silo-audit

Silo Audit — Does the Website Have a Silo Architecture?

This is an agent analysis layer ON TOP OF Screaming Frog: SF knows depth and inlinks, but it does not judge whether URLs form topical silos with hubs or whether semantic coverage is complete. This skill builds a URL tree, calculates structural metrics, and returns a verdict of chaos / basic silo / extended silo. The theory—3 levels, extended silos, the filter matrix, cross-links, E-E-A-T, and coverage of 5–15% / 20–30% / 70–90%—is in reference/silo-architecture.md; load it before evaluating a website.

When to Use It

• “Do we have a silo or not?” and “Evaluate the website structure/architecture.”
• “Are there topical clusters?” and “Is semantic coverage complete?”
• “Is the website flat or just a brochure site?” and “Have we reached topical authority?”
• “Site structure audit” and “silo structure.”

Workflow

1. Obtain crawl data. You need audit.json from sf-analyzer; see ../sf-analyzer/SKILL.md. If it is unavailable, run a crawl first. Otherwise, work from the sitemap: curl -s https://example.com/sitemap.xml | grep -oP '(?<=<loc>)[^<]+'. Build the path tree from indexable HTML in pages[]; exclude _next/, .js, and images.
2. Structural metrics calculated in Python from pages[].metrics:
   • URL path depth: len([s for s in urlparse(u).path.split('/') if s]). Calculate both the maximum and median. A flat structure has a median of ≤1 and almost everything at L0/L1. A silo typically has L1 section hubs → L2 internal pages, with L3 in an extended silo.
   • Hub pages: for every L1 prefix such as /services/ or /industries/, check whether its own /<prefix>/ index page links to child pages. Count children per prefix; a cluster is viable when it has ≥3–5 pages beneath a shared hub.
   • SF crawl_depth (metrics.crawl_depth): calculate the maximum and median. A depth beyond 4–5 clicks indicates a diluted silo or poor internal linking.
   • Orphan rate: divide the number of pages with metrics.inlinks == 0 or unique_inlinks == 0 by the total number of indexable pages. A rate above 10–15% indicates a structure with significant gaps.
   • Breadcrumbs: look for BreadcrumbList in SF structured data or <nav> breadcrumbs in saved HTML. Their presence confirms the silo hierarchy.
   • Cross-cluster links: use issues[].locations/inlinks to calculate the share of links whose source_url and target_url belong to different L1 clusters. A healthy silo keeps authority within each cluster but has some controlled cross-links; see the reference. It should have neither zero cross-links nor chaotic “everything to everything” linking.
   • Sitemap vs. crawl: use summary.sitemap to compare urls_in_sitemap with in_crawl_not_in_sitemap. Coverage gaps indicate incomplete silo branches.
3. Semantic coverage. Use the set of L1 prefixes and headings (metrics.title, h1) to identify which intents are covered and which have obvious gaps: catalog/services · industries/applications · dictionary/glossary · cases/portfolio · materials/blog hub · E-E-A-T pages (author/expert, about, methodology). A rough coverage percentage is covered intents divided by expected intents for the niche. Compare it with the reference thresholds: 5–15% — chaos, 20–30% — basic silo, and 70–90% — extended silo.
4. Verdict. Consolidate the metrics into one of three classes:
   • chaos/brochure site — flat structure with median depth ≤1, no hubs, a high orphan rate, and 5–15% coverage;
   • basic silo — L1 hubs, clusters of 3–5+ pages, breadcrumbs, and 20–30% coverage;
   • extended silo — L3 depth, a filter matrix, glossary + cases + E-E-A-T, controlled cross-links, and 70–90% coverage. Then provide a gap list: which hubs, clusters, and intents are missing; where orphan nodes occur; where cross-cluster links cause the silo to “leak”; and what to add—specific sections and hub pages—to reach the next level.