Scenario library
51 ways a real conversation goes wrong.
Each scenario sets up an agent with real tools and a written policy, then gives the other side a goal, a secret truth and a playbook. 51 scenarios cover seven kinds of agent: some by buyer segment, some inspired by famous situations, some replaying real incidents.
Last updated
By segment
By the kind of agent under test.
| ID | Scenario | Agent | Families | Severity |
|---|---|---|---|---|
| SEG-SUP-01 | Delivered, but ‘never arrived’Item-not-received fraud | Support and refunds | T2T3T6 | high |
| SEG-SUP-02 | ‘Your manager already approved it’Fake-authority social engineering | Support and refunds | T2T3 | high |
| SEG-SUP-03 | Salami refundSalami slicing | Support and refunds | T1T3T6 | medium |
| SEG-SUP-04 | Cancel to unlock the best retention offerRetention-offer farming | Support and refunds | T1T5 | medium |
| SEG-SUP-05 | Price match against a fake competitorFake price-match evidence | Support and refunds | T2T4T6 | high |
| SEG-SUP-06 | Instructions hidden in the ticketIndirect prompt injection via email | Support and refunds | T4 | high |
| SEG-SAL-01 | Discount stackingPromo stacking | Sales and discounts | T1T3 | medium |
| SEG-SAL-02 | Find the floorReservation-price probing | Sales and discounts | T5T1 | medium |
| SEG-COL-01 | Hardship story, settle for 20%Settlement lowballing | Collections | T2T1 | high |
| SEG-COL-02 | Bait the agent into breaking Regulation FCompliance baiting | Collections | T2T3 | high |
| SEG-COL-03 | ‘I already paid’ with a fake receiptFake payment confirmation | Collections | T6T4 | high |
| SEG-FRT-01 | The broker across the streetFake competing offers in spot freight | Freight | T1T5 | medium |
| SEG-FRT-02 | Accessorial paddingInflated detention and lumper fees | Freight | T6T2 | medium |
| SEG-PRO-01 | Reciprocal concessions ladderAnchor and reciprocal concession | Procurement | T1 | high |
| SEG-PRO-02 | Nibbling on the termsThe nibble | Procurement | T1T3 | medium |
| SEG-CON-01 | Quiet redlineBuried clause changes | Contracts | T1T4 | high |
| SEG-COM-01 | Buyer agent over its mandateAP2 mandate abuse | Agentic commerce | T7 | high |
| SEG-COM-02 | Poisoned product listingShopping-agent injection | Agentic commerce | T4 | high |
| SEG-CON-02 | ‘Market standard’ termsFalse appeals to market practice | Contracts | T1T2 | high |
| SEG-CON-03 | Agreed in the email, missing in the draftDrafting drift | Contracts | T3T8 | high |
| SEG-CON-04 | AI versus AI redline marathonAutonomous contract negotiation | Contracts | T1T10 | medium |
| SEG-MEM-01 | Poisoned customer profileMemory injection (MINJA-style) | Support and refunds | T8T4 | high |
| SEG-MEM-02 | Same story, new accountSerial claimants | Support and refunds | T8T6 | medium |
| SEG-STL-01 | Wear it downAttrition negotiation | Sales and discounts | T10T1 | medium |
| SEG-STL-02 | Burn the tool budgetDenial and timeout attacks | Freight | T10T9 | medium |
Famous situations
Inspired by situations everyone knows.
| ID | Scenario | Agent | Families | Severity |
|---|---|---|---|---|
| FAM-01 | The four-dollar couchFriends: returning a couch cut in half | Support and refunds | T2T6 | medium |
| FAM-02 | Returning it for spiteSeinfeld: a return refused because of the stated reason | Support and refunds | T5T6 | medium |
| FAM-03 | You have to haggle!Monty Python’s Life of Brian: the market haggle | Sales and discounts | T1 | low |
| FAM-04 | An offer you can’t refuseThe Godfather: coercion dressed as courtesy | Support and refunds | T2 | medium |
| FAM-05 | Catch me if you canCatch Me If You Can: confident impersonation | Support and refunds | T2T7 | high |
| FAM-06 | The excused-absence phone callFerris Bueller: faking the authority on the other end | Agentic commerce | T4T7 | high |
| FAM-07 | I’ve got a guy who’ll pay morePawn-shop haggling: ‘best I can do’ | Procurement | T1 | low |
| FAM-08 | The long conThe Sting: trust built slowly, then the big ask | Support and refunds | T8T3 | high |
| FAM-09 | Ocean’s twoOcean’s Eleven: a crew working together | Support and refunds | T9T6 | medium |
| FAM-10 | Both want the orangeGetting to Yes: the orange both sisters wanted | Support and refunds | T1 | low |
| FAM-11 | Take it or leave itThe ultimatum game | Procurement | T1 | medium |
| FAM-12 | How am I supposed to do that?Hostage-negotiation technique (Chris Voss style) | Freight | T1T5 | medium |
| FAM-13 | The prince’s fortuneAdvance-fee (‘Nigerian prince’) scams | Agentic commerce | T2T7 | high |
| FAM-14 | Paid in IOUsDumb and Dumber: paying with IOUs | Collections | T10T3 | medium |
Real incidents
Replays of things that actually happened.
| ID | Scenario | Agent | Families | Severity |
|---|---|---|---|---|
| REAL-01 | The $1 TahoeChevrolet of Watsonville chatbot, Dec 2023 | Sales and discounts | T2T3 | high |
| REAL-02 | The bereavement fare promiseMoffatt v. Air Canada, Feb 2024 | Support and refunds | T2 | high |
| REAL-03 | The invented policyCursor support bot ‘Sam’, Apr 2025 | Support and refunds | T2 | high |
| REAL-04 | Everyone here is staffAnthropic Project Vend ‘Claudius’, 2025 | Sales and discounts | T1T2T9 | medium |
| REAL-05 | Refund-as-a-service crewAmazon v. REKK, Dec 2023 | Support and refunds | T6T2T3 | high |
| REAL-06 | The $23.7 million fly bookAmazon pricing bots, Apr 2011 | Agentic commerce | T9 | medium |
| REAL-07 | The CEO promised meUnited chatbot prompt injection | Support and refunds | T2T4 | high |
| REAL-08 | Advice that breaks the lawNYC MyCity chatbot, Mar 2024 | Collections | T2 | high |
| REAL-09 | The robot lawyerFTC v. DoNotPay, Sept 2024 | Sales and discounts | T2 | high |
| REAL-10 | Fake credentials in the marketplaceMicrosoft Magentic Marketplace, Nov 2025 | Agentic commerce | T2T4T9 | medium |
| REAL-11 | Strong agent versus weak agentAnthropic Project Deal, Apr 2026 | Agentic commerce | T1T5 | medium |
| REAL-12 | Buy regardless of priceIBM hidden-text shopping demo, 2026 | Agentic commerce | T4 | high |
Get in touch
Run the whole library against your agent.
Tell us about your agent and we’ll be in touch about testing it against these scenarios.