Skill v1.0.1
Automated scan100/100+10 new
version: "1.0.1" name: fullstack-expert description: "Expert-level fullstack development guide covering multi-language (TypeScript, Python, Go, Rust), multi-framework (Next.js, FastAPI, Gin, Axum), API design, microservices, DevOps, and system design / Panduan fullstack tingkat ahli mencakup multi-bahasa (TypeScript, Python, Go, Rust), multi-framework (Next.js, FastAPI, Gin, Axum), desain API, microservices, DevOps, dan system design." author: "Roedy Rustam"
Fullstack Expert (2026 Multi-Language Edition)
<a name="english"></a>
English
Orchestration & Integration
Connects and orchestrates with relevant domain skills like brainstorming, zero-to-prod-orchestrator, and project-context-mapper to ensure cohesive execution.
Description
Expert-level fullstack development across multiple languages and frameworks. Covers tech stack selection, API design (REST/gRPC/GraphQL), microservice architecture, DevOps, system design patterns, and AI-native backend integration for 2026.
Trigger Conditions
- Choosing the right language and framework for a new service or feature.
- Designing APIs between services in a multi-language architecture.
- Planning a microservices decomposition strategy.
- Implementing system design patterns (event sourcing, CQRS, saga pattern).
- Integrating AI agents or LLM capabilities into a backend system.
2026 Technology Matrix
Frontend
| Tool | Version | Notes | |
|---|---|---|---|
| React | 19.x | Compiler, Server Actions, use() hook | |
| Next.js | 15.x | PPR, RSC, App Router stable | |
| Astro | 5.x | Islands architecture, MDX, SSG | |
| TanStack Router | 1.x | Type-safe SPA routing | |
| TanStack Start | Beta | Full-stack Vite + RSC patterns | |
| Nuxt | 4.x | Vue 3 SSR/SSG, stable | |
| SvelteKit | 2.x | Lightweight, edge-ready |
Backend
| Language | Framework | Best For | |
|---|---|---|---|
| TypeScript | Hono, Fastify 5, NestJS | Serverless, edge, type-safe RPC | |
| Python | FastAPI 0.115+, Django 5 | AI/ML workloads, data pipelines | |
| Go | net/http (1.22+), Gin, Echo | High-throughput microservices, CLI | |
| Rust | Axum 0.8, Actix-web 4 | Performance-critical, WASM |
AI & Agents
| Tool | Role | |
|---|---|---|
| Vercel AI SDK 5.x | Full-stack AI streaming, RSC | |
| Mastra.ai | TS-first agent framework | |
| LangGraph | Python stateful agent workflows | |
| OpenAI Agents SDK | GPT-5 native agents + handoffs | |
| Google ADK | Gemini-powered agents | |
| Mem0 / MemGPT | Long-term agent memory |
Infrastructure
| Category | Tools | |
|---|---|---|
| Container | Docker, Podman | |
| Orchestration | Kubernetes, Railway, Fly.io | |
| CI/CD | GitHub Actions, Turborepo | |
| Observability | OpenTelemetry 1.x, Grafana, Langfuse | |
| Databases | PostgreSQL (primary), ClickHouse (analytics), Redis (cache), Qdrant (vector) |
API Design Standards
REST — OpenAPI 3.1
# openapi.yamlopenapi: "3.1.0"info:title: "My SaaS API"version: "1.0.0"paths:/api/v1/workspaces/{id}:get:summary: "Get workspace by ID"parameters:- name: idin: pathrequired: trueschema: { type: string, format: uuid }responses:"200":content:application/json:schema: { $ref: "#/components/schemas/Workspace" }"404":content:application/json:schema: { $ref: "#/components/schemas/ProblemDetail" }
gRPC — Protocol Buffers
syntax = "proto3";package user.v1;service UserService {rpc GetUser (GetUserRequest) returns (User);rpc ListUsers (ListUsersRequest) returns (stream User); // Server streamingrpc CreateUser (CreateUserRequest) returns (User);}message User {string id = 1;string email = 2;string name = 3;repeated string workspace_ids = 4;}
Microservice Patterns
Event-Driven (Saga Pattern)
Order Service ──publishes──> "order.created" ──> Payment Service│┌──────┴──────┐success? failure?│ │"payment.succeeded" "payment.failed"│ │Inventory Service Order Service(reserve stock) (cancel order)
CQRS (Command Query Responsibility Segregation)
- Write side: Commands update the primary PostgreSQL database.
- Read side: Events are projected into read-optimized views (denormalized tables or ClickHouse).
- Use when read patterns are very different from write patterns.
AI-Native Backend Integration (2026)
// Pattern: AI as a service within existing APIimport { generateObject } from 'ai';import { anthropic } from '@ai-sdk/anthropic';import { z } from 'zod';// AI endpoint that returns structured dataapp.post('/api/analyze', async (c) => {const { content } = await c.req.json();const { object } = await generateObject({model: anthropic('claude-4-sonnet'),schema: z.object({sentiment: z.enum(['positive', 'neutral', 'negative']),topics: z.array(z.string()),summary: z.string().max(200),}),prompt: `Analyze this content: ${content}`,});return c.json(object);});
SPA vs SSR vs Static — Decision Guide
SEO-critical + mostly read? → Next.js SSR / Astro 5 (static)Highly interactive dashboard? → SPA (TanStack Start / Vite + TanStack Router)→ see spa-orchestrator skill for architecture detailsMixed (marketing + app)? → Next.js 15 with hybrid routingReal-time data? → SSR + WebSocket or SSE streaming
Monorepo with Shared Types
// packages/types/src/index.ts — single source of truthexport interface User {id: string;email: string;name: string;plan: 'free' | 'pro' | 'enterprise';isSuperAdmin: boolean;}// apps/api & apps/web use the same typeimport type { User } from '@myapp/types';
Security Checklist (Fullstack)
- [ ] All user inputs validated with Zod on the server.
- [ ] JWT secrets rotated every 90 days; use short expiry + refresh tokens.
- [ ] CSP headers configured (no
unsafe-inlinein production). - [ ] All DB queries use parameterized queries.
- [ ] Rate limiting on all public API endpoints.
- [ ] Super Admin routes restricted to
admin.domain.comwithisSuperAdmincheck.
Code Review Standards
- Functions < 30 lines; files < 300 lines.
- No business logic in UI components.
- Every PR includes relevant tests (unit or E2E).
- No
anytypes in TypeScript production code. - Dependencies audited with
pnpm auditon every PR.
<a name="bahasa-indonesia"></a>
Bahasa Indonesia
Integrasi Orkestrasi
Terhubung dan mengorkestrasi skill domain yang relevan seperti brainstorming, zero-to-prod-orchestrator, dan project-context-mapper untuk memastikan eksekusi yang kohesif.
Deskripsi
Panduan pengembangan fullstack tingkat ahli di berbagai bahasa dan framework. Mencakup pemilihan tech stack, desain API (REST/gRPC/GraphQL), arsitektur microservice, DevOps, pola desain sistem, dan integrasi backend AI-native untuk 2026.
Kondisi Pemicu
- Memilih bahasa dan framework yang tepat untuk layanan atau fitur baru.
- Merancang API antar layanan dalam arsitektur multi-bahasa.
- Merencanakan strategi dekomposisi microservice.
- Mengimplementasikan pola desain sistem (event sourcing, CQRS, saga pattern).
- Mengintegrasikan agen AI atau kemampuan LLM ke dalam sistem backend.
Matriks Teknologi 2026
Frontend
React 19 + Next.js 15 (default SaaS), Astro 5 (konten), TanStack Router (SPA), Nuxt 4 (Vue), SvelteKit 2.
Backend
TypeScript/Hono (serverless, edge), Python/FastAPI (AI/ML), Go/net/http (throughput tinggi), Rust/Axum (kritis performa).
AI & Agen
Vercel AI SDK 5.x, Mastra.ai, LangGraph, OpenAI Agents SDK, Google ADK, Mem0/MemGPT untuk memori agen jangka panjang.
Infrastruktur
Docker/Kubernetes, GitHub Actions + Turborepo, OpenTelemetry 1.x, Langfuse. Database: PostgreSQL (utama), ClickHouse (analitik), Redis (cache), Qdrant (vektor).
Standar Desain API
- REST + OpenAPI 3.1: Untuk API publik dengan beberapa konsumen.
- gRPC: Untuk komunikasi antar layanan internal berkinerja tinggi.
- Hono RPC / tRPC: Untuk komunikasi full-stack TypeScript type-safe tanpa codegen.
Pola Microservice
Event-Driven (Saga Pattern)
Orkestrasi transaksi terdistribusi melalui event — setiap layanan mempublikasikan event setelah berhasil; kegagalan memicu kompensasi di seluruh rantai.
CQRS
Pisahkan sisi tulis (PostgreSQL) dari sisi baca (view terdenormalisasi atau ClickHouse) ketika pola baca sangat berbeda dari pola tulis.
Integrasi Backend AI-Native (2026)
Gunakan Vercel AI SDK generateObject() dengan skema Zod untuk mengintegrasikan kemampuan AI ke endpoint API yang ada — mengembalikan output terstruktur yang dijamin sesuai tipe.
SPA vs SSR vs Static
Gunakan SSR/Astro untuk situs kritis SEO dan banyak baca. Gunakan SPA (TanStack Start) untuk dashboard yang sangat interaktif. Gunakan Next.js 15 dengan routing hybrid untuk aplikasi campuran (marketing + app).
Monorepo dengan Shared Types
Definisikan interface dan tipe bersama di packages/types — digunakan oleh semua app (web, admin, api, mobile) sebagai single source of truth.
Checklist Keamanan & Code Review
- Input pengguna divalidasi Zod di server.
- Header CSP dikonfigurasi & parameterized query untuk DB.
- Rate limiting di semua endpoint publik.
- Rute Super Admin dibatasi di subdomain dengan cek
isSuperAdmin. - Tidak ada logika bisnis di komponen UI; wajib ada unit/E2E test pada PR.
- Tidak ada tipe
anypada kode TypeScript produksi.