Link Weaver: see where your internal links really go, then design the missing ones

Ask it in your own words

• "How should I link my articles together?"
• "Which pages should link to my new post?"
• "Design a related posts block for my blog."
• "my translated pages are only reachable from the language switcher, is that a problem"

What happens next

1. It builds the graph. link_graph.py, the crawler, reads the site and its sitemap and records every link with its type, anchor and position.
2. It doubts its own sorting. Sorting a link by its place in the HTML is a heuristic. Claude compares a few pages with what you see before trusting any number.
3. It designs the links. Pillars and their clusters go into one relation table that you confirm. Link blocks and link briefs come from it.

What you get back

• A five-line audit summary, every figure labelled measured, reported, estimated or assumed
• The findings: orphans, pages held up only by menus and footers, overloaded hubs, broken links
• A relation table and a page map for your top pillars
• Link briefs: from URL, to URL, anchor, place in the page, reason
• A rollout plan, then a before and after table from a new crawl

Run on this very site

The link graph of the site as it was with five pages.

5 HTML pages parsed; 26 internal link occurrences; sitemap URLs: 5.
footer | 12 | 46%
header | 8 | 31%
editorial | 6 | 23%
Pages with <= 1 editorial inlink(s) (excluding home) | 3

Three quarters of the links sat in the header and the footer. Each skill now has its own page, linked from the list on the home page and from the skills it hands over to.

What it will not do

• It sees the raw HTML of the pages it fetched: a link added by JavaScript does not exist for it.
• It needs a sitemap to find orphans, because a crawl cannot reach a page nobody links to.
• Its internal rank models your own links, not Google's PageRank. No link change comes with a promise of ranking or indexing.

What it needs

Claude Code, Claude.ai, Claude Desktop or Cowork. The crawler needs code execution with Python 3.9+ and web access. Otherwise a manual route uses page fetches and Search Console's Links report, on a sample.

How to install a skill in Claude

Guides

• Discovered - currently not indexed: what it means and what to do, in order

Where it hands over

• Index Doctor: If the real problem is indexing as a whole (robots, noindex, canonicals, duplicates), start there.
• Polyglot: On a multilingual site, it reads this crawl's files to count the pages no same-language content links to.

Questions about this skill