Run any Skill in Manus with one click

dec-ai-native-patterns

Stars0

Forks0

UpdatedJuly 13, 2026 at 18:05

AI-native design engineering patterns — generative/adaptive UI, streaming as interaction, prompt-as-interface, human-in-the-loop / agentic UX, eval-driven development, designing for uncertainty, latency and cost as design constraints. Use when designing any AI/LLM-powered feature, building agentic or chat UX, handling streaming/probabilistic output, or planning for model error and cost.

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

jpoindexter

jpoindexter/dec-skills

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Web and Digital Interface DesignersComputer and Mathematical Occupations·SOC 15-1255

SKILL.md

readonly

name	dec-ai-native-patterns
description	AI-native design engineering patterns — generative/adaptive UI, streaming as interaction, prompt-as-interface, human-in-the-loop / agentic UX, eval-driven development, designing for uncertainty, latency and cost as design constraints. Use when designing any AI/LLM-powered feature, building agentic or chat UX, handling streaming/probabilistic output, or planning for model error and cost.
tags	["ai-native","llm","agentic-ux","streaming","evals","always-on"]

AI-Native Design Engineering

The new layer, increasingly central to the role. Probabilistic systems need UX that plans for being wrong.

Generative / Adaptive UI: Interfaces assembled or personalized at runtime by a model rather than fully predetermined.
Streaming as Interaction: Token-by-token output, partial results, and stop/regenerate controls are core interaction patterns, not edge cases.
Prompt-as-Interface: Natural language as an input modality alongside (not replacing) direct manipulation — with affordances that teach users what the system can do.
Human-in-the-Loop / Agentic UX: Design review, approval, correction, and undo around autonomous actions; surface confidence and provenance.
Eval-Driven Development: Treat model behavior as a testable surface — eval sets and regression checks for AI features, analogous to visual regression for UI.
Designing for Uncertainty: Probabilistic outputs need graceful failure, transparent limitations, and easy escape hatches. The system will be wrong; the UX plans for it.
Latency & Cost as Design Constraints: Model choice, caching, and streaming are UX decisions, not just infra ones.

How to apply

Make streaming first-class: show partial output immediately, always offer stop and regenerate.
For agentic actions, design the review/approve/undo loop before the autonomous capability — surface confidence and provenance.
Teach the user what the system can do via input affordances, not a blank prompt box.
Build an eval set early and run it like a regression suite; "looks good once" is not a test.
Plan the wrong-answer path (graceful failure + escape hatch) as carefully as the happy path.
Treat latency/cost/model choice as UX levers — cache, stream, and pick the smallest model that meets the bar.

More from this repository

same repository

dec-accessibility

jpoindexter/dec-skills

Accessibility foundations for design engineers — WCAG / POUR, semantic-HTML-first then ARIA, keyboard and focus management, inclusive design for permanent/temporary/situational constraints. Use when building or reviewing any interactive UI, adding ARIA, implementing modals/menus/forms, or auditing keyboard operability and focus order.

2026-07-130

dec-cognitive-load

jpoindexter/dec-skills

Cognitive Load Theory for interfaces — intrinsic (task difficulty), extraneous (poor-design overhead), and germane (useful mental-model building) load. Use when a screen feels overwhelming, deciding what to simplify vs preserve, chunking complex flows, or diagnosing whether friction comes from the task or the design.

2026-07-130

dec-core-principles

jpoindexter/dec-skills

Foundational design-engineering principles that govern where complexity lives and how interfaces stay fast and consistent — Tesler's Law, progressive disclosure, progressive enhancement, recognition over recall, optimistic UI, perceived performance, single source of truth, constraints as a feature. Use when deciding system architecture vs UI exposure, reducing user-facing complexity, choosing what to render/reveal, or justifying a constraint-driven system.

2026-07-130

dec-css-architecture

jpoindexter/dec-skills

CSS and styling architecture — scaling methodologies (BEM, ITCSS, CUBE CSS) and fluid/responsive technique (mobile-first, clamp() fluid type/space, container queries for component-level responsiveness). Use when structuring stylesheets at scale, choosing a CSS methodology, building responsive layouts, or deciding between media queries and container queries.

2026-07-130

dec-design-system-depth

jpoindexter/dec-skills

Design system depth beyond "tokens are data" — three-tier token architecture (primitive → semantic → component), W3C DTCG format and tooling, scales as systems, component API design, headless/unstyled primitives, documentation as product, governance and versioning. Use when architecting a design system, structuring tokens, designing component APIs, choosing headless primitives, or setting up contribution/versioning.

2026-07-130

dec-discovery-validation

jpoindexter/dec-skills

Product discovery and validation frameworks — Jobs To Be Done, Continuous Discovery / Opportunity Solution Tree, Lean UX / Build-Measure-Learn, Design Sprint, Working Backwards, Story Mapping, Crazy 8s. Use when scoping what to build, validating an idea before committing, structuring user research, or turning a fuzzy outcome into testable solutions.

2026-07-130

name	dec-ai-native-patterns
description	AI-native design engineering patterns — generative/adaptive UI, streaming as interaction, prompt-as-interface, human-in-the-loop / agentic UX, eval-driven development, designing for uncertainty, latency and cost as design constraints. Use when designing any AI/LLM-powered feature, building agentic or chat UX, handling streaming/probabilistic output, or planning for model error and cost.
tags	["ai-native","llm","agentic-ux","streaming","evals","always-on"]

AI-Native Design Engineering

The new layer, increasingly central to the role. Probabilistic systems need UX that plans for being wrong.

Generative / Adaptive UI: Interfaces assembled or personalized at runtime by a model rather than fully predetermined.
Streaming as Interaction: Token-by-token output, partial results, and stop/regenerate controls are core interaction patterns, not edge cases.
Prompt-as-Interface: Natural language as an input modality alongside (not replacing) direct manipulation — with affordances that teach users what the system can do.
Human-in-the-Loop / Agentic UX: Design review, approval, correction, and undo around autonomous actions; surface confidence and provenance.
Eval-Driven Development: Treat model behavior as a testable surface — eval sets and regression checks for AI features, analogous to visual regression for UI.
Designing for Uncertainty: Probabilistic outputs need graceful failure, transparent limitations, and easy escape hatches. The system will be wrong; the UX plans for it.
Latency & Cost as Design Constraints: Model choice, caching, and streaming are UX decisions, not just infra ones.

How to apply

Make streaming first-class: show partial output immediately, always offer stop and regenerate.
For agentic actions, design the review/approve/undo loop before the autonomous capability — surface confidence and provenance.
Teach the user what the system can do via input affordances, not a blank prompt box.
Build an eval set early and run it like a regression suite; "looks good once" is not a test.
Plan the wrong-answer path (graceful failure + escape hatch) as carefully as the happy path.
Treat latency/cost/model choice as UX levers — cache, stream, and pick the smallest model that meets the bar.