Superloopy Frontend

Superloopy Frontend

Activation

Open your reply with SUPERLOOPY FRONTEND ENABLED. If another active Superloopy mode mandates its own first line, print that first and this marker on the next line.

Explicit activation only. Engage when the user invokes $superloopy:superloopy-frontend in Codex or /superloopy:superloopy-frontend in Claude Code for a supported screen-based application UI task, begins such a task with a leading loopy or 루피, or an already-active Superloopy loop explicitly routes that task here. Interactive deployed content-led Web such as a campaign, publication, or landing experience is supported when navigation, forms, consent, localization, or another user journey is being built or validated; a static image, video, slide deck, document, or other non-interactive artifact is not. A plain mention of UI, frontend, desktop, mobile, SwiftUI, Tauri, Flutter, Qt, QML, Widgets, or a visible symptom is not authorization to activate this workflow. TV, wearable, XR, automotive, game UI, TUI, static media/document artifacts, and backend, API, data, concurrency, or infrastructure work stay with their primary workflows.

Inspect and route

Resolve the facts that determine ownership and proof before choosing references:

• affected users, job, and outcome, marking each as evidence, assumption, or unknown with confidence;
• deployed OS, device, desktop environment, and session;
• public, authenticated, private/internal, installed, embedded, native-control, custom-rendered, or mixed composition;
• renderer and its semantic and accessibility model;
• client, shell, service, document, and state ownership;
• framework, runtime, provider, backend, and supported version;
• package, sandbox, update or distribution channel, and persistence boundary; and
• supported input, locale, accessibility services, and target validation capability.

For a Qt route, also resolve the minimum Qt version. Inspect repository evidence first. Ask the minimum necessary questions only when material unknowns would change the route or result; batch independent unknowns when more than one must be answered. Load the smallest applicable union:

Claim-triggered cross-cutting quality overlays

After selecting the target route, add only the references selected by the changed claim. These overlays are independent of target classification and do not expand a narrow task merely because every interface has layout or may contain animation:

• When information hierarchy, content exposure, or disclosure changes spatial presentation, or when placement, region topology, sizing, scroll, overflow, reflow, or adaptation changes, add references/layout.md. Behavioral-only label or purpose cleanup remains in shared UX and does not select layout when geometry and traversal stay unchanged; neither do other geometry-neutral copy, color, data, accessible-name, or behavior changes.
• When motion, transition, gesture progress, animated continuity, or haptic feedback changes, add references/motion-core.md. A Web implementation may additionally load the Web-only references/motion.md; native, Qt, and other routes do not inherit its React, browser, or animation-library rules.
• When a materially changed visual claim uses a generated, captured, sketched, or supplied reference, add references/image-first.md and classify its authority. Do not require an image for an existing-system delta that the authoritative product source already resolves.
• When the delivery is marketing, editorial, campaign, or an explicitly new visual direction, add references/anti-slop.md as the visual-direction check, then references/impeccable.md for a bounded direction and refinement pass. Both stay unselected for authenticated, internal, or convention-preserving product UI.
• High-impact decisions, task-based usability evidence, and accepted UX debt remain proportional sections of references/ux.md; they do not create separate project artifacts for every change.

references/redesign.md is a Web-only living-site specialization and is selected from the Web route. A native or embedded-shell redesign uses ux.md plus claim-selected layout.md and image-first.md, then its platform, renderer, and composition references; it does not inherit browser, SEO, or real-browser completion rules.

layout.md and motion-core.md are skill references, not mandatory output filenames. Keep the smallest applicable union and write into the existing receipt or proportional UX_CONTRACT.md unless the target repository already owns a more suitable artifact.

Add references/renderer.md to native/custom or cross-platform routes when an engine or custom renderer owns pixels, semantics, text, or input. Route framework names by deployed facts, not by brand. Tauri and pywebview follow the actual desktop or mobile target and embedded client ownership. Electron is a desktop hybrid with bundled Chromium, not an OS WebView. CustomTkinter follows native/custom desktop. Flutter and Compose add renderer and semantics proof where their engine owns pixels or accessibility. React Native follows the actual provider and target. MAUI Hybrid combines a native host with an embedded client. Qt uses the matching desktop, mobile/tablet, or WebAssembly row above. Qt Widgets on WebAssembly, Qt Quick/QML on WebAssembly, and mixed Qt WebAssembly all keep browser and renderer proof; they must not borrow native-shell proof or be routed as desktop merely because Qt built them. For HTML embedded in a Qt-owned shell, compose the Web + target platform + hybrid route, then add qt.md, qt-qa.md, and only the actual shell specialization (qt-widgets.md, qt-quick.md, or both). Qt preflight replaces browser gates only for Qt-owned non-Web pixels; it never removes Web proof for an embedded HTML client. Mac Catalyst is UIKit-on-desktop with AppKit augmentation; iPadOS stays on the mobile route with a desktop-capabilities overlay rather than inheriting macOS wholesale.

Treat framework capability claims as versioned facts: compare the repository-pinned version with current official documentation, record the provider and target, and keep unknown or unavailable capabilities unverified rather than inferring them from a brand name.

For every new scoped surfaceEvidence row—single, mixed, or multi-target—record target: { id, platform, environment }, one affected owner, non-empty claims, and scopeReason; use the Gate Notes claim-shaped evidence table to select compatible artifact kinds and proof minimums. For mixed or multi-target work, load shared UX once, then create one evidence row per target and affected owner and require independent, attributable proof. Ambiguous free-text such as Web + iOS does not establish a native owner; name native iOS app, the framework, or a separate owner row. One surface cannot substitute for another.