YouTube Research Workflow Skill

Comprehensive research system for YouTube video/series planning. Uses 5 specialized agents to gather, analyze, and synthesize research into actionable content strategy.

CRITICAL: Direct Agent Spawning

You (the main Claude agent) MUST spawn the gatherer and strategist agents directly.

Agents cannot spawn subagents, so there is no orchestrator. You handle the orchestration by:

•Creating folders
•Spawning 4 gatherers in parallel
•Waiting for completion
•Spawning the strategist

Architecture

code

/youtube research "Topic"
         │
    [Main Claude Agent handles orchestration]
         │
         ├── @yt-topic-gatherer (Haiku)      ─┐
         ├── @yt-competitor-gatherer (Haiku) ─┼─ Phase 1: Parallel
         ├── @yt-seo-gatherer (Haiku)        ─┤
         └── @yt-community-gatherer (Haiku)  ─┘
                      │
                      ▼
         └── @yt-research-strategist (Opus)  ─── Phase 2: Sequential
                      │
                      ▼
              research-pack.md + series-structure.md

Agent Summary

Agent	Model	Purpose	Token Limit
`@yt-topic-gatherer`	Haiku	Subject matter, docs, features	1,000
`@yt-competitor-gatherer`	Haiku	YouTube videos, gaps	1,000
`@yt-seo-gatherer`	Haiku	Keywords, trends, titles	800
`@yt-community-gatherer`	Haiku	Reddit, forums, questions	800
`@yt-research-strategist`	Opus	Synthesis, series structure	N/A

Total: 5 agents (4 gatherers + 1 strategist)

Workflow Phases

Phase 1: Parallel Gathering (~2-3 min)

Main Claude agent spawns 4 Haiku agents simultaneously (single message with 4 Task calls):

Agent	Researches	Output Summary
Topic	Official docs, features, complexity	topic-summary.md
Competitor	Existing YouTube videos, gaps	competitor-summary.md
SEO	Keywords, trends, titles	seo-summary.md
Community	Reddit, forums, questions	community-summary.md

Each agent:

•Gathers comprehensive raw data
•Saves raw data to raw/ folder
•Creates condensed summary under token limit
•Saves summary to summaries/ folder

Phase 2: Strategic Synthesis (~3-5 min)

After all 4 gatherers complete, main Claude agent spawns Opus strategist:

•Reads all 4 summaries (~3,200 tokens)
•Uses sequential thinking for complex analysis
•Determines single video vs series
•Creates research-pack.md
•Creates series-structure.md (if series)

Output Structure

code

03-YouTube/research/YYYY-MM-DD-[slug]/
├── research-pack.md        # Strategic recommendations (always)
├── series-structure.md     # Episode breakdown (if series)
├── summaries/              # Condensed inputs (~3,200 tokens)
│   ├── topic-summary.md
│   ├── competitor-summary.md
│   ├── seo-summary.md
│   └── community-summary.md
└── raw/                    # Full research data
    ├── topic-raw.md
    ├── competitor-raw.md
    ├── seo-raw.md
    └── community-raw.md

Research Pack Contents

Section	Description
Executive Summary	4-5 bullet points + recommendation
Topic Overview	Features, complexity, prerequisites
Competitor Landscape	Existing videos, gaps, differentiation
Target Audience	Who, what they know, what they want
SEO Strategy	Keywords, title suggestions, tags
Community Insights	Questions to answer, pain points
Content Recommendation	Single video or series + reasoning
Production Notes	Demo requirements, difficulty score
Next Steps	What to do after research

Series Detection Criteria

Recommend Series When:

•Topic has 3+ distinct sub-topics
•Each sub-topic needs 10+ minutes coverage
•Clear learning progression exists
•Community has questions at multiple levels
•Too complex for single video

Recommend Single Video When:

•Topic is focused and contained
•Can cover in 15-30 minutes
•No natural breaking points
•Simple enough for one session

Tools Used by Agents

Topic Gatherer

•Perplexity (search, ask)
•Context7 (library docs)
•DeepWiki (GitHub projects)
•WebFetch

Competitor Gatherer

•Perplexity (search)
•WebSearch
•WebFetch

SEO Gatherer

•Perplexity (search)
•WebSearch

Community Gatherer

•Perplexity (search)
•WebSearch
•WebFetch

Strategist

•Sequential Thinking (complex analysis)
•Read, Write, Glob

Usage Examples

bash

# Research for potential series
/youtube research "XCloud tutorial series"

# Research for single video
/youtube research "Docker networking basics"

# Research for advanced topic
/youtube research "Kubernetes security hardening"

# Research for beginner content
/youtube research "Self-hosting for beginners"

Timing Expectations

Phase	Agents	Duration
Setup	Main agent	~10 sec
Phase 1	4 Haiku (parallel)	~2-3 min
Phase 2	1 Opus	~3-5 min
Total	5 agents	~5-8 min

Integration with YouTube Workflow

code

/youtube research "Docker Security"
    ↓
review: research-pack.md
    ↓
/youtube full "Docker Security"  (or Part 1 if series)
    ↓
/youtube publish "Docker Security"
    ↓
/social video "docker-security"

Token Efficiency

Problem Solved: If 4 gatherers each return 10,000 tokens, strategist receives 40,000+ tokens - too much.

Solution: Each gatherer summarizes to strict token limits:

•Topic: 1,000 tokens max
•Competitor: 1,000 tokens max
•SEO: 800 tokens max
•Community: 800 tokens max
•Total to strategist: ~3,600 tokens

Raw data preserved in raw/ folder for reference if needed.

Quality Checklist

Before research is complete:

• Folders created (summaries/ and raw/)
• All 4 gatherers ran in parallel (single message)
• All 4 summary files exist
• Summaries are within token limits
• Raw data preserved
• research-pack.md is comprehensive
• Series detection reasoning is clear
• series-structure.md exists (if series)
• Next steps are actionable