YouTube Summarize (Pipeline)

End-to-end pipeline that summarizes YouTube videos by orchestrating existing skills in sequence. Supports both single video URLs and channels (summarizes latest N videos).

Quick Start

/mk-youtube-summarize <URL|Channel> [count] [--force]

Examples

# Single video (existing behavior)
/mk-youtube-summarize https://www.youtube.com/watch?v=dQw4w9WgXcQ
/mk-youtube-summarize https://youtu.be/dQw4w9WgXcQ

# Channel's latest 3 videos (default)
/mk-youtube-summarize @RickAstleyYT
/mk-youtube-summarize https://www.youtube.com/@RickAstleyYT

# Channel's latest 5 videos
/mk-youtube-summarize @RickAstleyYT 5

# Get channel from video URL, then summarize latest 3
/mk-youtube-summarize https://www.youtube.com/watch?v=dQw4w9WgXcQ channel
/mk-youtube-summarize https://www.youtube.com/watch?v=dQw4w9WgXcQ channel 5

Pipeline Flow

Single Video Flow

URL (single video)
 │
 ▼
┌──────────────────────────────────────┐
│ /mk-youtube-transcript-summarize     │  ← Step 0.5: Check cache
│          --check <URL>               │
└────────────┬─────────────────────────┘
        ┌────┴────┐
        │         │
   exists:true  exists:false
        │         │
        ▼         ▼
┌─────────────┐  ┌───────────────────────────┐
│ Read & show │  │ /mk-youtube-get-info      │  ← Step 1
│  ✅ DONE    │  └────────────┬──────────────┘
└─────────────┘               │
                         ┌────┴────┐
                         │         │
                    has_subs    no_subs
                         │         │
                         ▼         ▼
                 ┌───────────────┐ ┌─────────────────┐
                 │/mk-youtube-   │ │/mk-youtube-     │  ← Step 2a or 2b
                 │get-caption    │ │get-audio        │
                 └───────┬───────┘ └────────┬────────┘
                         │                  │
                         │                  ▼
                         │         ┌─────────────────┐
                         │         │/mk-youtube-     │  ← Step 2c (audio path only)
                         │         │audio-transcribe │
                         │         └────────┬────────┘
                         │                  │
                         └────────┬─────────┘
                                  │
                                  ▼
                 ┌────────────────────────────────┐
                 │ /mk-youtube-transcript-summarize │  ← Step 3: MANDATORY
                 └────────────────────────────────┘

Channel Flow

Channel (@handle, channel URL, or video URL + "channel")
 │
 ▼
┌─────────────────────────────────┐
│ /mk-youtube-get-channel-latest  │  ← Step 0: Get latest N video URLs
│          [channel] [count]      │
└────────────┬────────────────────┘
             │
             ▼
┌─────────────────────────────────┐
│ For each video (1..N):          │
│   Step 0.5: --check cache       │
│   ├─ exists:true → Read & show  │
│   └─ exists:false → Full        │
│      pipeline (Steps 1 → 2 → 3) │
│   before moving to next video   │
└─────────────────────────────────┘
             │
             ▼
        ✅ All summaries displayed

Steps

Step 0: Detect Input Type & Extract URL

Determine whether input is a single video or a channel:

Single Video → proceed to Step 1:

• URL contains youtube.com/watch?v=
• URL contains youtu.be/
• URL contains youtube.com/shorts/ (single short)

Channel → proceed to Channel Path:

• URL contains youtube.com/@
• URL contains youtube.com/channel/
• URL contains youtube.com/c/
• Input starts with @ (e.g., @RickAstleyYT)
• Input is plain text without URL format (treated as @handle)
• User explicitly provides channel keyword after video URL

Step 0.5: Check for Existing Summary (unless –force)

Skip this step if --force is specified.

For single video input, use the Skill tool to invoke /mk-youtube-transcript-summarize with --check:

/mk-youtube-transcript-summarize --check <URL>

If exists: true:

1. Read the summary file at output_summary using the Read tool
2. Display the COMPLETE summary content (Full Display Rule applies)
3. Include metadata footer (title, channel, url from the check output)
4. STOP — do NOT proceed to Steps 1-3

If exists: false:

• Proceed to Step 1 as normal

For channel flow, the check happens per-video inside the loop after getting the video list. Each video URL is checked individually.

Channel Path: Get Latest Videos

Use the Skill tool to invoke /mk-youtube-get-channel-latest:

/mk-youtube-get-channel-latest <channel> [count]

• Default count: 3 (if not specified)
• Maximum recommended: 10 (to avoid excessive processing time)

The output is a JSON array of video objects. For each video in the array, execute the full Single Video pipeline (Steps 1 → 2 → 3) completely before moving to the next video.

Progress Display (recommended format):

Processing 3 videos from @RickAstleyYT...

[1/3] Rick Astley - Never Gonna Give You Up
→ ✅ Cached summary found, displaying...

[2/3] Rick Astley - Together Forever
→ Summary generated (new)

[3/3] Rick Astley - Take Me to Your Heart
→ ✅ Cached summary found, displaying...

✅ Completed: 3/3 summaries displayed (2 cached, 1 new)

Step 1: Get Video Info

Use the Skill tool to invoke /mk-youtube-get-info with the URL:

/mk-youtube-get-info <URL>

From the output, note:

• has_subtitles and has_auto_captions — determines the path for Step 2
• language — used for model selection if audio transcription is needed

Step 2: Get Transcript

Choose ONE path based on subtitle availability from Step 1:

Path A — Subtitles available (has_subtitles: true OR has_auto_captions: true)

Use the Skill tool to invoke /mk-youtube-get-caption:

/mk-youtube-get-caption <URL>

Save the text_file_path from the output for Step 3.

Path B — No subtitles (has_subtitles: false AND has_auto_captions: false)

First, use the Skill tool to invoke /mk-youtube-get-audio:

/mk-youtube-get-audio <URL>

Then, use the Skill tool to invoke /mk-youtube-audio-transcribe with the file_path from the audio output:

/mk-youtube-audio-transcribe <file_path> auto <language>

Pass the language from Step 1 for best model auto-selection (e.g., zh → belle-zh, ja → kotoba-ja).

Save the text_file_path from the output for Step 3.

Step 3: Generate Summary — MANDATORY — NEVER SKIP

Use the Skill tool to invoke /mk-youtube-transcript-summarize with the text_file_path obtained from Step 2:

/mk-youtube-transcript-summarize <text_file_path> [--force]

Pass --force if the user specified it for mk-youtube-summarize.

CRITICAL: You MUST use the Skill tool to invoke /mk-youtube-transcript-summarize. Do NOT generate summaries directly. The skill contains critical rules for:

• Compression ratio calibration
• Section structure requirements
• Numerical data preservation
• Language handling

Skipping this step or generating summaries without the skill will produce lower-quality output.

Display Rule: The /mk-youtube-transcript-summarize skill will generate and save the full summary. You MUST display the COMPLETE summary content to the user as-is. Do NOT create a condensed or abbreviated version after the summary is generated.

Processing Multiple Videos

When summarizing multiple videos (either from a channel or manually specified):

• Execute ALL steps (1 → 2 → 3) for EACH video independently
• Do NOT batch Step 1 for all videos and then do Step 3
• Complete the full pipeline per video before moving to the next
• Display progress after each video completion

Notes

• This is a pure orchestration skill — it does not have its own scripts
• Summary caching: Before running the full pipeline, checks if a cached summary exists via --check. Use --force to bypass and regenerate.
• Each sub-skill handles its own dependency management (yt-dlp, jq, whisper-cli, etc.)
• Each sub-skill also has its own file caching — repeated runs use local cached files
• For batch processing, consider limiting count to avoid context overflow
• Channel processing uses /mk-youtube-get-channel-latest which auto-detects channel from video URLs