convez.ai
SkillDesign & UIIntermediário
jpoindexter

Design & AI Skills — Biblioteca de Referência

jpoindexter/design-and-ai-skills

40 skills de referência para design e IA, portáveis em SKILL.md para agentes Claude

Ver no GitHub
InicialPluginScore · de 0 a 100

Coleção de 40 skills de nível production para design de produtos e engenharia de IA, organizadas em dois acervos: 27 skills de design system (tokens, tipografia, acessibilidade, componentes) e 13 skills de engenharia de IA (LLMs, agents, observabilidade). Cada skill é um arquivo SKILL.md autossuficiente com referências a padrões reais (WCAG, Material Design, RAGAS, OWASP), números concretos, tabelas de decisão e código portável para Claude Code e agentes compatíveis.

Instalação

$npx convez-ai jpoindexter/design-and-ai-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
2
Forks
2
Acompanhando
0
Saúde
28%
Último commit
há 3 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Implementar design systems com tokens W3C e parity Figma-to-code
  • Estruturar pipelines de produção para LLMs e sistemas de agents
  • Revisar acessibilidade (WCAG 2.2, APCA) em interfaces de produto
  • Arquitetar observabilidade, evals (RAGAS) e rastreamento de LLMs
  • Documentar padrões de design escaláveis para múltiplas plataformas

Veredito

Biblioteca sólida e bem estruturada de skills técnicos com foco em padrões reais (W3C, WCAG, OWASP) em vez de dicas genéricas. Ideal para times que já têm contexto de design systems e engenharia de IA e precisam de referências confiáveis e copy-pasteable; requer esforço para integração inicial, mas entrega alto valor em manutenção de padrões.

README

A tradução para o português deste README ainda está na fila.

Design & AI Engineering Skills

40 reference-grade skills for product design and production AI engineering — self-contained SKILL.md files that load into Claude Code, Codex, or any agent that reads a skills directory.

Two libraries: 27 design-system skills covering the full UX/design canon, and 13 AI engineering skills covering production LLM and agent systems.


What these are

Each skill is a single SKILL.md file with a YAML frontmatter block and a deep reference body. The body is written as an authoritative reference, not a tips list: real numbers, formulas, comparison tables, copy-pasteable code, do/don't pairs, named failure modes, and attribution to the actual standards and papers (Nielsen/Norman/WCAG, vLLM/PagedAttention, RAGAS, OWASP LLM Top 10, W3C DTCG, and so on).

When a skill is installed, the agent loads the full content into its context at the start of the relevant task — so it works from the actual standard, not a summary of a summary.


SKILL.md format

---
name: slug-matching-folder-name
description: One-line summary used by the agent to decide relevance.
tags: [category, topic, subtopic]
---
# Title

Body — markdown, tables, code blocks.

Frontmatter fields:

FieldRequiredNotes
nameyesMust match the containing folder name exactly
descriptionyesUsed for skill discovery; be specific
tagsyesArray of lowercase strings

Body conventions used in this library:

  • Sections numbered (§1, §2…) with a clear heading
  • Comparison tables for any decision with >2 options
  • Concrete numbers (budgets, thresholds, sizes, timings) wherever they exist
  • Explicit do/don't pairs for the most common mistakes
  • Named failure modes, not vague warnings
  • Attribution to the originating standard, paper, or framework (with year where useful)

Repository structure

design-and-ai-skills/
├── README.md                        ← you are here
├── index.html                       ← root viewer (both libraries)
├── scripts/
│   ├── build-skills-index.py        ← rebuild ai-engineering-skills/index.html
│   └── build-design-skills-index.py ← rebuild design-system-skills/index.html
├── design-system-skills/
│   ├── README.md                    ← skill list for this library
│   ├── index.html                   ← standalone viewer
│   └── <slug>/SKILL.md              ← one folder per skill (27 total)
└── ai-engineering-skills/
    ├── README.md                    ← skill list for this library
    ├── index.html                   ← standalone viewer
    └── <slug>/SKILL.md              ← one folder per skill (13 total)

Design System Skills — 27 skills

For any screen: web, iOS/Android, macOS/Windows, tablet, TV, watch, foldable. Current to 2025–2026 standards (WCAG 2.2, Material 3, Apple HIG, W3C DTCG, OKLCH, container queries, view transitions).

Foundations

SkillWhat it covers
design-tokensW3C DTCG format, 3-tier primitive→semantic→component model, Style Dictionary v4, Figma↔code parity, multi-brand theming
typography-systemModular and fluid scales, clamp(), variable fonts, OpenType, baseline rhythm, accessible ramps, cross-platform tokens
color-and-elevationOKLCH ramps, WCAG/APCA contrast, semantic tokens, elevation-as-lightness dark mode, copy-pasteable CSS
grid-and-spacing8pt spacing scale, column/baseline grids, responsive breakpoints, modern CSS Grid and container queries
iconography-and-imageryIcon grids, optical sizing, variable axes, SVG delivery, aspect ratios, art direction, modern image formats

Structure

SkillWhat it covers
atomic-designBrad Frost's atoms→molecules→organisms→templates→pages, Figma/code component tree organization, abstraction levels
layout-and-compositionVisual hierarchy, Gestalt, alignment, whitespace, scanning patterns, intrinsic CSS (Grid/Flexbox/subgrid/:has)
components-and-statesComponent anatomy, full interaction-state matrix, variants vs sizes vs states, sizing tokens, headless/data-state patterns
design-system-governanceOperating models, contribution and release process, Figma↔code parity, documentation, adoption metrics, anti-patterns
design-system-frameworksMaterial 3, Apple HIG, Fluent 2, Carbon, Polaris, Primer, Ant, Spectrum; Tailwind, shadcn/ui, Radix, React Aria; adopt-vs-adapt decision

Interaction & device

SkillWhat it covers
interaction-and-motionDuration scales, easing tokens, choreography, view transitions, scroll-driven animations, @starting-style, prefers-reduced-motion
forms-and-data-entrySingle-column layout, persistent labels, input type + keyboard + autocomplete per field, :user-invalid validation, multi-step wizards
responsive-and-multi-deviceMobile-first breakpoints, fluid clamp() type, container queries, logical properties, dvh/safe-area, phone/tablet/TV/watch/foldable
platform-conventionsiOS/HIG, Android/Material 3, macOS, Windows 11 Fluent — concrete numbers, comparison tables, adapt-vs-custom-brand decision

UX & inclusion

SkillWhat it covers
usability-heuristicsNielsen's 10 heuristics, Norman's fundamentals, Shneiderman's 8 golden rules, Rams' 10 principles, heuristic evaluation method
laws-of-ux-and-psychologyFitts, Hick-Hyman, Miller, Jakob, Tesler, Postel, Doherty, Peak-End, behavioral econ cluster — each with origin, UI application, design rule, pitfall
ux-writing-and-contentVoice and tone systems, every UI string type with before/after, terminology consistency, inclusive language, localization readiness
accessibility-and-inclusive-designWCAG 2.2 AA/AAA, APCA, keyboard/screen-reader/touch/motor, forms, motion/cognition, testing, EAA/Section 508 legal floor
inclusive-designMicrosoft's 3 principles + Persona Spectrum, social model of disability, cognitive/neurodivergent/sensory/cultural/economic/age patterns beyond WCAG
information-architecture-and-navigationOrganization/labeling/navigation/search systems, taxonomy depth-vs-breadth, nav patterns per device, wayfinding, card sort/tree test/first-click
data-visualizationPerceptual encoding ranking, chart-by-intent selection, data color scales, Tufte's data-ink discipline, accessible charts, interaction patterns
ethical-design-and-dark-patternsBrignull/Mathur dark-pattern taxonomy, every pattern with example + fix, consent/privacy UX, attention ethics, FTC/GDPR/DSA legal exposure
performance-and-perceived-speedCore Web Vitals (LCP/INP/CLS budgets and fixes), optimistic UI, skeletons, prefetch-on-intent, loading states, field vs lab measurement at p75
onboarding-and-empty-statesEvery view state (empty/loading/error/offline/success/blocked), first-run flows, show-don't-tell, seed data, activation to aha moment
multimodal-voice-and-hapticsTouch/visual/voice/audio/haptic modality combinations, conversational/LLM-agent UX, voice error recovery, iOS/Android haptics, TV/watch/AR
internationalization-and-localizationText expansion, RTL/bidi with CSS logical properties, Intl formatting, ICU pluralization, script typography, pseudolocalization, translation workflow
brand-identity-and-design-principlesOpinionated design principles, logo systems, attributes-to-visuals translation, brand-in-product, brand→token bridge, guidelines, rebrands

AI Engineering Skills — 13 skills

For engineers shipping LLM and agent systems in production. Grounded in specific systems and papers: vLLM/PagedAttention, FlashAttention, AWQ/GPTQ/SmoothQuant, RAGAS, OWASP LLM Top 10, OpenTelemetry GenAI.

Context & harness

SkillWhat it covers
harness-and-context-engineeringThe control loop, tools, memory, retries, budgets; context as a managed token budget; lost-in-the-middle, context rot, just-in-time retrieval, compaction, sub-agent isolation
adaptation-strategiesChoosing between ICL, RAG, fine-tuning (LoRA/QLoRA/DPO), and distillation; data requirements; catastrophic forgetting; combining methods

Inference & serving

SkillWhat it covers
inference-performanceTTFT/TPOT/throughput, prefill vs decode, the roofline model, continuous batching, PagedAttention, chunked prefill, disaggregation, FlashAttention, GPU sizing
inference-caching-and-kvAnthropic cache_control, OpenAI prefix caching, Gemini implicit/explicit; semantic caching; KV-cache internals (PagedAttention, RadixAttention, eviction, multi-tenant safety)
quantization-and-model-compressionFP8/INT8/INT4, GPTQ/AWQ/SmoothQuant/NF4/GGUF k-quants, KV quantization, speculative decoding (Medusa/EAGLE/n-gram), distillation, quality cliffs

Reliability & agents

SkillWhat it covers
structured-output-and-tool-callingJSON mode, constrained/grammar-guided decoding, json_schema coercion, Zod/Pydantic validation, validate→repair→escalate→fallback loop, idempotency for retries
agent-reliability-and-guardrailsPlan→act→observe failure physics, five budgets (iteration/tool-call/token/wall-clock/cost), runaway prevention, circuit breakers, three guardrail layers, checkpoint/resume
model-routing-and-fallbackRouting by cost/latency/quality, cascade escalation, multi-provider fallback, circuit breakers, hedging, degraded-mode UX

Retrieval & evals

SkillWhat it covers
rag-architectureFull ingest→chunk→embed→index→retrieve→rerank→assemble→generate pipeline, chunking strategies, hybrid search, reranking, context assembly, failure modes
llm-evals-and-retrieval-qualityGolden/regression/adversarial eval sets, LLM-as-judge biases, recall@k/MRR/nDCG, grounding/faithfulness/attribution, RAGAS scoring, CI gates

Ops, cost & safety

SkillWhat it covers
llm-observability-and-costDistributed traces and spans, OpenTelemetry GenAI conventions, drift detection, per-feature/tenant/user cost attribution, LangSmith/Langfuse/Phoenix/Helicone
llm-safety-and-multitenancyOWASP LLM Top 10, direct/indirect prompt injection, privilege separation, dual-LLM pattern, data-leakage prevention, cross-tenant cache/index/memory isolation
production-failure-modes-and-tradeoffsFull failure taxonomy (symptom/root-cause/detection/mitigation), latency/quality/cost/reliability tradeoff map, production-readiness checklist, incident response

Install

Fresh clone (recommended)

git clone https://github.com/jpoindexter/design-and-ai-skills
cd design-and-ai-skills
./bootstrap.sh

bootstrap.sh does two things in one step:

  1. Installs all 40 skills to ~/.claude/skills/
  2. Wires a post-merge git hook so future git pulls auto-reinstall

Reload Claude Code after running. You're done — no manual steps needed after this.

Existing clone / manual install

./install.sh

Finds every SKILL.md under design-system-skills/ and ai-engineering-skills/ and copies each to ~/.claude/skills/<slug>/. Use this if you cloned before bootstrap.sh existed, or want to reinstall without pulling.

Preview what will be installed first:

./install.sh --dry-run

Keeping skills up to date

Once bootstrap.sh has been run once, git pull handles everything:

git pull  # post-merge hook fires automatically → all 40 skills reinstalled

No manual install.sh needed after a pull.

Install a single skill

cp -R design-system-skills/usability-heuristics ~/.claude/skills/

Other agents

Each <slug>/ folder contains one SKILL.md. Copy the folder into whatever directory your agent reads for skills. The format (frontmatter + markdown body) is not Claude-specific — any agent that can read a skills directory with YAML frontmatter will work.


Browse

Each library ships a self-contained index.html viewer — no server, no build, works from file://.

ViewerPath
Both librariesdocs/index.html
Design system skillsdocs/design-system-skills.html
AI engineering skillsdocs/ai-engineering-skills.html

Open any of them directly in a browser. Grouped sidebar, filter box, collapsible skill cards.


Regenerate the viewers

After adding or editing a skill, rebuild the corresponding index.html. Requires mistune:

pip install mistune

# Rebuild AI engineering viewer → docs/ai-engineering-skills.html
python3 scripts/build-skills-index.py ai-engineering-skills

# Rebuild design system viewer → docs/design-system-skills.html
python3 scripts/build-design-skills-index.py

Adding a skill

  1. Create <library>/<slug>/SKILL.md — slug must be kebab-case and match the name: in frontmatter.
  2. Fill in the frontmatter (name, description, tags).
  3. Write the body. Conventions: numbered sections, comparison tables for multi-option decisions, concrete numbers where they exist, attribution to originating standards/papers, named failure modes, do/don't pairs.
  4. Rebuild the viewer for that library (see above).
  5. Add the skill to this README and to <library>/README.md.

License

MIT — use, adapt, share.

Ver o original no GitHub

Bom para

Relacionados