惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

S
SegmentFault 最新的问题
B
Blog
P
Proofpoint News Feed
美团技术团队
The GitHub Blog
The GitHub Blog
Y
Y Combinator Blog
A
About on SuperTechFans
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Vercel News
Vercel News
有赞技术团队
有赞技术团队
小众软件
小众软件
H
Hackread – Cybersecurity News, Data Breaches, AI and More
Google DeepMind News
Google DeepMind News
Martin Fowler
Martin Fowler
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
aimingoo的专栏
aimingoo的专栏
H
Help Net Security
罗磊的独立博客
L
LangChain Blog
GbyAI
GbyAI
腾讯CDC
T
The Blog of Author Tim Ferriss
Microsoft Security Blog
Microsoft Security Blog

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
I Built Reusable Claude Code Skills to Ship Production We...
Matt Senter · 2026-05-15 · via DEV Community

I Built Reusable Claude Code Skills to Ship Production Websites Faster

Most AI-generated websites don’t fail because the homepage looks bad.

They fail because all the boring production details get skipped.

Things like:

  • SEO metadata
  • prerendering
  • analytics
  • social share images
  • CSP headers
  • Lighthouse optimization
  • mobile polish
  • sitemap generation
  • robots.txt
  • deployment configuration
  • IndexNow
  • caching
  • environment setup

The actual React components are often the easy part now.

The operational infrastructure is what still slows everything down.

After building a few AI-assisted projects with Claude Code, I realized I was repeatedly solving the same problems over and over again. Not just visually, but operationally.

So instead of writing larger prompts, I started building reusable Claude Code skills.

The result became an open source repository of production-oriented workflows:

senternet-site-skills on GitHub

The repository contains reusable production-focused skills for things like SEO, prerendering, mobile optimization, social sharing, CSP configuration, Lighthouse tuning, analytics setup, deployment workflows, and more.


The Problem With “Vibe Coding”

I actually like AI-assisted development.

A lot.

Claude Code is incredibly powerful for:

  • rapid iteration
  • frontend generation
  • restructuring layouts
  • writing utility code
  • integrating APIs
  • content scaffolding

But after the initial excitement wears off, a pattern emerges:

The AI can generate pages faster than you can operationalize them.

You end up spending huge amounts of time fixing:

  • SEO issues
  • deployment inconsistencies
  • broken metadata
  • poor mobile behavior
  • missing analytics
  • performance regressions
  • social preview problems
  • incomplete production setup

And ironically, these are often the least exciting tasks for humans to repeatedly perform manually.

That made them perfect candidates for reusable skills.


From Giant Prompts to Reusable Workflows

At first, I tried solving this with increasingly large prompts.

Things like:

“Please make sure this page is mobile responsive, optimized for SEO, uses prerendering, has proper metadata, social sharing support, CSP headers, analytics integration, and production-safe deployment settings…”

That approach quickly became unreliable.

Claude would focus heavily on one instruction while quietly ignoring another. Sometimes it would partially implement features. Other times it would regress working functionality while trying to “help.”

The breakthrough came when I stopped treating prompts like conversations and started treating them like infrastructure.

Instead of giant prompts, I created focused reusable skills:

  • senternet-site-metatags
  • senternet-site-prerender
  • senternet-site-mobile-optimize
  • senternet-site-share-images
  • senternet-site-csp
  • senternet-site-lighthouse
  • senternet-site-indexnow
  • senternet-site-firebase

Each skill had:

  • narrowly scoped responsibilities
  • deterministic expectations
  • operational guardrails
  • reusable implementation logic

The outputs became dramatically more stable.


The Most Important Idea: Operational Consistency

One thing I learned very quickly:

AI is surprisingly good at generating components.

It is much worse at consistently maintaining production infrastructure across multiple projects.

Humans naturally remember things like:

  • “Did we set OpenGraph tags?”
  • “Are we prerendering this route?”
  • “Did we configure robots.txt?”
  • “Will this social image crop correctly?”
  • “Are Lighthouse scores still acceptable?”
  • “Did analytics get added to the production layout?”

AI often forgets these details entirely unless explicitly guided.

That’s where reusable skills become powerful.

Instead of depending on memory or repetitive prompting, operational standards become encoded into reusable workflows.

The goal wasn’t full automation.

It was reducing forgotten work.


The “Uber Skill” Concept

One of the most useful patterns ended up being what I started calling “uber-skills.”

Instead of handling a single isolated task, these workflows orchestrate multiple setup steps together.

For example:

  • detect what’s already configured
  • skip completed setup
  • identify missing production features
  • apply only incremental improvements
  • avoid destructive rewrites

That last point became especially important.

One of the biggest failure modes with AI coding tools is:

asking for one change and getting an accidental full-project refactor.

The skills helped constrain that behavior significantly.

---
name: senternet-create-site
description: Orchestrate the full Senternet site build by running foundation, favicon, SEO, analytics, prerender, image, and performance skills in order.
---

# Create a Complete Optimized Marketing Site

Spin up or upfit a fully optimized marketing site by executing all site skills in sequence.

---

## Mode Detection

Before asking anything, check whether the user provided a path to an existing directory:

- **Existing directory** — run in **upfit mode**: navigate to that directory and detect what's already implemented before each step. Skip steps that are complete, patch steps that are partial.
- **No directory provided** — run in **create mode**: ask the intake questions below and build from scratch. If the user does not have a design zip, directory, or HTML export, default to a barebones Hello World site using the requested project and directory names.

## Upfit Feature Inventory

In upfit mode, surface a visible feature inventory before the optional phases so the user can see what is already enabled and what is still available to add.

- For each optional capability, detect whether the repo already has it.
- Report each one as either `enabled` or `available`.
- If one or more optional capabilities are `available`, present them to the user as a single enablement menu instead of separate yes/no prompts.
- Only offer options for capabilities that are not already enabled.

Optional capabilities to inventory in upfit mode:

- Transactional email via Resend
- Reddit pixel for ad campaigns
- Spanish `/es/` multilingual support
- Ad landing pages for paid campaigns
- SEO blog
- Competitor comparison / alternative pages
- reCAPTCHA Enterprise protection for forms

Enter fullscreen mode Exit fullscreen mode


What Actually Improved

The biggest gains weren’t:

  • writing code faster
  • generating prettier components
  • reducing typing

The biggest gains were:

  • fewer regressions
  • fewer forgotten deployment details
  • fewer SEO mistakes
  • fewer repetitive setup tasks
  • more consistent production readiness
  • reduced decision fatigue

In other words:

The AI became more useful once the workflow became more structured.


Real-World Usage

These workflows eventually became part of the production process behind projects like:

Both projects benefited from repeatedly applying the same operational standards:

  • metadata handling
  • mobile optimization
  • share image workflows
  • SEO structure
  • deployment consistency
  • performance optimization

Without reusable skills, I found myself re-solving the same infrastructure problems on every project.


Where AI-Assisted Development Still Struggles

Even with reusable skills, there are still clear limitations.

Claude Code can still:

  • over-refactor working code
  • hallucinate architecture decisions
  • regress responsive layouts
  • invent unnecessary abstractions
  • partially apply instructions
  • miss subtle UX inconsistencies

And frontend polish still requires human judgment.

A lot of human judgment.

But structured workflows dramatically reduce the chaos.


My Current Take

I increasingly think the future of AI-assisted development looks less like prompting and more like operational engineering.

The developers getting the best results probably won’t be:

  • the people writing the longest prompts
  • the people trying to remove all constraints
  • the people chasing fully autonomous agents

It’ll be the people building:

  • reusable workflows
  • constrained systems
  • composable tooling
  • deterministic infrastructure
  • operational guardrails

The real leverage comes from encoding consistency.

Not just generating code.


Final Thoughts

AI coding tools are already extremely capable.

But there’s still a huge difference between:

generating a demo

and:

repeatedly shipping production-ready websites.

For me, reusable Claude Code skills became a way to bridge that gap.

Not by replacing engineering discipline.

But by making it easier to apply consistently.