What problem does it solve? Concurrent requests that read shared state, decide based on that read, and then write can silently violate business invariants like balance limits, quotas, and uniqueness. This Skill guides an agent to systematically find these race condition windows before they cause double-spends, over-limit grants, or duplicate records in production. ## Core Features & Use Cases - Pattern Detection: Maps READ → DECISION → WRITE sequences and identifies the shared state and invariant each decision protects. - Interleaving Analysis: Models how two simultaneous requests can both pass a check-then-act gate, and verifies whether defenses like CAS, SELECT FOR UPDATE, unique constraints, or serializable transactions close the window. - Confidence-Graded Reporting: Classifies findings as CONFIRMED, HIGH CONFIDENCE, POSSIBLE, or SPECULATIVE with structured audit-report output. - Use Case: While reviewing a wallet debit endpoint, the Skill traces the balance check and deduction, shows that two concurrent requests both read balance=100 and both write balance=50, and recommends a conditional UPDATE or SELECT FOR UPDATE as the fix. ## Quick Start Ask the agent to audit the checkout or balance-debit flow for race conditions where two simultaneous requests could both pass the same check.
