What problem does it solve? Generated content that must be identical on every device — a daily pick, a procedural layout, a shareable result — breaks when generators read wall clocks, use ambient randomness, or depend on unspecified PRNG behavior. This Skill enforces the discipline that makes output a pure function of an injected key, and pins it with a committed golden-vector table so regressions fail loudly instead of silently rewriting what users already saw. ## Core Features & Use Cases - Deterministic seed derivation: Hash the key with FNV-1a-64, XOR per-feature and per-mode salts, mix through SplitMix64, and draw from a PRNG the repo owns — never dart:math's Random. - Golden vector tables: Commit (key, params) → fingerprint rows computed from an independent oracle, covering boundaries, leap days, and historical keys, regenerated only by a reviewed local command while CI verifies but never blesses. - Versioned generator cutovers: Ship generator improvements as new versions with a cutover key, keeping old versions in the binary so stored outcomes referencing past content still reproduce. - Use Case: A Flutter app shows every user the same daily content. Use this Skill to build the generator as a pure function of an ISO date key, ban DateTime.now() and ambient Random() via the check script, and pin the output with fingerprint vectors so a refactor that reorders draws fails the build. ## Quick Start Apply the seeded-determinism rules to my Dart daily-content generator and create a golden vector table with an independent oracle for its fingerprints.
