What problem does it solve? Subtle logic bugs in state machines, date math, sign conventions, dedup windows, and pagination cursors often survive code review because nobody exhaustively modeled the edge cases. This Skill systematically finds and fixes those provable bugs before they reach production. ## Core Features & Use Cases - Exhaustive Logic Modeling: Enumerates every state, transition, and edge case (empty inputs, boundary dates, negative amounts, timezone/DST edges, off-by-one windows) for one tricky unit per run. - Proof-Driven Bug Fixing: A bug is only accepted when a failing test reproduces it AND documented intent (comments, docs, ADRs, git history) confirms the divergence, preventing opinion-based rewrites. - Minimal Root-Cause Fixes with Regression Tests: Ships one fix at a time via /ship, including the failing test as a permanent regression test and a caller sweep for dependencies on old behavior. - Use Case: Run it against a monorepo package containing balance computation logic; it models the sign conventions, derives a breaking input, proves the divergence from beancount semantics with a failing test, and ships the fix. ## Quick Start Invoke /routine-logic-bugfixer with an optional package or path argument to have it hunt for and fix one provable logic bug in that scope.
