What problem does it solve? Multi-step operations that touch databases, external APIs, queues, and caches can fail between steps, leaving systems in partial or corrupted states that the happy path never produces. This Skill guides an agent to systematically investigate what happens when an operation fails mid-flight instead of only asking whether it works. ## Core Features & Use Cases - Failure Point Mapping: Decomposes operations into steps, labeling each as idempotent, reversible, or an irreversible external effect, then analyzes the state left behind if failure occurs after each step. - Six Failure Classes: Investigates partial success, timeouts, lost responses, retries, crashes, and rollback failures with concrete attack patterns for each. - Evidence-Based Reporting: Produces findings via an audit report template with confidence levels from CONFIRMED (injected failure observed) down to SPECULATIVE, plus explicit false-positive criteria for transactions, sagas, and idempotent designs. - Use Case: When reviewing a checkout flow that writes an order, charges a payment gateway, and sends a confirmation email, use this Skill to identify that a database error after the charge leaves the customer billed with no recorded order. ## Quick Start Audit the checkout operation in this codebase for partial failure states, unsafe retries, and missing rollback coverage.
