Youtube Planner

{
  "sources": [
    "@realrobtheaiguy",
    "@aiadvantage",
    "@matthewberman"
  ]
}

How to run

youtube-planner

Or target specific date:

youtube-planner --date 2026-02-16

Workflow

1. Fetch latest videos for each tracked channel (Python script)

Use direct Python/curl approach for efficient batch processing:

• Read channels from sources.json
• Fetch latest 15 videos from each channel (free endpoint) using curl in parallel
• Save each to workspace/group/youtube/{date}/{channel-name}/latest.json
• First run: Filter to videos published in last 30 days
• Subsequent runs: Filter to videos published in last 24 hours

First run detection: Check if any metadata files exist in workspace/group/youtube/. If empty, this is the first run.

2. Process all videos (Python script)

Use direct Python script to process all videos efficiently:

1. Collect all videos from latest.json files that match the time range
2. For each video:
   • Check transcript cache: find workspace/group/youtube -path "*/{channel-slug}/{title-slug}/transcript.txt"
   • If not found, fetch transcript via curl to TranscriptAPI
   • Save transcript.json and transcript.txt
   • Always save new metadata snapshot to metadata/{timestamp}.json
3. Use rate limiting (0.25s between API calls) to respect API limits
4. Process in batches or all at once depending on count

Python approach is faster and more reliable than spawning hundreds of agents.

3. Generate dashboard

Run the dashboard generator:

python3 .claude/skills/youtube-planner/generate-dashboard.py

Output: workspace/group/youtube/dashboard.html

Dashboard features

• Sortable columns: Click headers to sort by VPH, views, date
• Default sort: Highest VPH (views per hour since publish)
• Columns: Thumbnail, Title, Channel, Published, Views, VPH, Trend
• Trend sparklines: Mini chart showing view count progression from all metadata snapshots
• Click thumbnails: Opens video in new tab

Trend visualization

The Trend column shows an inline sparkline chart based on all metadata files in metadata/:

• X-axis: Time (from first snapshot to latest)
• Y-axis: View count
• Visual: SVG mini line chart (80px × 30px)
• Color: Green if accelerating, red if decelerating, gray if flat

Example progression:

metadata/2026-02-16-0800.json → 1,200 views
metadata/2026-02-16-1200.json → 2,500 views
metadata/2026-02-16-1600.json → 4,100 views

Sparkline shows upward curve → green chart

File organization

workspace/group/youtube/
  dashboard.html
  2026-02-16/
    rob-the-ai-guy/
      latest.json
      google-geminis-new-upgrades/
        transcript.txt
        transcript.json
        metadata/
          2026-02-16-0800.json
          2026-02-16-1200.json
          2026-02-16-1600.json

VPH calculation: views / hours_since_publish

Trend direction: Compare VPH from last 2 snapshots:

• If latest VPH > previous VPH by 10%+: accelerating (green)
• If latest VPH < previous VPH by 10%+: decelerating (red)
• Otherwise: flat (gray)

Files created

Rules

• First run behavior: Pull all videos from last 30 days to bootstrap data
• Subsequent runs: Only process videos from last 24 hours to track recent uploads
• Transcript is cached forever (never re-fetch unless user says "refresh")
• Metadata is NEVER cached — always save new timestamped file in metadata/ subdirectory
• Dashboard updates every time skill runs
• Channel list is in sources.json at skill base directory
• Sparklines render for videos with 2+ metadata snapshots

Adding new sources

When the user asks to add a new channel (e.g., "add @ChannelHandle to sources"), automatically:

1. Update sources.json:

   # Read current sources
   sources=$(cat .claude/skills/youtube-planner/sources.json)
   # Add new channel to the list
   # Use Edit tool to add the channel
   

2. Fetch last 30 days of data:

   python3 .claude/skills/youtube-planner/catch-up-channel.py @ChannelHandle
   

3. Regenerate dashboard:

   python3 .claude/skills/youtube-planner/generate-dashboard.py
   

Important: Always use the catch-up script for new channels to bootstrap with 30 days of historical data, even if other channels are in "24-hour mode".

Example usage

# First run: Fetches all videos from last 30 days
youtube-planner

# Subsequent runs: Only fetch videos from last 24 hours
youtube-planner

# Force specific date range (override auto-detection)
youtube-planner --date 2026-02-15

# Add new tracked channel (handled automatically)
# User says: "add @DanKoeTalks to sources"
# Skill will: update sources.json → fetch 30 days → regenerate dashboard

Output

At the end, confirm:

• Number of channels checked
• Number of new transcripts fetched (API credits used)
• Number of metadata snapshots saved
• Dashboard location: workspace/group/youtube/dashboard.html