| name | research |
| description | Deep research for an article section. Scales research depth based on article.json targets. Gathers sources, finds real-world examples, collects supporting evidence, and identifies case studies. |
| allowed-tools | Read, Write, WebSearch, WebFetch, Glob |
Section Research Agent
Research the section: $ARGUMENTS
Setup
-
Load Article Context
- Read
article.json to find section matching $ARGUMENTS
- Note the section's
scaffold, required_elements, and research_questions
- Understand the
narrative_role (hook, foundation, deep-dive, etc.)
- Check
article_type to know what kinds of evidence to prioritize
- Check
content_type to understand the format
- Read
research_config for this article's research depth settings:
sources_per_section — how many sources to gather
search_queries_min — minimum number of search queries
include_video_research — whether to search for videos
-
Review What's Needed
- Word target for this section (from section's
word_target)
- Required examples and links (from section's
required_elements)
- Specific research questions to answer
- Research depth from
research_config
Research Standards
All targets come from article.json — do NOT use hardcoded values:
- Sources per section: from
research_config.sources_per_section
- Examples per section: from section's
required_elements.examples_minimum
- Search queries: at least
research_config.search_queries_min
- Supporting evidence appropriate to the article type
- Expert quotes where available
- Contrasting viewpoints if they exist
Research Process
Phase 1: Broad Search
Run at least research_config.search_queries_min search queries. Scale up for deeper articles.
Query patterns:
- "[topic] 2025 2026" - Recent content
- "[topic] guide" - Educational content
- "[topic] examples" - Real implementations
- "[topic] best practices" - Expert recommendations
- "[topic] case study" - Real-world applications
- "[topic] vs [alternative]" - Comparisons
- "[topic] research data statistics" - Hard evidence
- "[topic] expert opinion" - Authority voices
For shorter formats (email, news), focus on the most relevant 2-3 query patterns. For long-form and research articles, use all patterns and add topic-specific queries.
Phase 2: Find Real-World Examples
Search specifically for examples matching the section's required_elements.examples_minimum:
- Named companies that succeeded or failed with this approach
- Named people (founders, practitioners, experts) with relevant stories
- Research studies with concrete findings and data
- Case studies with measurable outcomes
- Historical examples or precedents
- Conference talks or interviews with practitioners
For each example found, note:
- Name (person, company, study) and URL
- What makes it notable
- Specific details, numbers, or outcomes
- Why it matters for the article's argument
Phase 3: Deep Dive (WebFetch)
For the most promising sources (scale to article depth):
- Long-form/Research: Fetch 5+ top sources
- Tutorial/Case-study: Fetch 3-4 top sources
- News/Opinion: Fetch 2-3 top sources
- Email: Fetch 1-2 top sources
For each:
- Fetch full content
- Extract key insights, not just summaries
- Pull exact quotes with attribution
- Note any data/statistics
- Gather evidence appropriate to the article type
Phase 4: Topic-Specific Evidence
Based on article_type in article.json, prioritize gathering:
| Article Type | Evidence to Prioritize |
|---|
| Technical | Code samples, documentation, benchmarks, repo examples |
| Business | Revenue figures, growth metrics, strategy breakdowns |
| Marketing | Campaign results, conversion data, channel comparisons |
| Opinion | Research papers, expert positions, counterarguments |
| Tutorial | Step-by-step processes, tool comparisons, common pitfalls |
| Cultural | Quotes, surveys, trend data, cultural artifacts |
Phase 5: Video/Media Content
Only run this phase if research_config.include_video_research is true.
Search for relevant videos:
- Conference talks (often have unique insights)
- Interviews with practitioners
- Documentary-style explorations of the topic
Note: Title, URL, timestamp of key moments, why it's valuable
Output Files
1. Research Notes: research/[section-id].md
Structure your notes for easy writing:
# Research: [Section Title]
## Executive Summary
3-4 sentences capturing the most important findings for this section.
## Key Insights (for narrative)
1. **Insight with hook potential**: [explanation with source]
2. **Surprising finding**: [explanation with source]
3. **Common misconception**: [what people get wrong]
## Real-World Examples
### [Example Name 1]
- **URL**: [link]
- **What happened**: [description]
- **Key detail**: [specific numbers, outcomes, or quotes]
- **Why it's notable**: [what makes this a good example]
### [Example Name 2]
... continue for required number of examples ...
## Supporting Evidence
### [Evidence piece 1]
- **Source**: [url]
- **Type**: [data/quote/case study/code sample]
- **Content**: [the actual evidence]
- **Context**: [how to use this in the article]
## Expert Quotes
> "Quote that captures key insight"
> — [Name], [Role] at [Company] ([Source](url))
## Statistics & Data
- [Stat 1 with source]
- [Stat 2 with source]
## Video Resources
- **[Video Title](url)** - [Why it's valuable, key timestamp]
## Contrasting Views (if any)
- View A: [perspective with source]
- View B: [alternative perspective with source]
## Questions Answered
### Q: [Research question 1]
**A:** [Detailed answer with citations]
### Q: [Research question 2]
**A:** [Detailed answer with citations]
## Raw Sources
| Title | URL | Type | Key Value |
|-------|-----|------|-----------|
| ... | ... | doc/blog/video | ... |
2. Update sources.json
Append comprehensive source data:
{
"section": "section-id",
"researched_at": "timestamp",
"sources": [
{
"title": "Source Title",
"url": "https://...",
"type": "documentation|blog|video|research|case-study|interview",
"authority": "high|medium",
"key_quotes": ["quote 1", "quote 2"],
"accessed": "YYYY-MM-DD"
}
],
"examples": [
{
"name": "Example Name",
"url": "https://...",
"description": "What it is",
"relevance": "Why include in article"
}
]
}
3. Update article.json
- Set
research_complete: true
- Add discovered sources to main sources array
4. Append to progress.txt
[timestamp] Research: [section-id]
- Sources: N (target: [sources_per_section from research_config])
- Examples found: N (target: [examples_minimum from section])
- Key insight: [one sentence]
- Research quality: [PASS/NEEDS MORE]
Quality Check
Before marking complete, verify against article.json targets:
If any check fails, do more research before completing.
Completion
Summarize:
- Sources gathered (vs target from article.json)
- Examples found (vs target from article.json)
- Evidence collected
- Research quality assessment
- Next step:
/research [next-section] or /outline when all done