Yt Research

Examples:

• yt-research "양자 컴퓨팅의 현재와 미래"
• yt-research "AI 반도체 전쟁" --deep
• yt-research "일본 경제 버블" --channel tech-shorts
• 리서치 "메이플스토리 20주년 역사"
• 자료 수집 "한국 출산율 위기" --deep

Commands

• yt-research "<topic>" — standard depth research
• yt-research "<topic>" --deep — deep research
• yt-research "<topic>" --quick — quick overview
• yt-research "<topic>" --channel <id> — channel-specific research
• yt-research list — show recent research in research/ directory
• yt-research show <topic_slug> — show research detail from saved JSON

Phase 1: Web Search

Use WebSearch to find relevant sources. Run searches in both Korean and English for broader coverage.

Search Strategy

1. Primary search: Topic as-is in original language
2. Translated search: Topic translated to the other language (KO↔EN)
3. Targeted searches: Append qualifiers based on depth:
   • quick: "{topic}", "{topic} 정리", "{topic} summary"
   • standard: Add "{topic} 통계", "{topic} statistics", "{topic} 논란", "{topic} timeline"
   • deep: Add "{topic} 논문", "{topic} research paper", "{topic} official report", "{topic} 반론"

Source Prioritization

Rank sources by type (highest to lowest reliability):

1. official — Government reports, official documentation, press releases
2. academic — Papers, journal articles, university publications
3. news — Major news outlets (Reuters, AP, 연합뉴스, etc.)
4. wiki — Wikipedia, 나무위키 (cross-reference required)
5. video — YouTube transcripts, documentary clips
6. blog — Expert blogs, Medium, Brunch
7. community — Reddit, forum posts (low reliability, use for sentiment only)

Source Count by Depth

• quick: 3-5 sources
• standard: 5-10 sources
• deep: 10-20 sources

Phase 2: Source Processing

Use WebFetch to get full content for each discovered source.

For each source, extract:

• Key claims/facts: Factual statements that can be verified
• Statistics/numbers: Any quantitative data with context
• Quotes: Direct quotes from named individuals
• Dates: Events, deadlines, milestones
• Proper nouns: People, organizations, places — spelling must be verified

Extraction Rules

• Always note the exact location (paragraph, section) where data was found
• Preserve original language for quotes; add translation if needed
• Flag any claim that appears in only one source
• Note when a source cites another source (trace to origin when possible)

Phase 3: Structuring

Organize all collected data into the following JSON structure:

{
  "topic": "주제명",
  "topic_en": "Topic in English",
  "depth": "standard",
  "created_at": "2026-03-16T14:30:00+09:00",
  "summary": "1-2 paragraph overview of the topic covering the most important findings and context.",
  "claims": [
    {
      "id": "claim_001",
      "statement": "주장 내용",
      "statement_en": "Claim in English",
      "category": "fact|opinion|prediction|historical",
      "evidence": [
        {
          "source_id": "src_001",
          "excerpt": "근거 발췌",
          "excerpt_en": "Evidence excerpt in English",
          "reliability": "high|medium|low"
        }
      ],
      "contradictions": [
        {
          "source_id": "src_003",
          "excerpt": "반론 내용",
          "detail": "How this contradicts the claim"
        }
      ],
      "fact_check_flags": ["claim_flag_id_if_any"]
    }
  ],
  "sources": [
    {
      "id": "src_001",
      "title": "출처 제목",
      "url": "https://...",
      "type": "news|academic|official|blog|wiki|video|community",
      "date": "2026-01-15",
      "language": "ko|en",
      "reliability": "high|medium|low",
      "author": "Author name if available",
      "publisher": "Publisher/outlet name"
    }
  ],
  "timeline": [
    {
      "date": "2024-01",
      "event": "사건 설명",
      "event_en": "Event description",
      "source_id": "src_001"
    }
  ],
  "key_numbers": [
    {
      "value": "42%",
      "context": "맥락 설명",
      "context_en": "Context in English",
      "source_id": "src_001",
      "date": "2025-12"
    }
  ],
  "key_people": [
    {
      "name": "이름",
      "name_en": "Name in English",
      "role": "직함/역할",
      "relevance": "Why this person matters to the topic"
    }
  ],
  "fact_check_flags": [
    {
      "id": "flag_001",
      "claim_id": "claim_001",
      "issue": "single_source|outdated|conflicting|unverified|sensational|number_unverified",
      "detail": "설명",
      "severity": "high|medium|low",
      "recommendation": "Suggested action to resolve"
    }
  ]
}

Phase 4: Fact-Check Flagging

Automatically detect and flag potential issues. Run these checks on every claim:

Auto-Detection Rules

Severity Levels

• high — Must verify before using in script. Could damage credibility if wrong.
• medium — Should verify. Add hedging language if used without verification.
• low — Note for awareness. Minor risk.

Phase 5: Output

File Output

Save results to the project directory:

• Research JSON: research/{topic_slug}/research.json — full structured data
• Research Brief: research/{topic_slug}/brief.md — human-readable summary

Topic Slug Generation

Convert topic to filesystem-safe slug:

• Korean: Use romanization or short English equivalent
• Remove special characters, replace spaces with hyphens
• Lowercase, max 50 chars
• Example: "양자 컴퓨팅의 현재와 미래" → quantum-computing-present-future

Brief Format (brief.md)

# Research Brief: {topic}

**Depth**: {depth} | **Sources**: {count} | **Date**: {date}

## Summary
{1-2 paragraph overview}

## Key Claims
1. {claim} — [{reliability}] (source: {source_title})
2. ...

## Key Numbers
- {value}: {context} (source: {source_title}, {date})
- ...

## Timeline
- {date}: {event}
- ...

## Fact-Check Flags
### High Severity
- [{claim_id}] {detail} — {recommendation}

### Medium Severity
- [{claim_id}] {detail}

## Sources
1. [{title}]({url}) — {type}, {date}, {reliability}
2. ...

Console Output

After saving files, display:

1. Summary paragraph
2. Claim count and source count
3. Fact-check flags (high severity highlighted)
4. File paths for saved output

Multi-Channel Support

Load channel-specific config to adjust research behavior.

Channel Config (channels/{channel_id}/config.json)

{
  "channel_id": "tech-shorts",
  "name": "채널명",
  "default_depth": "standard",
  "topic_domains": ["tech", "science", "AI"],
  "tone": "casual|professional|educational",
  "language_priority": "ko|en|both",
  "research_preferences": {
    "prefer_korean_sources": true,
    "include_community_sentiment": false,
    "max_source_age_months": 12
  }
}

When --channel is provided:

• Load config and apply default_depth if no explicit depth given
• Filter sources by topic_domains relevance
• Adjust source language priority
• Apply max_source_age_months to outdated flagging threshold

Integration

• Input to yt-script: Research JSON serves as the primary input for script generation. The script skill reads research/{topic_slug}/research.json directly.
• Standalone use: Can be triggered independently for general research without video production intent.
• Pipeline: yt-research → yt-script → yt-thumbnail (planned)

Data Location

• Research output: research/{topic_slug}/research.json
• Research brief: research/{topic_slug}/brief.md
• Channel configs: channels/{channel_id}/config.json