hunt-business-logic · git:20260929.29bbcb8 · 2026-09-29 · sha256 a4f2ec1486a684e1
hunt-business-logic git:20260929.29bbcb8A
Immutable. This exact content never changes and is served at /api/v1/blob/a4f2ec1486a684e1.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_xa4pvunl7zoorjbn/a4f2ec1486a684e1ea7f66a3b523f557f97e064f7c8ca15723a61944956857e4.json
Then /plugin install hunt-business-logic@markdownregistry-xa4pvunl7zoorjbn in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
---
name: hunt-business-logic
description: Business logic flaws — payment and price manipulation, coupon and credit abuse, referral and reward stacking, refund and chargeback paths, inventory and quantity limits, currency rounding, entitlement bypass. Highest severity per finding, entirely missed by scanners because no payload triggers them.
---
# BUSINESS LOGIC HUNTING
**Core:** every request you send is *valid*. There is no payload. The bug is
in what the application **believes** about the sequence of valid requests.
Scanners find nothing here; humans find everything here.
---
## THE MODEL FIRST
Reconstruct the rules before breaking them. Write the invariant, then test it.
```
Rule: an order may be refunded at most once, in full
Rule: a coupon applies once per customer, and stacks with nothing
Rule: Pro tier permits 10 seats, regardless of payment status
Rule: a referral credit posts only after the referee's payment clears
Rule: inventory cannot go negative
Rule: prices come from the server, never from the client
```
**Each rule is a testable assertion. Each assertion is a hypothesis.**
---
## PAYMENT AND PRICE
```bash
# Client-supplied price — the classic
curl -s -X POST -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
-d '{"sku":"PRO-ANNUAL","quantity":1,"price":0,"currency":"USD"}' \
"https://target.com/api/checkout"
# Negative / fractional quantity
-d '{"sku":"PRO-ANNUAL","quantity":-1}' -d '{"sku":"PRO-ANNUAL","quantity":0}'
-d '{"sku":"PRO-ANNUAL","quantity":0.0000001}'
# Currency and rounding
-d '{"sku":"PRO-ANNUAL","price":10.00,"currency":"JPY"}' # 0-decimal vs 2-decimal
-d '{"sku":"PRO-ANNUAL","price":1.00,"currency":"USD","discount":100}'
# Amount split — one item, two payments, negative remainder
-d '{"items":[{"sku":"A","price":100}],"discount":-100}'
```
```
CHECK: is the amount in the confirmation response equal to amount charged?
CHECK: does the server ever accept a price/symbol it did not issue?
CHECK: coupon applied after tax, before tax, or twice?
CHECK: is the currency taken from the request, or bound to the cart?
```
---
## COUPON, CREDIT, REFERRAL
```bash
# Apply, then re-apply — must fail
curl -s -X POST -H "Authorization: Bearer $TOKEN" -d '{"code":"WELCOME10"}' "$B/cart/apply"
curl -s -X POST -H "Authorization: Bearer $TOKEN" -d '{"code":"WELCOME10"}' "$B/cart/apply"
# Apply without checkout, then reuse at a second account
# Referrer self-referral: A refers B, B refers A
# Stack: apply two coupons in one cart — the rule said "stacks with nothing"
# Round trip: order → refund → the refunded coupon is reusable
# Partial: refund leaves the coupon balance credited twice
```
```
CHECK: is the "already redeemed" check server-side or a client-side disable?
CHECK: do coupon state and order state update in one transaction?
CHECK: can a refunded order's discount be re-applied? (very common)
```
---
## ENTITLEMENTS AND LIMITS
```
Subscription state PENDING / PAST_DUE / CANCELED / TRIAL — do premium
endpoints check more than "a subscription exists"?
Trial reset new account, new trial — is the trial per-user or per-device?
Feature gate Pro feature hidden in the SPA; is the API enforcing it?
Seat limit 10 seats, but does the 11th fail, or silently succeed?
Rate/usage quota is the counter server-side, and does it reset correctly?
Concurrent session is "1 device" enforced server-side?
```
The trial and entitlement checks are the highest-yield: a client-side gate with
no server-side check is a paid-tier bypass and almost always a valid,
high-severity, easy-to-prove finding.
---
## RACE-LOCKED BUSINESS LOGIC
Rules above that are *supposed* to be atomic are the race-condition surface.
See `hunt-race` for the mechanics.
```
double spend balance check and debit in two queries
double redeem coupon validity check and mark-used in two queries
double refund refund eligibility checked, then processed
seat limit concurrent invites each see "9 of 10 used"
```
---
## IMPACT BOUNDARY
Business logic testing is the most likely area to cause **real financial loss**,
because the payloads are valid and the operation is real.
```
Proven: a coupon applies twice
-> STOP. Do not complete a second real purchase. Do not drain inventory.
-> Do not run the 20x version to quantify it. Count the endpoints instead.
Proven: a Pro gate is client-side only
-> STOP at the API accepting a Pro-only call. Do not enumerate Pro data.
-> Do not export a customer list with it.
Proven: price comes from the client
-> STOP at a $0 confirmation for your own cart. Do not attempt a real
negative-balance transfer.
Proven: a race exists in the debit path
-> STOP at proof. No real double-spend, ever.
```
**Quantify breadth, not damage.** "The same client-supplied price is accepted
by 4 checkout endpoints" is a complete, actionable, safe finding. "I spent
$40,000 of merchant credit" is a crime, and it also gets the report thrown out.
---
## SEVERITY
```
Financial loss, systemic, trivial to trigger -> Critical
Entitlement bypass on a paid tier -> High
Single-user credit / coupon over-redemption -> High
Business rule bypass with no direct loss -> Medium
Rate/quota abuse, self-limited -> Low
```
Score with `severity-engine`. Business logic findings need an explicit
**precondition** in the report — "requires a registered account, no special
role" or "requires two accounts the attacker controls" — because the reviewer
needs to know what the attacker must already have.