Content Researcher

Research a topic by reading the best-ranking content on the web, distilling it into a structured knowledge base, and surfacing gaps that existing content misses. The output is two files — a knowledge base and a list of under-discussed questions with answers — that together form the foundation for writing superior content.

Inputs

• Topic — the subject to research (e.g. "email deliverability", "React performance optimization")
• Content type (optional) — what the user plans to write: ultimate guide, how-to, comparison, listicle, tutorial, etc. Defaults to "comprehensive guide" if not specified. This shapes the search queries and what counts as a good source.

Output files

Both files live in the project's current working directory by default. If the user specifies a different location, use that instead.

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

Step 1. Expand the topic into search queries

Take the user's topic and content type, then generate five search queries. Each query targets a different angle or entity related to the topic — this ensures you don't just find the same article five times. Tailor the query suffixes to the content type.

Example for topic "technical SEO" (ultimate guide):

• technical SEO ultimate guide
• site architecture SEO complete guide
• core web vitals optimization guide
• crawl budget management comprehensive guide
• structured data markup complete tutorial

Example for topic "React vs Vue" (comparison):

• React vs Vue 2025 comparison
• React vs Vue performance benchmark
• React vs Vue developer experience comparison
• migrating from Vue to React pros cons
• React vs Vue for enterprise applications

Example for topic "setting up ESLint" (how-to):

• how to set up ESLint from scratch
• ESLint configuration best practices
• ESLint flat config migration guide
• ESLint with TypeScript complete setup
• ESLint rules recommended setup tutorial

Step 2. Search and filter

Web-search each query. From the combined results, select the top 5-8 pages that are genuinely substantive (structured with headings, covering the topic in depth). Skip thin pages — they won't contribute meaningful knowledge. Match source quality to content type: for comparisons, prioritize hands-on benchmarks and experience reports; for how-tos, prioritize step-by-step tutorials with code/screenshots.

Step 3. Read and extract (subagent per source)

For each selected source, spawn a subagent. Read ./references/guide-reader.md for the subagent prompt. Each subagent reads the source and returns extracted knowledge entries plus discovered questions.

After each source is processed, merge its output into the two shared files.

Step 4. Answer remaining questions (subagent per question)

Once all sources have been processed, review under-discussed.md. For each unanswered question, spawn a subagent. Read ./references/question-researcher.md for the subagent prompt.

Append each answer directly below its question in under-discussed.md:

## {question}

{answer}

Sources: {urls}