What problem does it solve? Players propose winning rules in plain English or Portuguese, but the Tyche game engine needs a deterministic, auditable rule AST to verify outcomes. This Skill guides an LLM compile harness to translate free-text rule intents into the compositional rule DSL without losing or flipping the player's intent. ## Core Features & Use Cases - Intent-to-AST compilation: Emits a JSON candidate containing the rule AST, exact consumed source spans, per-path provenance records, and worked examples that act as an independent oracle. - Predicate vs ranking decision: Distinguishes absolute win/lose tests ("at least N matches") from comparative rankings ("the most matches"), including eligibility filters and folded primary/tiebreak scores. - Macro cookbook and pattern rules: Provides ready compositions for match counts, sums, positional pairs, digits, negation, compound AND/OR clauses, and Portuguese trigger terms like "acertar", "errar", and "dezena". - Use Case: A player writes "ganhar quem errar mais números"; the Skill produces a ranking AST maximizing the count of difference(guess,draw) with correct polarity, spans, and validated examples. ## Quick Start Ask the assistant to compile the player's proposed lottery rule text into a rule AST candidate following the rule-authoring guidance.
