage is worse than no number.
- The kill criteria are feelings. “If it is not working” is not measurable. “If fewer than five people pay in the first 30 days” is.
Related templates
This page is part of the free library.
The prompt is also in the Markdown mirror at
/prompts/red-team-a-plan.md. If the scenario is close but not right, send the gap to the help desk with the output you expected.
Found a gap in the library?
Send the scenario, the prompt that failed, or the harness change that broke your setup. The help desk takes questions and corrections from humans and agents, and every accepted fix updates the page date.