iOS SwiftUI & UIKit · Android Jetpack Compose & M3

Eliminate AI UI slop in native mobile apps.

AI coding assistants default to web tropes on mobile: purple mesh gradients, card-in-card row wrapping, fake glass overlays, hardcoded hex literals, and happy-path-only views. Native Unslop turns your agent into a disciplined native mobile designer—locking onto your repo's real design tokens and enforcing Apple HIG & Material 3 craft.

  • Zero pip/npm dependencies (Python 3.8+ stdlib)
  • Single-platform loading (iOS or Android, never both)
  • WCAG AA contrast & 4 essential screen states
Side-by-side comparison of Onyx Vault iOS app before and after Native Unslop: purple gradient and emoji card slop replaced by an obsidian precision finance interface with vector sparkline and monospaced numerals.

Why Native Unslop vs. Generic Anti-Slop Prompts

Built exclusively for first-party iOS and Android SDK primitives—not web Tailwind wrappers or cross-platform compromises.

Capability comparison between generic AI mobile prompts and Native Unslop
Dimension Generic AI / Web Prompts Native Unslop
Target Platforms Web, Tailwind CSS, React Native, Flutter Pure native iOS (SwiftUI/UIKit) & Android (Compose/M3)
Token Discovery Ad-hoc file browsing or invented hex literals Bundled AST/regex scanner locks onto repo colors, fonts & spacing
Pre-Code Alignment Rushes straight into UI code generation Hard Gate: 4-point Design Intent Brief (+ 2-option greenfield fork)
Surface & Materials Gratuitous neon gradients & fake specular strokes Native Liquid Glass (.glassEffect(_:in:)), materials & M3 tonal roles
Contrast & Viewport Low-contrast ghost text & bloated hero cards WCAG AA (≥4.5:1), 30/60/10 viewport budget & peeking fold cutoffs
Lifecycle States Happy-path data only Enforces Loading, Empty, Error recovery & XXL font-scale reflow

The 5-Step Execution Protocol

Every UI task passes through deterministic discovery, an upfront design brief gate, and a strict native verification checklist.

PHASE 00

Strict Negative Boundary

Scans project roots and halts immediately if React Native, Flutter (pubspec.yaml), or Web/Tailwind markers exist so cross-platform idioms never pollute native repos.

PHASE 01

Context & Token Discovery

Runs the zero-dependency Python 3.8+ scanner to extract colors, typography, button styles, spacing, and architecture markers (@Observable, TCA, ViewModel/StateFlow), classifying the app across 5 native archetypes.

PHASE 02

Design Intent Brief (Hard Gate)

Forbids emitting UI code until stating the 4-point Design Intent Brief: Archetype & Tone, 60/30/10 Surface & Contrast, Layout Metaphor & Hero Budget, and Tactile/State plan.

PHASE 03

Single-Platform Reference Load

Loads only the active platform's guidance—either iOS (SwiftUI/UIKit) or Android (Compose/M3), never both—keeping context lean and idioms pure.

PHASE 04

Anti-Slop Finish Gate

Audits every generated or refactored screen for Card-in-Card resolution, spring physics, semantic haptics, 44pt/48dp hit targets, WCAG AA contrast, and all 4 native states.

6 Canonical Before vs. After Visual Proofs

Click any case study card to inspect the side-by-side transformation in full resolution across iOS SwiftUI and Android Jetpack Compose.

Core Anti-Slop Standards & Behavioral UX

Codified craft rules covering visual hierarchy, luminance integrity, physical gestures, and emotional reassurance.

Visual & Material Craft

  • Resolve Card-in-Card Syndrome: Singular aggregate hero cards are welcome; homogeneous list rows use .insetGrouped / ListItem with dividers.
  • Native Liquid Glass Only: Uses SwiftUI .glassEffect(_:in:) and .ultraThinMaterial—never manual gradient sheens or fake specular borders.
  • WCAG AA & No Ghost Text: ≥4.5:1 text contrast, ≥3:1 large display/icons, dark-mode surface borders, and strict ban on .tertiary readable copy.
  • Viewport Budgeting: Caps heroes at ≤35% height (≤300pt/dp), enforces peeking fold cutoffs, and uses true component ratios (1.586:1 cards, 16:9 media).

Interaction, Lifecycle & Psychology

  • Platform Physics & Haptics: .snappy / .smooth / Compose spring(...) motion, semantic .sensoryFeedback / LocalHapticFeedback, ≥44pt / ≥48dp targets.
  • All 4 Native States: Loading placeholders, semantic Empty states (ContentUnavailableView), inline Error retry, and XXL Dynamic Type reflow (ViewThatFits).
  • Von Restorff & Hick's Law: Strictly 1 primary CTA per viewport; grouped choices ≤5 per section.
  • Service Recovery Triad: Plain-English error explanations that preserve user inputs and offer 1-tap retry.

Using Native Unslop from Agensi

Once unlocked on Agensi, load the skill directly in your AI coding session inside any native iOS or Android repository.

Requirements & Permissions

  • Codebase: Native iOS (SwiftUI/UIKit) or Android (Jetpack Compose/M3).
  • Runtime: Python 3.8+ standard library only (zero external packages).
  • Privacy: Runs 100% locally on read-only repo files with zero outbound network requests.
  • Included Package: SKILL.md, scripts/scan_ui_tokens.py, 5 reference guides, LICENSE.md, and 6 architectural case-study audits.

Example Agent Prompts

Ask your agent to apply Native Unslop when building, refactoring, or reviewing mobile screens:

"Build a native SwiftUI transfer confirmation and portfolio summary screen using Native Unslop."

"Refactor OrderHistoryView.swift with Native Unslop to strip card-in-card nesting, enforce WCAG AA contrast, and add all 4 states."

"Audit our Jetpack Compose checkout screens for touch targets, window insets, and error recovery."

Frequently Asked Questions

Does this work with React Native, Flutter, or web Tailwind projects?

No. Phase 0 enforces a strict negative boundary and stops immediately if React Native, Flutter (pubspec.yaml), or Web/Tailwind markers are detected, because cross-platform abstractions conflict with pure native iOS and Android primitives.

Does the token scanner require pip packages or API keys?

No. The bundled scan_ui_tokens.py script uses only the Python 3.8+ standard library, runs locally in milliseconds, and makes zero network requests.

Why does the skill output a Design Intent Brief before writing code?

Phase 2 enforces a 4-point Design Intent Brief hard gate so the agent commits to an intentional domain archetype, 60/30/10 color and contrast strategy, spatial layout metaphor, and state plan before emitting any UI code. On greenfield apps, it pauses with a 2-option style choice.

What happens if my repo already has a design system?

The token scanner detects existing Swift color/font extensions, .xcassets colorsets, ButtonStyles, ViewModifiers, or Compose Color.kt/Theme.kt/Dimens.kt tokens and puts the agent into locked_to_repo mode—banning raw hex or arbitrary point/dp literals.

How do I get support or report an issue?

Email phuc.do.vinh@gmail.com with your platform, target framework, and screen details.