Keyword Cluster Builder

Real searchers don't all type the same keyword for the same need. They type variations, slang, misspellings, half-questions. A page that targets only the head term misses every other phrasing — sometimes dozens of them, sometimes the highest-intent ones. This skill encodes the five-layer mental scaffold that finds the full set: direct variations, specific contexts, method/capability queries, decision queries, and related questions.

Step zero is intent. A polished cluster built for the wrong meaning of the root is the most damaging output the skill can produce because it looks professional — the user may not notice it solved the wrong problem. If the root is ambiguous ("mercury", "python", "apple") and business context doesn't resolve it, the skill stops and asks. It also tags every candidate keyword by intent type — informational, commercial investigation, transactional, navigational — because terms with different intent cannot rank on the same page, no matter how topically related they look. When a request spans multiple intents, you get separate clusters, one per intended page, plus a priority order — never one intent-mixed table that ranks for nothing.

The output is a five-column table (keyword, layer, estimated volume, relevance), plus recommended H2 topics and FAQ questions drawn directly from the cluster. If you supply a SEMrush or Ahrefs CSV, real volumes go in; rows the skill expanded semantically are marked — (semantic add) so the no-fabrication line stays visible. The skill never invents a number to fill a blank cell.

A few deliberate constraints: it won't pad to hit a number. If a narrow root genuinely yields only 8 strong terms, you get 8 and a reason — an honest small cluster beats 30 terms half of which dilute the page. It won't force a transactional term into an informational cluster. And it won't write the post for you — that's the SEO Blog Writer's job. This produces the brief.