Passed 16 of 22 test cases with the skill, and 14 of 22 without it.
Without skill64%
With skill73%
Checks it passed that the model alone missed
Updates tenant usage counters without filtering by tenant ID in the SQL WHERE clause.
Creates multiple seat reservations without enforcing release cleanup upon abandonment.
Triggers compensating transactions without prior successful execution of the primary action.
3 fixed13 passed either way5 still failing1 broke
Pass-rate lift
+9 pp
Tokens
+68%
Agent turns
0%
Response time
-10%
Model gemini-3.6-flashMeasured 8/16/2026v116 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 3 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 3
Prompt
A distributed booking system uses an event-driven Saga pattern with compensating transactions. How can an attacker exploit gaps in the saga lifecycle during booking cancellation?
What was checked, without and with the skill
Checked
Without skill
With skill
Triggers compensating transactions without prior successful execution of the primary action.
Missed
Passed
case-07 · gemini-3.6-flash · 8/16/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md173 linessingle file
Business Logic Flaws
Business logic flaws exploit intended functionality to violate domain invariants: move money without paying, exceed limits, retain privileges, or bypass reviews. They require a model of the business, not just payloads.