How to up-level your content creation with Claude Skills
Produce a full website sitemap, 44 pages of content, internal/external linking, and clean markdown in a single evening with your own Claude Code project.
I’ve been using Claude.ai, Claude Code, and Claude’s plugins for PowerPoint and Excel heavily the past few weeks. Before that, I was—and still am—using ChatGPT, Gemini, and Claude on a daily basis.
A friend who built and sold a service business outside tech recently used Claude Code to build an entire back-office system for his new business. He isn’t a coder, just a high-agency individual—driven and always looking to learn new things. This was an eye-opening moment for me. If people outside of software engineering are building applications to run real businesses without the ability to code, it was time to double down on what’s possible.
Open a chat with Claude.ai to get started, here was the prompt i used.
I want Claude to help me build out webpage content based on google business profile categories. I’ve got a process documented with instructions on how to build the content. How should I set up the Claude code environment with skills, Claude.md, or other guardrails to act a system of agents working on the research and content creation, similar to a human agency?
Here is what Claude came back with:
Claude Code Environment Architecture
1. CLAUDE.md (Project Root)
This is your master orchestration file — think of it as the “agency playbook.” It should define:
Project overview — what the system does (generates webpage content mapped to Google Business Profile categories)
Agent roles — define distinct personas/responsibilities (more on this below)
Workflow sequence — the order of operations from research → draft → review → publish-ready output
Global rules — tone, brand voice, SEO constraints, output formats, quality standards
File conventions — where inputs live, where outputs go, naming conventions
2. Skills as “Agent Specializations”
Create a /skills/ directory where each skill folder represents an agent role:
/skills/
├── researcher/
│ └── SKILL.md # Instructions for category research, competitor analysis, keyword extraction
├── content-strategist/
│ └── SKILL.md # Maps GBP categories to page structures, defines content outlines
├── copywriter/
│ └── SKILL.md # Writes the actual page content following brand/SEO guidelines
├── seo-editor/
│ └── SKILL.md # Reviews for keyword density, meta tags, schema markup, internal linking
├── qa-reviewer/
│ └── SKILL.md # Final quality check — grammar, accuracy, brand compliance
└── templates/
├── service-page.md
├── location-page.md
└── category-landing.mdEach SKILL.md should contain:
Role definition — “You are acting as a [role]...”
Inputs expected — what this agent receives
Process steps — exactly how to do the work (paste your documented process here)
Output format — structured deliverable (markdown, JSON, etc.)
Quality criteria — what “done” looks like
3. Workflow Orchestration via CLAUDE.md
Define the pipeline explicitly:
## Workflow
### Phase 1: Research (skills/researcher)
- Input: GBP category name + business info
- Output: `/research/{category}-research.md` (keywords, competitor insights, search intent)
### Phase 2: Strategy (skills/content-strategist)
- Input: Research output + template
- Output: `/outlines/{category}-outline.md` (page structure, H1-H3, CTA placement)
### Phase 3: Draft (skills/copywriter)
- Input: Outline + brand voice guide
- Output: `/drafts/{category}-draft.md`
### Phase 4: SEO Review (skills/seo-editor)
- Input: Draft
- Output: `/drafts/{category}-seo-reviewed.md` (with inline comments)
### Phase 5: QA (skills/qa-reviewer)
- Input: SEO-reviewed draft
- Output: `/final/{category}-final.md`4. Invoking It
When you run Claude Code, you’d prompt something like:
“Process the GBP category ‘Emergency Plumbing Service’ through the full content pipeline. Follow the workflow in CLAUDE.md, using each skill in sequence. Save all intermediate outputs.”
This gives you traceability (every phase produces a file) and the ability to intervene at any stage.
Below is the full system I built yesterday that created a sitemap, researched the google business profile categories, created internal and external links, reviewed the content for fact checking and no-ai readability, and outputted markdown files for uploading.
I highly recomend directing Opus 4.6 and directing it to leverage Sonnet 4.6 for sub-agents. The orchestration agent will continue to leverage Opus 4.6 for thinking and checking sub tasks. If Claude Code seem daunting to you, just open a chat with Claude.ai and start asking how to build a similar project. You’ll be shocked how much you can acomplish by asking.
Happy building!
How my Web Content System Works
This posts explains how the Handyman website content production system is built and how it turns a business profile into a full website’s worth of SEO-optimized pages.
Part 1: How CLAUDE.md and SKILL.md Files Work
What is CLAUDE.md?
CLAUDE.md is a plain-text instruction file that sits at the root of a project folder. When Claude (the AI) opens a project, it reads this file first — before doing anything else. Think of it as a playbook. It tells Claude what the project is, what rules to follow, and how to do the work.
In this project, CLAUDE.md defines:
The site architecture — a template showing how pages are organized (homepage at the top, category pages below it, service pages below those, and so on)
The production pipeline — seven phases that every page moves through, from planning to final review
Content rules — reading level, sentence length limits, voice and tone, and a list of banned words and phrases
Page types — five types of pages the system can produce, each with a different role and format
How to run the system — commands for building a full site, adding pages, resuming mid-project, or checking progress
CLAUDE.md is the single source of truth. Every phase of the pipeline refers back to it. If a rule is in CLAUDE.md, it applies everywhere.
What are SKILL.md Files?
SKILL.md files are specialized instruction sets — one per phase of the pipeline. Each SKILL.md lives in its own folder under skills/. When Claude enters a specific phase of work, it reads the corresponding SKILL.md to know exactly what to do, what to check, and what to produce.
Think of CLAUDE.md as the company handbook and each SKILL.md as a job description for a specific role. The handbook applies to everyone; the job description tells one role exactly how to do its part.
This project has seven SKILL.md files:
Each SKILL.md contains the specific steps, templates, checklists, and rules that its phase requires. For example, the outline strategist’s SKILL.md contains four different outline templates — one for each page type — so the outline format automatically matches whether the page is a service page, a neighborhood page, or a supporting article.
Part 2: How the Project Processes Inputs Into Finished Pages
What Goes In
The system starts with three input files:
Client Profile (inputs/client-profile.md) — The business’s Google Business Profile data: business name, city, primary category, secondary categories, services offered, neighborhoods served, and landmarks nearby.
Authority Domains (inputs/authority-domains.md) — A list of approved websites (mostly .gov and .edu sources) that can be used for external citations. Each page gets one external link from this list.
Internal Links (inputs/internal-links.md) — An inventory of every page that has been completed so far, with its URL, title, and a short description. This file grows as pages are finished. New pages link to existing ones, and the inventory is updated so future pages can link back.
What Comes Out
For each page, the system produces files at each stage:
research/{slug}-research.md — Research brief with PAA questions, local details, and competitor analysis
outlines/{slug}-outline.md — Structured outline with H2/H3 headings, section purposes, and word targets
drafts/{slug}-draft.md — Progressive draft built section by section
drafts/{slug}-factcheck.md — Fact-check report listing every claim and its verification status
drafts/{slug}-linked.md — Draft with internal and external links inserted
final/{slug}-final.md — Production-ready content with meta tags, quality report, and similarity checks
The final file is what goes to the client. It includes the finished content, two to three meta title options, two to three meta description options, a quality report, and a status marker.
The Site Map
The site map (sitemap/sitemap.md) is the master tracking document. It lists every page in the site, its position in the hierarchy, its target keyword, its production status, and its linking relationships. The site map is created in Phase 0 and updated after every page is completed.
Part 3: The Seven Phases
Every page moves through the same seven-phase pipeline. Each phase has a specific purpose, a specific output, and a checkpoint before the next phase begins.
Phase 0 — Site Planning
Purpose: Turn the client’s business profile into a complete site architecture.
What happens:
The site planner reads the client profile and maps every secondary category to a category page
Every service is assigned to its parent category as a service page
High-priority services that don’t fit neatly under one category are elevated to top-level pages
Neighborhood and landmark pages are planned from the client’s service area
Supporting content (educational articles) is planned for the highest-value services
A production order is set — the most impactful pages are built first
The full site map is saved and presented for approval
Output: sitemap/sitemap.md
Checkpoint: The site map must be approved before any pages are written.
Phase 1 — Research
Purpose: Gather everything needed to write a specific page.
What happens:
The researcher checks the site map to understand where this page fits — its parent, its siblings, and its children
Input requirements are validated based on the page type (service pages need PAA questions; neighborhood pages need geo targets; etc.)
PAA (People Also Ask) questions are identified for the topic
Local details relevant to McAllen and the Rio Grande Valley are gathered
The page’s linking context is noted — what should link to this page and what this page should link to
Output: research/{slug}-research.md
Phase 2 — Outline
Purpose: Create a structured blueprint for the page before any writing begins.
What happens:
The outline strategist reads the research brief and selects the correct template based on the page type
Category pages get a template with service listing sections
Service pages get a template with cause/solution sections, a DIY-vs-professional section, and an FAQ
Neighborhood pages get a template focused on local housing stock and common service needs
Supporting content gets a template built around answering a specific question in depth
Every section in the outline has a purpose statement, a target word count, and notes on what to include
Output: outlines/{slug}-outline.md
Checkpoint: The outline must be approved before writing begins.
Phase 3 — Content Writing
Purpose: Write the page content section by section, following the approved outline exactly.
What happens:
The content writer works through the outline one section at a time
Every sentence must be 20 words or fewer
Every paragraph is 2 to 4 sentences
The reading level targets Grade 5-6
The voice is “we/our” (the business) speaking to “you” (the reader) — practical, calm, and local
A banned word list of 80+ words and phrases is enforced (no marketing fluff, no AI-sounding transitions)
The similarity guard is active — if a sentence could appear on another page in the site, it gets rewritten with specifics
Output: drafts/{slug}-draft.md
Checkpoint: Each section is reviewed before the next one is written.
Phase 4 — Fact Check
Purpose: Verify every factual claim in the draft and add citations.
What happens:
First pass: Every fact, statistic, measurement, and technical claim in the article is listed alongside its source
Second pass: The draft is re-read to catch any facts that were referenced but not yet cited
Unverifiable claims are flagged for removal or rewording
Citations are added in numbered format (e.g., [1]) with full source details at the bottom of the page
Output: drafts/{slug}-factcheck.md (report) + updated draft with citations
Phase 5 — Link Curation
Purpose: Connect this page to the rest of the site and to one external authority source.
What happens:
The link curator reads the site map to understand this page’s linking hierarchy:
Category pages link down to their child service pages
Service pages link up to their parent category and down to supporting content
Supporting content links up to its parent service page
Neighborhood pages link to relevant services and categories
Sibling pages can cross-link where it makes sense in context
The internal link inventory is checked for all available link targets
Links are inserted naturally within the content — no forced link blocks or lists
One external link is selected from an approved authority domain (typically .gov or .edu)
The internal link inventory is updated with this new page as a target for future pages
Output: drafts/{slug}-linked.md + updated inputs/internal-links.md
Phase 6 — Quality Review
Purpose: Final audit for AI detection risk, content quality, and production readiness.
What happens:
AI detection assessment: The content is evaluated for patterns that make text sound AI-generated — uniform sentence length, hedge phrases, filler transitions, generic structure. If risk is elevated, specific rewrite strategies are applied (varying sentence rhythm, adding local specifics, breaking predictable patterns).
Similarity check: The page is compared against every sibling page in the site map (pages at the same hierarchy level). Sentence-level overlap is flagged. The target is less than 20% overlap with any related page.
Quality checklist (17 points): Reading level, sentence length, paragraph length, banned words, voice, geo mentions, price claims, keyword placement, internal links, external links, FAQ format, scope boundaries, and completeness are all verified.
Meta tags: Two to three title options and two to three description options are generated with character counts.
Output: final/{slug}-final.md + updated sitemap/sitemap.md (page marked COMPLETE)
Summary
This system turns a single business profile into a complete set of website pages by running each page through a structured seven-phase pipeline. CLAUDE.md provides the rules and architecture that apply everywhere. Seven SKILL.md files provide the specific instructions for each phase. The site map tracks progress and relationships between pages. The internal link inventory grows as each page is completed, connecting the site together.
For this website, this system produced 44 pages: 4 category pages, 23 service pages, 7 neighborhood pages, 5 supporting content articles, and 5 additional service pages — all following the same pipeline, all cross-linked, and all checked against each other for quality and uniqueness.