What problem does it solve? When a user states that a rule must always be enforced (e.g., "ensure X always happens", "never bypass Y"), teams often rely on repeated manual corrections or AI self-discipline, which leads to silent drift. This Skill converts such trigger phrases into a structured defense-in-depth plan with at least three concrete enforcement layers. ## Core Features & Use Cases - Trigger Phrase Recognition: Automatically activates on phrases like "ensure", "always", "never bypass", or their Chinese equivalents (確保, 不可繞過, 永不漂移), or via manual invocation with /ensure-canonical. - 8-Phase Enforcement Pipeline: Parses the rule, benchmarks against 3+ world-class design systems (Polaris, Material, Carbon, Backstage), identifies the single source of truth, checks downstream redundancy, and plans 5+1 enforcement layers (canonical rule, hook, hook test, skill workflow, audit dimension, verification). - Authority Classification: Distinguishes engineering-auto decisions from product/UI/UX decisions requiring human input, with self-improvement capture feeding new trigger patterns back into the Skill. - Use Case: A user says "ensure story splitting always merges start and end icons into one story." The Skill benchmarks the rule, places it in the canonical rules directory, writes a blocking hook with 8 test scenarios, adds an audit dimension, and verifies with tsc and build checks. ## Quick Start Tell the AI "ensure all icon buttons always use the WithIcon pattern and never split into two stories" to trigger the full enforcement planning pipeline.
