nicheking-watch
Decode any YouTube video against your specific niche. Point Claude at a competitor's video, your own back catalog, or an inspiration video — get back a structured editing-pattern breakdown scored against the top performers in YOUR niche, plus draft hooks and titles adapted to YOUR voice.
/nicheking-watch https://youtu.be/<competitor-video>
That's the entire workflow. Claude downloads the video, extracts scene-aware frames, pulls the timestamped transcript, fetches your niche profile + research patterns from Niche King, and writes a decode that goes straight into your library.
What you get back
Not a generic "here's what's in this video" summary. A YouTube-creator-specific editing decode:
- Hook decode — when the hook arrives, what pattern type, the actual transcript, and a draft hook adapted to your niche and voice
- Pacing decode — first cut timestamp, average cut length, B-roll cadence, pattern-interrupt timestamps, talking-head-to-B-roll ratio
- Structure — chapter breaks with estimated retention per section
- CTA placements — exact timestamps for early / mid / end CTAs with the actual copy used
- Visual patterns — on-screen text style, color grade, framing default, signature moves visible in frames
- Niche match score (0-100) — which of THIS video's patterns match the top performers in YOUR niche's research, which are missing, and which moves DON'T translate
- Adapt for creator — single most important takeaway + tactical steps + 3-5 draft titles for your next video drawing from this video's hook angle
- Risk notes — what NOT to copy because it won't fit your niche/voice/audience
The decode saves to your Niche King library so the editorial-brief tool, script writer, and any other AI editor connected via MCP can reference it later.
Why this exists
Niche King is the complete YouTube strategy platform — niche → topics → research → strategy → AI integration → create → improve. We analyze 500+ videos per niche through transcripts and thumbnail vision. But transcripts can't tell you cut cadence, B-roll usage, signature visual moves, on-screen text patterns. Frames can.
nicheking-watch is the missing visual layer. It runs on YOUR machine using YOUR Claude (or Cursor / Windsurf / Codex) subscription, extracts the visual data Niche King's server-side research can't, and feeds it back into Niche King's niche pattern matching.
Install
| Surface | Command |
|---|
| Claude Code | /plugin marketplace add Video-Niche-King-LLC/nicheking-watch then /plugin install nicheking-watch@nicheking-watch |
| claude.ai (web) | Download nicheking-watch.skill from the latest release → Settings → Capabilities → Skills → + |
| Codex | git clone https://github.com/Video-Niche-King-LLC/nicheking-watch ~/.codex/skills/nicheking-watch |
Setup
You need a Niche King API key (free for any active subscriber, generated at https://app.nicheking.video/integrations).
mkdir -p ~/.config/nicheking
cat > ~/.config/nicheking/.env <<EOF
NICHEKING_API_KEY=sk-nk-your-key-here
EOF
chmod 600 ~/.config/nicheking/.env
For Whisper fallback (only used when a video has no caption track) you can optionally add a Groq or OpenAI key:
cat >> ~/.config/nicheking/.env <<EOF
GROQ_API_KEY=gsk_... # preferred — cheaper, faster
OPENAI_API_KEY=sk-... # alternative
EOF
Usage
# Basic — decode a competitor's video
/nicheking-watch https://youtu.be/<video>
# Decode a local file
/nicheking-watch ~/Videos/competitor.mp4
# Focus on a specific range (useful for long videos)
/nicheking-watch https://youtu.be/<video> --start 5:00 --end 25:00
# Bump frame budget for tightly-edited videos
/nicheking-watch https://youtu.be/<video> --max-frames 120
# Skip Whisper (frames-only mode if no captions)
/nicheking-watch https://youtu.be/<video> --no-whisper
# Without an API key — generic decode without niche scoring
/nicheking-watch https://youtu.be/<video> --no-niche
Flags: --start, --end, --max-frames, --resolution, --scene-threshold, --max-gap, --whisper groq|openai, --no-whisper, --no-niche, --out-dir.
What this costs you
- Niche King: $0 — the API endpoints (fetch context + save decode) are free for any active subscriber.
- Video download + native captions: $0 — yt-dlp + ffmpeg, both free.
- Whisper fallback: Only kicks in when a video has no caption track. ~$0.01-0.05 per video on Groq.
- Claude / Codex inference: Charged to your existing AI subscription, not Niche King. The skill runs locally on your machine.
A typical 10-minute video decode costs about $0.50-$2 in Claude inference (depends on frame count and your tier) and $0 if it has captions, plus a fraction of a cent in Whisper if it doesn't.
Re-running the same video
