curiositech/frontend-architect

Frontend Architect

Expert in frontend stack decisions, Cloudflare deployment, and bridging design systems to component implementations.

Decision Points

Framework Selection Tree

1. Performance Budget Assessment

If bundle size must be <100KB:
  → Use Astro (islands architecture)
If first contentful paint must be <800ms:
  → Use Astro or SvelteKit
If you need full SSR + interactivity:
  → Use Next.js or Remix
If team only knows React:
  → Use Next.js or Vite + React

2. Deployment Target Decision

If deploying to Cloudflare Pages:
  → Astro (native) > Next.js (OpenNext) > SvelteKit
If need edge compute + database:
  → Next.js + Turso or Astro + D1
If team uses Vercel:
  → Next.js (best DX) > Astro
If self-hosting:
  → Any framework, consider maintenance overhead

3. Team Capability Assessment

If team <3 developers:
  → Pick familiar stack, avoid learning curves
If team >5 developers:
  → Consider TypeScript monorepo (Turborepo)
If designers hand off static mockups:
  → Astro + MDX for content
If designers work in Figma with dev mode:
  → Next.js + shadcn/ui

4. Content Type Classification

If >80% static content:
  → Astro
If heavy user interactions (dashboards, SPAs):
  → Next.js App Router or Vite + React
If marketing site with CMS:
  → Astro + Sanity/Contentful
If internal tool prototype:
  → Next.js (faster iteration) or Vite (faster builds)

Failure Modes

Schema Bloat

Detection: Bundle size >500KB from framework overhead alone Diagnosis: Using full-stack framework for static content Fix: Switch to Astro. If you see hydration for simple interactions, you're over-engineering

Deployment Mismatch

Detection: Build takes >5 minutes or frequent deployment failures Diagnosis: Wrong adapter for platform (e.g., Node.js app on edge runtime) Fix: Next.js on Cloudflare needs OpenNext. Check runtime compatibility early

Access Control Theatre

Detection: Complex auth logic in frontend for internal tools Diagnosis: Implementing security in wrong layer Fix: Use Cloudflare Access for auth, simple role checks in app. Don't build OAuth from scratch

Framework FOMO

Detection: Considering migration without clear performance/DX problems Diagnosis: Chasing trends instead of solving real problems Fix: Measure current pain points. Only migrate if you can quantify 2x improvement

Premature Microservices

Detection: 3+ repos for a team of <5 developers Diagnosis: Over-engineering for future scale that may never come Fix: Consolidate to monorepo. Split when you have dedicated teams per service

Worked Examples

Example 1: Content Site with CMS (Next.js vs Astro)

Scenario: Marketing site, 20 pages, blog, team of 2 developers, Contentful CMS

Decision Process:

1. Performance requirement: Marketing needs <90 Lighthouse score
2. Content type: 80% static, 20% interactive (newsletter signup, contact forms)
3. Team skill: React familiar, never used Astro
4. Timeline: 4 weeks to launch

Framework Analysis:

Next.js Route:
- Pros: Team familiar, rich ecosystem, good CMS integrations
- Cons: Bundle size ~200KB, hydration cost, overkill for mostly static
- Performance: 75-85 Lighthouse (hydration penalty)

Astro Route:
- Pros: <50KB bundle, 95+ Lighthouse, partial hydration
- Cons: 1-week learning curve, smaller ecosystem
- Performance: 90-100 Lighthouse

Decision: Choose Astro

• Performance requirement (>90 Lighthouse) eliminates Next.js
• 1-week learning curve acceptable for 4-week timeline
• Use Astro with React islands for forms (familiar components)

Implementation:

npm create astro@latest -- --template minimal
npx astro add react tailwind
# Use .astro for pages, .tsx for interactive islands

Example 2: Internal Tool Prototype

Scenario: Admin dashboard prototype, 5 beta users, 2-week sprint

Decision Process:

1. Polish requirement: 70% (internal tool, not customer-facing)
2. Iteration speed: Critical (prototype validation)
3. Access control: Must restrict to beta users
4. Tech debt tolerance: High (will rebuild if successful)

Architecture Decision:

Framework: Next.js (fast iteration, familiar)
Styling: shadcn/ui (components without design time)
Auth: Cloudflare Access (zero auth code)
Deployment: Cloudflare Pages with protection
Database: Turso (edge-compatible, easy setup)

Access Control Setup:

1. Cloudflare Access Application:
   - Domain: internal-admin.pages.dev
   - Policy: Allow emails from allowlist
   
2. Feature flags in app:
   - Check cf-access-authenticated-user-email header
   - Grant features based on email (admin vs beta)

What expert catches vs novice misses:

• Novice: Builds custom auth, spends 50% of sprint on login
• Expert: Uses Cloudflare Access, focuses 100% on core features

Quality Gates

• [ ] Framework choice justified with performance/team/timeline constraints
• [ ] Deployment target verified with test deploy (not just documentation)
• [ ] Bundle size measured and under target (<200KB for marketing, <500KB for apps)
• [ ] Team has completed hello-world in chosen framework
• [ ] Access control strategy defined for internal tools
• [ ] Performance budget established with Lighthouse targets
• [ ] Migration path documented if changing from existing stack
• [ ] Cost estimate including hosting, CDN, and edge compute usage
• [ ] Rollback plan defined for deployment issues
• [ ] Component library integration tested with actual design tokens

NOT-FOR Boundaries

Don't use this skill for:

• Writing component styling → Use frontend-developer instead
• Design critique or visual assessment → Use design-critic instead
• Backend API architecture → Use api-architect instead
• Database schema design → Use database-architect instead
• CSS debugging or animation → Use frontend-developer instead
• DevOps beyond deployment config → Use devops-automator instead
• Performance optimization of existing code → Use performance-optimizer instead

Delegation triggers:

• If discussing CSS selectors or animations → frontend-developer
• If evaluating visual hierarchy or user experience → design-critic
• If designing REST/GraphQL endpoints → api-architect
• If optimizing database queries → database-architect