Vale construir sua versão de Surfer SEO?

Build me a content-brief and draft-scoring tool to replace Surfer SEO, personal scale. Requirements:

- A Node CLI with two commands: brief "<keyword>" and score <draft.md> "<keyword>".
- brief: fetch the top 10 results from Serper.dev (key in .env), download each page,
  extract title, headings, and body text with cheerio + @mozilla/readability.
- Compute across the winners: median word count, recurring H2/H3 topics, and the 30
  most frequent meaningful terms via simple tf-idf. No NLP service.
- Send that to an LLM (key in .env) to write briefs/<keyword>-YYYY-MM-DD.md: suggested
  outline, target word count, terms to cover, questions to answer.
- score: compare my draft against the brief's term list and outline, print a coverage
  percentage and a table of missing terms in the terminal.
- Cache every SERP response and fetched page in SQLite (better-sqlite3) so re-runs
  cost zero API credits.
- No accounts, no telemetry, everything local except the Serper and LLM calls.
- Out of scope: a writing editor with live scoring, site audits, and rank tracking.
  This outputs Markdown briefs, and page scraping will break on some sites · Serper
  is the one paid dependency, say both in the README.
- README: where to get the Serper key and the rough API cost per brief.