SEO Interlinking Audit -- Existing-Site Link Health

Diagnoses the internal-link graph of a site that already exists and already has content -- orphan pages, pages too deep to get crawled/prioritized, anchor-text over-optimization, and which high-authority pages could donate link equity to the weak ones found. Read-only: it never edits a post, never calls a WordPress write ability, and never inserts a link on its own.

This is the complementary skill to site-architecture and seo-cluster (which plan new structure or new content clusters) and to seo-programmatic (which covers templated pages generated at scale). This skill audits what is already live.

Quick Reference

Required inputs

Two Screaming Frog exports are needed -- not the same report. See references/screaming-frog-export.md for exact menu paths if this needs explaining to a client or a new session:

1. All Inlinks (bulk export, From/To/Anchor Text columns) -- this is the actual link graph. The "Internal" tab's own Inlinks/Outlinks columns are just counts, they cannot rebuild who-links-to-whom or the anchor text used.
2. Internal HTML (or "Internal All" filtered to HTML) -- Address, Indexability, Status Code -- used to restrict the audit to indexable, live, HTML pages.

GSC data is optional but strongly recommended -- without it, orphans/deep pages are ranked only by crawl signals, not by whether Google already sends them traffic. Export a standard GSC Search Analytics CSV (Page, Clicks, Impressions, Position) for the same property, or ask the user which GSC access method applies to this project (service account vs. local OAuth token -- see this environment's own GSC access notes) if a live pull is wanted instead of a CSV.

If the user hasn't provided these files yet, ask for them before running the script -- do not guess file paths.

Workflow

1. Collect the two Screaming Frog CSVs, the homepage URL, and (if available) the GSC CSV or a live GSC pull for the same property.
2. Run the script:

   python scripts/audit_interlinking.py --inlinks <all-inlinks.csv> --overview <overview.csv> --home <https://example.com/> --gsc <gsc.csv>

3. Read the JSON the script prints (summary, orphans, deep_pages, anchor_text_flags, hub_candidates).
4. Before writing a single recommendation, check this environment's project memory for that site's own interlinking house rules (anchor style, how many inline links, whether it uses a closing "Recomendamos"/"También te puede interesar" block, etc.). Different projects in this environment have different, specific rules -- never invent a generic format. If no project-specific rule is found, say so explicitly rather than assuming one.
5. Render a single prioritized Markdown report, ordered:
   • Orphans with real GSC clicks/impressions first (highest impact: Google already wants to send traffic there, it just can't find the page well internally).
   • Then orphans with no GSC signal yet.
   • Then deep pages (same GSC-first ordering).
   • Then anchor-text over-optimization flags. Each recommendation names a specific hub candidate to link from (pick the closest topical match among hub_candidates, not just the single highest-traffic page for everything) and a suggested anchor text consistent with the project's house rules from step 4.
6. Stop there. Do not open the CMS, do not edit a post, do not call any WordPress ability to add the link -- hand the report back for the user (or a separate, explicitly-confirmed step) to execute.

Thresholds (script defaults, all overridable via flags)

Explicitly out of scope

• SE Ranking -- retired from this stack entirely, never a data source here.
• New site or new cluster planning -- see site-architecture, seo-cluster, blog-cluster.
• Templated/programmatic pages at scale -- see seo-programmatic.
• Writing anything -- no post edits, no link insertion, no WordPress writes of any kind. Diagnosis and prioritized recommendations only.

Reference files

• references/screaming-frog-export.md -- load on demand: exact Screaming Frog report names, menu paths, and column names this skill expects, for when a session needs to explain the export steps or troubleshoot a CSV that doesn't parse.