Youtube Researcher

YouTube Researcher

Overview

Use this skill for YouTube topic research, niche scouting, competitor scans, channel review, and transcript collection. Prefer the bundled script over ad hoc API calls so the workflow stays fast, repeatable, and consistent across chats.

Quick Start

Run the helper script directly:

python3 "$HOME/.codex/skills/youtube-researcher/scripts/youtube_research.py" --help

Common commands:

python3 "$HOME/.codex/skills/youtube-researcher/scripts/youtube_research.py" search "ai coding agents" --limit 10 --json

python3 "$HOME/.codex/skills/youtube-researcher/scripts/youtube_research.py" search "b2b saas onboarding" --limit 8 --transcripts 5 --excerpt-chars 900 --json

python3 "$HOME/.codex/skills/youtube-researcher/scripts/youtube_research.py" channel "@ycombinator" --limit 12 --with-transcripts 3 --json

python3 "$HOME/.codex/skills/youtube-researcher/scripts/youtube_research.py" transcript "https://www.youtube.com/watch?v=dQw4w9WgXcQ" --lang en

Workflow

1. Search the niche

Start with search when the user is exploring a topic, niche, idea cluster, or competitor landscape.

• Use --limit to pull more videos.
• Use --channel "name" to keep only results whose channel name matches a substring.
• Use --sp when the user wants a YouTube-native filter token copied from a filtered YouTube URL.
• Use --no-cache for fast-changing topics when freshness matters.

2. Pull transcripts only where useful

Transcripts cost more requests than search, so avoid fetching them for everything unless the user clearly needs them.

• Use --transcripts N on search to fetch transcripts for the top N search results.
• Use transcript for one-off deep dives.
• Use --mode native if the user wants only existing captions and wants to avoid AI-generated fallback.
• Use --mode auto for the default fast path.

3. Inspect channels directly

Use channel when the user wants to understand a creator, competitor, or niche-specific publisher.

• It fetches Supadata channel metadata plus recent video IDs.
• It enriches returned videos with metadata so titles, views, dates, and thumbnails are available without extra work.
• Use --with-transcripts N to attach transcript excerpts for the first N channel videos.

4. Synthesize in-chat

After using the script, do the analysis in the chat:

• recurring hooks or thumbnail/title patterns
• topic clusters and content gaps
• channel positioning
• common claims, framing, and audience promises
• relevant videos worth deeper review

Output Guidance

• Prefer --json when the result will be analyzed further in the chat.
• Prefer plain output when the user mainly wants a quick list or transcript excerpt.
• For “latest” or time-sensitive requests, pair --no-cache with concrete dates in the final answer.

Keys

This shareable version expects API keys from environment variables:

• SUPADATA_API_KEY
• SERPAPI_KEY

Do not commit API keys into this repo. Export them in the shell, use a local .env, or store them in the team's preferred secret manager.

References

• references/api-notes.md for the current endpoint notes and parameter reminders