Should you build your own Frase?

Build me a content-brief generator to replace Frase. Requirements:

- A Node CLI: `brief "<keyword>"` writes ./briefs/<keyword-slug>.md.
- Fetch the top 10 results for the keyword from Serper.dev (key in .env).
- Scrape each result with Playwright plus Readability: title, H2/H3 outline,
  word count, and any questions found on the page.
- Send the aggregate to an LLM (key in .env) to produce the brief: suggested
  title and angle, a target outline, questions to answer, terms competitors
  cover, and a word-count range.
- `score <draft.md> "<keyword>"` re-checks my draft against the brief: which
  terms and questions are covered or missing, with a plain coverage percentage.
- Cache scraped pages in ./cache/ keyed by URL so re-runs cost nothing.
- Flat files only, no database; no accounts, no telemetry, the only network
  calls are the SERP API, page fetches, and the LLM.
- Out of scope: an editor UI, rank tracking, team workflows. Briefs are
  Markdown, I write in my own editor.
- README: which keys to get and the rough per-brief cost of the SERP API.