What problem does it solve? Verifying that an Expo skill actually works—triggers on the right prompts, generates compiling code, and renders correctly on real devices—requires a slow manual loop of building apps, booting simulators, and eyeballing screenshots. This Skill automates that entire end-to-end evaluation pipeline. ## Core Features & Use Cases - Trigger Accuracy Testing: Runs realistic prompts through claude -p subprocesses and measures recall of whether the skill under test gets triggered. - A/B Code Generation: Executes each eval prompt with-skill and without-skill in fresh Expo fixtures, then gates output on tsc, lint, and expo export. - Runtime Screenshots: Runs generated apps on iOS simulator, Android emulator, or web via Expo Go or development builds, captures screenshots, and grades them against expectations or a reference image. - Results Viewer: Generates a self-contained HTML report comparing with-skill vs without-skill scores, screenshots, and design-quality rubric grades. - Use Case: After editing the expo-ui skill's description, run an eval to confirm trigger recall improved and that generated apps still render correctly on both platforms. ## Quick Start Ask the assistant to evaluate the expo-ui skill end-to-end with runtime screenshots on iOS and Android using the latest Expo SDK.
