What problem does it solve? Concurrent and distributed components fail through rare multi-fault interactions — a network partition plus a slow disk plus a node crash at the same moment — that flaky integration tests cannot reproduce and chaos engineering can observe but not replay. This Skill wires a single seeded random source through every nondeterministic boundary so any failing run replays exactly from its recorded seed. ## Core Features & Use Cases - Seeded Deterministic Boundaries: Replaces network I/O, disk I/O, clock, and scheduling with simulated implementations driven by one seeded PRNG, while real production code runs unchanged in-process. - Fault Injection and Invariant Checking: Drops, reorders, and delays messages, corrupts disk operations, crashes nodes, and skews clocks, asserting a steady-state invariant after every simulated step. - Seed-Based Regression Discipline: Commits each failing seed as a permanent named regression test, modeled on TigerBeetle's VOPR and FoundationDB's simulation framework. - Use Case: When building a consensus or replication protocol, run thousands of seeded simulations to surface a conflicting-commit bug, then replay seed 12345 forever as a regression test proving the fix. ## Quick Start Ask the AI to apply deterministic simulation testing to your distributed component by wiring seeded simulated network, disk, and clock boundaries and asserting your invariant every step.
