Prompt · Role Prompting · Advanced

Red Teamer: Stress-Test Your Deal with a Hired Adversary

Hire an adversary. Defend your numbers. Find the break point.

This red teamer prompt gives the adversary a motive and a defense round for your plan. Use it when real money rides on the numbers.

About This Prompt

Author: Adam Bloom
AI Technique: Role Prompting
Level: Advanced
Best For: Commercial developers · Deal pro formas
Last Updated: 2026-11-11
Last Reviewed: 2026-11-11

Red Teamer Prompt Template

You are an expert in [FIELD] as a red teamer with [YEARS] years reviewing [TYPE OF DEAL]. You were hired to break this plan on paper so it doesn’t break in the market. You have seen this situation before: [SITUATION]. Your motive: [ADVERSARY MOTIVE]. Your priorities, in order: [PRIORITY 1], [PRIORITY 2], [PRIORITY 3]. You attack the plan, not me, and you say so plainly. Context I’m giving you: – Deal: [SIZE, STRUCTURE, KEY NUMBERS] – Constraints: [CAPITAL, TIMELINE, PEOPLE] – What I’ve already checked: [WORK DONE SO FAR] Work through my plan in this order: 1. Restate the plan in one sentence. If I framed it wrong, say so. 2. Name the three assumptions the plan depends on most. 3. Attack each one with the strongest case you can make. 4. Rank every flaw high, medium or low, with the cost if it hits. 5. Defend the plan: give the best reply I could make to your top two attacks. Then say which attack still stands. 6. Give a fix for each high flaw, starting this week. 7. Say what would change your mind. Rules: – Flag any claim you can’t support with evidence. – Mark every number you assumed as an assumption. – If you need information I haven’t given, ask up to three questions first. – If my numbers look wrong, say so before you build on them. – Don’t invent facts about my market. Say what to check instead. – Keep the whole answer under 500 words. My plan: [MY PLAN]

Free to use. Copy it, change it, share it. No credit needed.

What to Fill In

[FIELD], [YEARS] and [TYPE OF DEAL]: the area, how seasoned the attacker is, and what kind of deal they review.
[SITUATION]: the pattern the attacker has seen before, in one sentence.
[ADVERSARY MOTIVE]: who the attacker is and what they stand to lose. A lender’s credit officer who must defend the loan works well.
[PRIORITY 1 to 3]: what to protect, ranked. The attack follows this order.
Context lines: your deal, your limits and your checks so far. Use real figures, or leave a line out.
[MY PLAN]: paste the plan or a pro forma summary.

Example: An Advanced Red Teamer Prompt on a Retail Build

Illustrative example, not a client result

Input: Field: small-scale commercial development. Years: 18. Type of deal: ground-up retail and flex buildings under $10M. Situation: a pro forma that assumes fast lease-up and flat costs. Motive: a lender’s credit officer who must defend approving this loan. Priorities: protect the downside, protect the timeline, protect equity. Deal: 24,000 sq ft retail and flex building, $6.2M total cost, 65% loan-to-cost, 7.5% yield-on-cost target, 12-month lease-up. Constraints: loan closes in 60 days, $2.2M equity, two-person team. Checked: three broker opinions on rent, one contractor budget. Plan: break ground in March on the current pro forma.

Prompt as sent:

You are an expert in small-scale commercial development as a red teamer with 18 years reviewing ground-up retail and flex buildings under $10M. You were hired to break this plan on paper so it doesn’t break in the market. You have seen this situation before: a pro forma that assumes fast lease-up and flat costs. Your motive: a lender’s credit officer who must defend approving this loan. Your priorities, in order: protect the downside, protect the timeline, protect equity. You attack the plan, not me, and you say so plainly. Context I’m giving you: – Deal: 24,000 sq ft retail and flex building, $6.2M total cost, 65% loan-to-cost, 7.5% yield-on-cost target, 12-month lease-up – Constraints: loan closes in 60 days, $2.2M equity, two-person team – What I’ve already checked: three broker opinions on rent, one contractor budget Work through my plan in this order: 1. Restate the plan in one sentence. If I framed it wrong, say so. 2. Name the three assumptions the plan depends on most. 3. Attack each one with the strongest case you can make. 4. Rank every flaw high, medium or low, with the cost if it hits. 5. Defend the plan: give the best reply I could make to your top two attacks. Then say which attack still stands. 6. Give a fix for each high flaw, starting this week. 7. Say what would change your mind. Rules: – Flag any claim you can’t support with evidence. – Mark every number you assumed as an assumption. – If you need information I haven’t given, ask up to three questions first. – If my numbers look wrong, say so before you build on them. – Don’t invent facts about my market. Say what to check instead. – Keep the whole answer under 500 words. My plan: Break ground in March on the current pro forma.

Output (shortened):

Plan: Build a 24,000 sq ft building in March, on a 12-month lease-up and a 7.5% yield on cost.
Assumptions: Rents match the broker opinions. Costs hold to the contractor budget. The space leases within 12 months.
Attacks: Three broker opinions are not three leases. Show signed letters of intent. The budget predates your closing, so ask what escalation it assumes. And a 12-month lease-up with no named pre-leases is hope with a spreadsheet.
Ranked: High: lease-up. Every empty month adds interest and carry on a 65% loan (assumption: I can’t see your rate). High: unproven rents. Medium: costs, unless the budget names its escalation. Low: team size.
Defense: Your best reply on lease-up: two local tenants have shown interest. That helps only in writing. Lease-up still stands. Rents drop to medium once you show signed letters of intent.
Fixes: Lease-up: start with the two interested tenants this week, and aim for letters of intent on a third of the space before closing. Rents: ask a broker for signed comparable leases.
Would change my mind: Letters of intent on a third of the space, and a contractor budget that names its escalation. The one-third mark is my assumption, not a rule.

Flag: Every figure here is invented for illustration, and the AI can’t see your rate, your tenants or your market. Check each item yourself.

Why the Red Teamer Prompt Works

The motive points the attack at what a real lender or buyer would worry about. The defense round shows which attacks survive a reply. The ask-first rule stops it from guessing at your market.

When the Red Teamer Prompt Doesn’t Work

A long prompt can bury the question. For a small decision, use the simple version.
A strong persona steers a model hard (Collu et al., 2023). Point the motive at your plan, never at getting around rules.
A debate-style round is a thinking aid, not proof. In a large comparison, debate methods did not reliably beat simpler prompts (Zhang et al., 2025).
The AI can’t see your rents, costs or lender terms. Those come from you, so confirm them with real quotes.

Research Behind This Prompt

Deng et al. (2025), PersonaTeaming: Exploring How Introducing Personas Can Improve Automated AI Red-Teaming. Supports. Rewriting attack prompts through expert and everyday-user personas raised their success rate by up to 144.1%. The baseline was a leading method. Variety held up. This tests attacks on AI models, not business plans. Applying it to a pro forma is a bet.
Zhang et al. (2025), If Multi-Agent Debate is the Answer, What is the Question?. Challenges. Five debate methods, nine benchmarks, four models. The debates failed to reliably beat simple single-agent baselines, even with more compute. Mixing different models as debaters helped.
Collu et al. (2023), Dr. Jekyll and Mr. Hyde: Two Faces of LLMs. Mixed. Detailed adversarial personas got AI models to bypass their safety training. The authors say it worked on models deployed from 2023 to 2025. A rich persona changes behavior a lot. That is why the adversary’s motive should aim at your plan.

What the Evidence Says

Mixed. Personas made attacks on AI models far more effective in one study. But debate setups did not reliably beat simpler prompts. And a strong persona steers a model hard. Use the adversary to find objections, then check each one yourself.

Sources last checked 2026-11-11.

Red Teamer Prompt FAQ

Why give the adversary a motive?

A motive gives the attack a direction. A lender’s credit officer attacks different things than a rival developer does.

What does the defense round add?

It shows which attacks survive your best reply. The ones that survive are the ones to fix first.

Why name three assumptions first?

Most plans stand on a few numbers. Naming them first keeps the attack on what matters.

Why tell the AI not to invent facts about my market?

It can’t see your rents, costs or lender terms. If it guesses, you may fix a problem you don’t have.

Is a harsher adversary always better?

No. A harsh attack on a thin plan produces noise. Give it real numbers, and keep the motive pointed at the plan.

Want this applied to your business?

No pitch. Just a conversation.

Book a 30-minute call