content-researcher

Inputs

Accept one of:

• Topic — the subject to research (e.g. "email deliverability"). Optionally include content type (ultimate guide, how-to, comparison, listicle, tutorial — defaults to "comprehensive guide").
• SERP — a keyword and a list of pages (title + snippet) from a prior SERP analysis.

Outputs

• knowledge-base.md — structured knowledge (~30 entries, each with a depth label)
• under-discussed.md — questions not well covered by existing content, with researched answers

Both go in the working directory by default, or wherever the caller specifies.

Steps

1. Find sources (subagent) — If SERP provided, use those pages and go to step 2. Otherwise, follow references/search-and-filter.md to get 5-8 source pages.
2. Read and extract (subagent per source) — For each page, spawn a subagent with the topic, a source ID ([1], [2], ...), and the current knowledge base. Follow references/guide-reader.md. Merge each subagent's output into the two shared files.
3. Answer remaining questions (subagent per question) — Review under-discussed.md for unanswered questions. For each, spawn a subagent following references/question-researcher.md.

For steps labelled with (subagent), run them as subagents.