- Home
- Skills
- Mobile Development
- Omni-UX
Omni-UX
Cross-platform UX and UI rules for agents: Apple, Windows, Android and web contracts, surface playbooks, design tokens and an anti-slop audit run before delivery.
Free
Works with the AI Agents you already use
Omni-UX
Example session with this skill installed
Review this React admin dashboard before we ship it. It uses Tailwind, an indigo-to-purple gradient header, Inter for everything, a 3-card KPI grid and glass-effect side panels. Tell me what has to change and what is fine.
- Read your context and instructions
- Compiled the omni-ux
Summary
The dashboard hits the instant-death signature (4 of the 12 tells together), so the visual layer needs a regenerate, not a patch. The data layer and table structure are fine and can stay.
Findings
| # | Severity | Finding | Rule |
|---|---|---|---|
| 1 | P0 | Gradient text on the page title | slop-rules: gradient-text |
| 2 | P0 | No prefers-reduced-motion handling; the KPI counters animate on every load | motion: reduced-motion mandatory |
| 3 | P0 | Line and bar charts render on white cards inside the dark theme | dashboards: charts re-theme in dark |
| 4 | HIGH | Indigo/purple gradient as the only accent | color: one accent hue, not purple by default |
| 5 | HIGH | Inter for display, body and data | type: distinctive display face, tabular-nums on data |
| 6 | HIGH | 3-card uniform KPI grid with equal visual weight | dashboards: one hero metric |
| 7 | HIGH | Glassmorphic side panels over scrolling content | anti-slop: glassmorphism |
| 8 | MEDIUM | Widgets show a number with no action or question attached | dashboards: every widget = question + action |
| 9 | MEDIUM | Loading state is a spinner on a blank page | states: skeleton matching layout |
| 10 | LOW | Focus ring removed on the filter chips (outline: none without a replacement) | accessibility: focus-visible |
What stays
- Table: TanStack Table with virtualization, column resize and a selection model. Keep it.
- Saved views and the record-list-as-home pattern match the CRM/admin playbook.
- Spacing is on the 8px scale throughout.
Changes, in order
- Pick one accent hue from the product's own identity and build OKLCH ramps for light and dark. Remove the gradient header; use a flat surface token.
- Set a display face that is not Inter (one of the registry's pairings), keep Inter or the system face for body, and turn on tabular figures for all numeric cells.
- Promote one metric to hero size; demote the other two to a secondary row.
- Replace glass panels with solid elevated surfaces (Mica-style on Windows, standard materials on macOS, plain surface tokens on web).
- Re-theme the charts with the dark palette tokens; axis and grid lines use
--border-muted. - Add a skeleton that matches the dashboard grid; wrap counters and transitions in a reduced-motion check.
- Restore
:focus-visibleon the chips.
Verdict
Blocked at P0. Re-run the audit after steps 1 to 7; expected result is zero P0, zero HIGH.
Connects securely to your tools. The creator never sees your data.
About this skill
Omni-UX is a design guide written for coding agents. It covers the four platform families an app can ship on (Apple with Liquid Glass, Windows with Fluent 2, Android with Material 3 Expressive, and the web on a shadcn/ui baseline) and the seven kinds of product people build on them: mobile apps, desktop apps, dashboards, CRM and admin tools, marketing sites, e-commerce and AI-native surfaces.
The SKILL.md walks the agent through five decisions in a fixed order: platform contract, surface playbook, flow pattern, tokens, then an anti-slop audit. Each step is a short table in the skill itself, with a pointer to the full chapter when more detail is needed. The package ships those chapters: 32 guide files on principles, platforms, surfaces, patterns, tokens and stacks, four dense reference sheets (platform cheatsheet, copy-ready token values, a decision tree and the full audit checklist) and five JSON registries (platforms, component crosswalk, tokens, flows and 30+ slop rules with severities).
The anti-slop part is the reason this guide exists. Generated interfaces tend to collapse to the same defaults: a purple gradient, Inter everywhere, a centered hero with two buttons, a three-card grid, glass panels and copy that says "supercharge". Omni-UX treats each of these as a defect with a severity. Three or more together means regenerate. P0 findings (gradient text, emoji icons, invented metrics, missing reduced-motion support, outline:none, white charts on dark dashboards) never ship. The final check is the swap-the-logo test: if a competitor's logo would fit the design without anyone noticing, the design is not done.
Accessibility sits above everything else: WCAG 2.2 AA contrast, 44/48 point targets, full keyboard operation with visible focus, name/role/state for screen readers, all six async states (loading, empty, zero results, error, offline, success), reduced motion, 200% zoom, and color never used alone.
What is in the package
- SKILL.md: the five-step procedure, platform contracts, surface rules, state rules, token discipline, audit gates, accessibility floor and a component crosswalk across iOS, Windows, Android and shadcn.
- references/: platform-cheatsheet.md, quick-tokens.md, decision-tree.md, anti-slop-audit.md.
- guide/: 32 chapters (principles, Apple per OS, Windows, Android, web, seven surfaces, eight flow patterns, five token chapters, five stack notes for SwiftUI, Jetpack Compose, React web, React Native/Flutter and cross-platform).
- registry/: platforms.json, components.json, tokens.json, flows.json, slop-rules.json.
Markdown and JSON only. No scripts, no network access, no environment variables.
The source repository (public, GitLab) also contains an optional MCP server with eight tools including ux_slop_audit and ux_search, a Claude Code plugin with /ux-review, /ux-design, /ux-flow and /ux-platform commands, and a ux-auditor subagent. Those are not part of this package; the link is in the SKILL.md.
The guide content is derived from Microsoft's apple-ux-guide (CC-BY-4.0) and the Anti-AI-UI doctrine (MIT), updated for the 2025 to 2026 Apple HIG, Material 3 Expressive, Fluent 2 and current shadcn/ui patterns.
This is guidance, not a renderer: it does not produce images, icons, logos or illustrations and it does not write backend code. Platform rules reflect the 2025 to 2026 state of the Apple HIG (Liquid Glass), Material 3 Expressive and Fluent 2; vendors change these over time. The audit is a checklist the agent applies, so results depend on what the agent can see of the UI (code, screenshots or a description). The MCP server, slash commands and subagent live in the source repository and are not part of this package.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago
- Free with an Agensi account
Permissions
The skill reads only its own files (SKILL.md, references/, guide/, registry/). It makes no network requests, runs no scripts and reads no environment variables. Reading the user's project files happens only when the user asks for a review of them.