youtube-videos

name	youtube-videos
description	Use when the user provides a YouTube playlist link/ID or YouTube video links/IDs and wants those recordings added to an event videos.yml file.
allowed-tools	event_talks event_lookup youtube_videos speaker_lookup

Adding YouTube Videos to videos.yml

Use this skill when the user gives you:

• a YouTube playlist URL or playlist ID, or
• one or more YouTube video URLs or video IDs,

and wants those recordings added to data/{series-slug}/{event-slug}/videos.yml.

Important Notes

• Review docs/ADDING_VIDEOS.md before editing.
• If you need unpublished or scheduled talks instead of published recordings, switch to docs/ADDING_UNPUBLISHED_TALKS.md.
• videos.yml entries with video_provider: "youtube" must end with real values for:
  • date
  • published_at
  • video_id
  • speakers
• Videos without speakers will not display.
• Keep talks ordered by their actual presentation order. schedule.yml auto-maps empty slots from videos.yml in order, so published date order is not enough.
• Prefer the RubyEvents YouTube tools first. Use the Rails scripts when they are the better fit for a full-playlist import.
• When a published YouTube recording matches an existing video_provider: "scheduled" entry, keep the existing title and only update raw_title from YouTube metadata.

Related Scripts

These scripts are documented in docs/ADDING_VIDEOS.md and are useful context for deciding the import path:

• scripts/prepare_series.rb: fills youtube_channel_id in series.yml from youtube_channel_name
• scripts/create_events.rb: creates event.yml files from playlists on a YouTube channel
• scripts/extract_videos.rb: regenerates an event videos.yml from the playlist ID in event.yml

Use scripts/extract_videos.rb when all of the following are true:

1. the user wants a whole playlist imported for one event
2. the playlist is the canonical event playlist
3. replacing or regenerating videos.yml is acceptable

This matters because extract_videos.rb:

• uses the playlist ID stored in event.yml
• honors metadata_parser from event.yml
• sorts by published_at, so you may still need to reorder the final file for schedule order
• does not finish the job by itself because date and sometimes speakers still need manual completion

Tools to Prefer

• rubyevents-event_lookup to find the target event
• rubyevents-youtube_playlist to inspect playlist metadata
• rubyevents-youtube_playlist_items to fetch playlist videos
• rubyevents-youtube_video for a single video
• rubyevents-youtube_videos for batches of video IDs

Helpful supporting tools when needed:

• rubyevents-speaker_lookup to check whether speakers already exist
• rubyevents-event_talks to compare against existing talk records

Workflow

1. Identify the target event and inspect:
   • data/{series}/{event}/event.yml
   • data/{series}/{event}/videos.yml if it already exists
2. Decide whether this is:
   • a full playlist import, or
   • a partial import / append / update from selected videos
3. Extract IDs from the user input:
   • playlist URL -> list=...
   • watch URL -> v=...
   • short URL -> path segment after youtu.be/
4. Fetch metadata:
   • playlist: use rubyevents-youtube_playlist and rubyevents-youtube_playlist_items
   • video list: use rubyevents-youtube_video or rubyevents-youtube_videos
5. Convert YouTube metadata into RubyEvents entries.
6. Merge into videos.yml without creating duplicates.
7. Reorder the file to match real talk order if the event has a schedule or multi-day structure.
8. Run bin/lint.

Building Entries

Each imported YouTube talk should usually end up with fields like:

- id: "speaker-or-title-event-slug"
  title: "Talk title"
  raw_title: "Original YouTube title"
  date: "YYYY-MM-DD"
  published_at: "YYYY-MM-DD"
  description: "Talk description"
  video_provider: "youtube"
  video_id: "abcdefghijk"
  speakers:
    - "Speaker Name"

Follow the repo's YouTube parsing conventions:

• remove the event name from the title when it is redundant for newly created YouTube entries
• keep raw_title when you normalize the title
• if you are updating an existing scheduled talk, preserve its curated title and only refresh raw_title
• split speaker suffixes the same way the parsers do:
  • " by "
  • " & "
  • ", "
  • " and "
• keep keynote or lightning-talk session titles intact when the parser would normally do so

If event.yml specifies a custom metadata_parser, trust that behavior when using scripts/extract_videos.rb.

Choosing the Edit Strategy

Full playlist for one event

Prefer this path when the playlist represents the whole event:

1. confirm the target event
2. confirm the playlist ID
3. if needed, update event.yml so its id matches the playlist ID
4. run:

bin/rails runner scripts/extract_videos.rb <series-slug> <event-slug>

Then finish the file:

• replace placeholder or missing date values with real talk dates
• fix any missing or incorrect speakers
• reorder talks to match the event program, not YouTube publish order
• preserve or restore fields like track, slides, or other manual metadata if the file already had them

Selected videos or partial playlist import

Do not use scripts/extract_videos.rb for this.

Instead:

1. fetch only the videos you need with the RubyEvents YouTube tools
2. map them into videos.yml entries
3. append new items or update matching items in place

Match existing entries by video_id first. If there is no existing video_id match, use the talk id only when you are sure it is the same talk. If the matched entry is a scheduled talk, preserve its existing title and only replace raw_title, published_at, video_provider, and video_id.

Generator Guidance

The talk generator is useful for unpublished talks, but it is not the final source of truth for YouTube imports because it creates entries with:

• video_provider: "scheduled"
• video_id: "<talk-id>"

Only use the generator to bootstrap a missing videos.yml if that is genuinely helpful. For published YouTube recordings, make sure the final saved entry is a real YouTube entry.

Date and Speaker Rules

• Never use the YouTube publish date as the talk date unless the event itself happened that day.
• For a single-day event, it is usually safe to use the event date for every imported talk.
• For multi-day events, do not guess when the correct day is unclear; ask the user or infer it from the schedule/title/playlist organization.
• If speakers cannot be derived confidently from the title or description, ask the user instead of inventing them.

Speaker Records

If you add a speaker who is not already in data/speakers.yml, add the speaker record too when the task clearly includes that scope. Prefer including a GitHub handle when you can find one.

Final Checks

• videos.yml is valid YAML
• every imported YouTube entry has real date, published_at, video_id, and speakers
• no duplicate video_id entries were introduced
• talk order matches the program, especially when schedule.yml exists
• bin/lint passes