internal-link-audit

Internal-linking tool for an Astro site built with the new-website kit. Builds the site's inbound-link graph from dist/ and reports which pages are ORPHANED (in the sitemap but no internal link points

SKILL.md

GitHub

Internal Link Audit

Pages rot in the OTHER direction too: not the links that go dead, but the pages that nothing links TO. A page can be in the sitemap, pass every QA spec, and still be orphaned — reachable by no internal link, so crawlers find it late (or never) and no human browsing the site ever lands on it. The classic cause: a footer/nav link group that lists only ONE of a sibling set (e.g. one acme-vs-<competitor> comparison), leaving the rest linked from nowhere.

This is the internal-link twin of outgoing-link-audit:

offline CI gate active sweep (this)
outgoing (dead/rebranded) tests/links.spec.ts outgoing-link-audit
internal (orphan/thin) tests/orphans.spec.ts internal-link-audit

Read-only with respect to the site: it diagnoses and suggests. You apply fixes only after the user confirms.

The strategy it enforces

This skill does not invent its own linking rules — it measures the built site against the strategy in site-architecture (its "Internal Linking Strategy" section, plus its "3-Click Rule"; read them before suggesting fixes). The strategies, and how this audit checks each:

Strategy (from site-architecture) How the audit enforces it
No orphan pages — every page has ≥1 inbound internal link ORPHANS list (0 inbound) — the hard gate tests/orphans.spec.ts also fails on this
Hub-and-spoke / topic clusters — hub links all spokes, spokes link the hub and each other sibling sets that don't cross-link surface as orphans/thin; suggest the hub + cross-links
Important pages get the most inbound links — home, pricing, key features the inbound COUNT ranking shows whether your money pages actually lead
Descriptive, keyword-bearing anchor text — not "click here" judgment: read the anchor text of the links you add/keep (see website-seo-geo)
3-Click Rule — important pages within ~3 clicks of home DEEP list (>3 clicks from home, computed by BFS)
Contextual links beat boilerplate — in-body editorial links > nav/footer THIN list flags pages held up only by the global nav/footer

(Linking to the clean canonical URL — no .html, no redirect/trailing-slash hop — is link hygiene the navigation spec already guards; apply it to every link you add, step 4.)

When to run

Before a release / launch, after adding any batch of pages (the moment a set of siblings gets created but only some get linked), or when Google Search Console flags "Crawled - currently not indexed" / "Discovered - not indexed" / orphaned pages.

Unlike outgoing-link-audit, this makes no network calls — it reads the local build — so it's cheap to run any time.

Shortcut: the shared script does the graph

scripts/check_internal_links.sh builds (if needed), extracts every internal anchor from dist/, and prints the inbound-link graph grouped as ORPHANS / THIN / OK. Run it first:

bash scripts/check_internal_links.sh

Then apply judgment to its output (the part a script can't do — which related page should link to each orphan, and with what anchor text) and hand off the fixes.

Procedure

1. Build the inbound-link graph

Audit the built dist/ so it matches what Cloudflare serves. For every page, the script counts the distinct other pages that link to it (self-links and the home root excluded; query/fragment/www/trailing-slash variants collapse to one target, matching the test) and its click depth from home (BFS). Classify each page:

  • ORPHAN — 0 inbound internal links. The high-value finding: a page no one can reach by browsing. This is exactly what tests/orphans.spec.ts fails on.
  • THIN — exactly 1 inbound link, and that link is the global nav/footer (no contextual in-body link). Not broken, but weakly connected: little internal PageRank flows to it (the "contextual links beat boilerplate" strategy).
  • DEEP — more than ~3 clicks from home. Crawlers spend less budget on deep pages and pass them less equity (the "shallow crawl depth" strategy). Orthogonal to inbound count — a page can be both THIN and DEEP.
  • OK — ≥2 inbound links and within 3 clicks.

Then sanity-check the inbound RANKING against intent: your most important pages (home, pricing, primary feature/conversion pages) should be among the most-linked. A key conversion page sitting at 1 inbound link is a strategy miss even though it isn't technically orphaned.

2. Suggest WHERE to add the links (the judgment step)

For each orphan/thin page, the fix is not "link it from anywhere" — it's to add a link where it's topically earned, so the link helps a real reader:

  • Sibling sets cross-link. If /acme-vs-crm, /acme-vs-chat, /acme-vs-calendar, … exist, every comparison page should link to its siblings (a "Compare with other tools" block), and a hub/overview page should list them all. A footer "Product" group that names only acme-vs-crm is the bug — add the siblings to that group in src/config.ts, or build the hub page.
  • Contextual in-body links beat nav links. A sentence on a related page that links the orphan in context ("see how we compare to a CRM") is worth more than a footer entry, to both readers and crawlers.
  • Pick anchor text that carries the target's keyword, not "click here"/"read more" — it's a ranking signal for the target page (see website-seo-geo). Vary it naturally across links; don't paste the exact-match phrase every time.
  • Pull DEEP pages shallower. Link a >3-click page from a hub or section page that itself sits near home, so it lands within ~3 clicks.
  • Feed the important pages. If a money/conversion page is under-linked relative to its importance, add links to it from high-traffic related pages — important pages should have the most inbound links, not the fewest.
  • Use the free links. Breadcrumbs and a "related pages" block are inbound links on every page at no editorial cost — a structural fix for whole clusters at once.
  • Always link the canonical URL — clean path, no .html, no trailing slash, no redirect hop (so equity isn't lost to a 301 and navigation.spec stays green).

Judgment, not just counts: a deliberately-unlinked page (a paid-campaign landing page reached only from an ad, a print-QR page) is correctly orphaned — it should go in ORPHAN_EXEMPT in tests/orphans.spec.ts with a reason, not get a forced link. Use the model to tell a real orphan from an intentional one.

3. Report

Group by severity. Lead with ORPHANS (the actionable ones), then THIN, then DEEP, then OK as a count. For each orphan/thin/deep page, name the specific page(s) that should link to it and the suggested anchor text.

4. Hand off the fixes

For each finding, after the user confirms:

  1. Add the link in the source page (src/pages/*.astro, a content-collection entry, or a nav/footer group in src/config.ts). Use keyword-bearing anchor text and the clean URL (no .html, no trailing slash).
  2. For a deliberately-unlinked page, add it to ORPHAN_EXEMPT in tests/orphans.spec.ts with the reason instead.
  3. Re-run npm run build && npm test to confirm green (orphans.spec.ts must pass), then commit.

Scope notes

  • Audits the active branch you're on; dist/ is the build of src/, which is what Cloudflare serves.
  • Internal links only. Outgoing (external) link liveness is outgoing-link-audit; dead in-site links and #fragment integrity are the navigation + anchors specs in website-qa. This skill answers the orthogonal question: is every page a link target, linked the way the strategy says?
  • Strategy lives in site-architecture (its Internal Linking Strategy section) — this skill is its enforcement arm against the built site, not a second rulebook. When the two could drift, site-architecture wins; flag the gap rather than re-deciding here.
  • The hard gate (tests/orphans.spec.ts) only knows pass/fail on reachability-from- home; this skill is the richer report (inbound COUNT per page, THIN pages, link suggestions) you act on to keep it green.
  • Page-relative links: scripts/check_internal_links.sh resolves only root-relative (/about) and absolute-on-host links — the kit's convention. It does not resolve page-relative hrefs (about, ./team, ../pricing), so a page reachable only via one is over-reported as ORPHAN/THIN by the script. The gate tests/orphans.spec.ts does resolve them and is authoritative — trust the gate over the script for those, and prefer root-relative links so the two never diverge.