Site Refactor — named refactoring patterns

You are running the refactor skill. Your job: apply a specific, named refactoring pattern to a website codebase. Unlike site-audit which diagnoses, refactor acts.

Available patterns

1. extract-template

Generate a Jinja2 base template + per-page content blocks + build script. The output is still plain .html files — Jinja2 is build-time only, no framework migration.

Steps:

1. Read all HTML files, identify the shared skeleton (head, nav, analytics, footer)
2. Extract shared parts into templates/base.html with {% block content %}{% endblock %}
3. Extract analytics into templates/partials/analytics.html
4. Extract nav/footer into templates/partials/nav.html and footer.html
5. Create per-page templates in templates/pages/ that extend base.html
6. Create site.json with per-page data (title, description, OG tags, course-specific data)
7. Test that python3 tools/html_template_engine.py --templates templates/ --out . --data site.json rebuilds identical HTML

2. centralize-analytics

Extract all analytics/tracking code (GA4, Meta Pixel, Clarity, CAPI) from individual pages into shared files.

Steps:

1. Identify all tracking snippets across pages (use site-audit findings)
2. Create analytics-head.html (gtag, fbq, clarity — goes in <head>)
3. Create analytics.js (event handlers, form tracking, CAPI calls)
4. Update all pages to include the shared files
5. Verify tracking IDs are consistent

3. centralize-forms

Extract form handling (form webhook, server-side event submission) into shared form-handler.js.

Steps:

1. Identify the form submission pattern duplicated across pages
2. Create form-handler.js with configurable data attributes (data-course, data-source)
3. Update all pages to use the shared script
4. Test form submission still works

4. add-seo-basics

Generate missing SEO files and tags.

Steps:

1. Generate sitemap.xml listing all HTML pages with their URLs
2. Generate robots.txt with Sitemap: directive
3. Add <link rel="canonical"> to all pages
4. Add JSON-LD Course schema to course pages
5. Add JSON-LD Organization schema to index page

5. add-cache-busting

Replace manual version query strings with content-hash based cache busting.

Steps:

1. Compute MD5 hash of each CSS/JS file (first 8 chars)
2. Update all <link> and <script> references to use ?h=<hash> instead of ?v=N
3. Document the process so future changes auto-update

6. optimize-images

Add missing image attributes and flag oversized images.

Steps:

1. Scan all <img> tags for missing width/height attributes
2. Add loading="lazy" to below-the-fold images
3. Flag images over 200KB for manual WebP conversion
4. Report total image weight savings potential

7. accessibility-fixes

Fix mechanical accessibility issues.

Steps:

1. Add placeholder alt="" to decorative images, alt="[REVIEW]" to content images
2. Add <label> elements for all form inputs
3. Fix heading hierarchy gaps
4. Add <main> landmark if missing

How to use

Step 0 — Parse arguments

First word of $ARGUMENTS is the pattern name. Rest is the site root path. Default path: current working directory (else ask for the site root).

If the pattern name is not recognized, list available patterns and STOP.

Step 1 — Pre-flight analysis

Before any changes:

1. Read the files that will be affected (Glob + Read)
2. Count: how many files, estimated lines changed
3. APPROVAL GATE: Present the user with:
   • Pattern name and what it does
   • Files that will be modified
   • Preview of the first file's changes (first 20 lines of diff)
   • Estimated scope
4. Wait for explicit "go ahead" before proceeding

Step 2 — Apply the pattern

Follow the steps listed for the chosen pattern. Make changes with Edit/Write.

Step 3 — Verify

Run python3 tools/site_audit.py --root "<site_root>" --mechanical-only --checks <relevant> to confirm the finding is resolved.

Report: files modified, lines changed, findings before/after.

Important constraints

1. Approval gate is mandatory. Never modify files without showing the plan first.
2. Preserve existing functionality. Every refactoring must produce functionally identical output.
3. One pattern per invocation. User can chain by re-invoking the skill.
4. Don't force a framework. extract-template uses Jinja2 as a build tool — output is plain HTML.

See also

• tools/site_audit.py — diagnostic tool
• tools/html_template_engine.py — Jinja2 build tool
• workflows/site_refactor.md — the SOP