gauntlet-looplisted

# Gauntlet Loop Turn a goal into a closed maker → critic loop with a real external quality bar. The critic is separate, binary, and allowed to say the work loses. Never call a green maker pass a win. ## Choose a mode - **Prompt mode:** the user wants a paste-ready one-shot prompt. Generate the prompt; do not do the work. - **Run mode:** the user says build, implement, polish, or run it. Execute the loop in the current task. - **UI mode:** any frontend, UI, visual, interaction, or design task. Apply the UI checks below in either mode. If mode or the goal is genuinely unclear, ask one short question. Do not ask about architecture before the bar is chosen. ## Set the bar first If the user supplied a reference, use it. Otherwise offer **2–3 candidates and stop**. A valid bar is: 1. **Named** — a specific page, product, repository, paper, post, or artifact. 2. **Fetchable** — the critic can actually open, run, read, or screenshot it. 3. **Comparable** — both outputs share a format, viewport, length, or benchmark. For code or tooling, pair the reference with a measurable check when one exists. Never invent a vague bar such as “best-in-class.” Do not copy proprietary code or content; compare qualities and behavior. ## Lock the loop contract Before making changes, write or state: - **Goal:** the user-visible outcome. - **Bar:** the exact reference and how it will be obtained. - **Comparison unit:** the slice being judged. - **Done condition:** a runnable command, tes