Deep Research Skill

Conduct multi-source web research on a given topic and produce a structured, citation-backed report.

Workflow

1. Clarify Scope

If the topic is vague, ask the user to narrow down:

• What specific aspect? (기술 비교, 트렌드, 경쟁사, 도입 사례, etc.)
• Target audience or context? (스타트업, 대기업, 개인 프로젝트, etc.)
• Depth? (quick overview vs deep dive)

If the topic is clear enough, proceed directly.

2. Research Plan

Before searching, outline 3-5 research angles to cover. For example, a tech comparison might cover:

1. Core features & philosophy
2. Performance benchmarks
3. Ecosystem & community
4. Learning curve & DX
5. Production adoption & case studies

3. Multi-Source Search

Execute 5-10 WebSearch queries from different angles:

• Direct topic searches
• "vs" comparisons
• "{topic} pros cons {current year}"
• "{topic} production experience"
• Reddit/HN discussions for real-world opinions
• Korean sources via Naver/Korean keywords when relevant

For each promising result, use WebFetch to extract key details.

4. Synthesize

Compile findings into a structured report:

# Research Report: {Topic}
> Researched: {date}

## TL;DR
{3-5 bullet executive summary}

## {Section 1 — varies by topic}
{Analysis with specific data points}

## {Section 2}
...

## {Section N}
...

## Comparison Table (if applicable)
| Criteria | Option A | Option B | ... |
|---|---|---|---|

## Recommendation
{Clear recommendation with reasoning}
{Conditions or caveats}

## Sources
- [Title](URL) — {1-line summary of what was extracted}
- ...

5. Deliver

• Present the report directly in chat
• If the user wants it saved, write to a file (suggest research/{topic-slug}.md)

Research Quality Rules

• Recency: Prefer sources from the last 12 months. Flag outdated information.
• Diversity: Mix official docs, blog posts, community discussions, benchmarks. Don't rely on a single source.
• Specificity: Include concrete numbers (stars, downloads, benchmark results, adoption stats) over vague claims.
• Honesty: If information is conflicting or uncertain, say so. Don't present opinions as facts.
• Attribution: Every claim should trace back to a source in the Sources section.

Language

• Write the report in the user's language (Korean if they asked in Korean)
• Keep technical terms in English with Korean explanation where helpful
• Source titles can remain in their original language