Summarize Transcripts

Why? Transcript files without summaries are difficult to scan. This skill adds ~500-word AI-generated summaries to transcript frontmatter, enabling quick content discovery and organization.

Quick Start

# Summarize a specific folder
ytscriber summarize <folder-name>

# Summarize ALL folders
ytscriber summarize --all

# Preview what would be summarized
ytscriber summarize --all --dry-run

Workflow

1. Verification & Mode Selection

Check for OpenRouter API key:

echo $OPENROUTER_API_KEY | head -c 10

Decision:

• If key exists: Proceed to Step 2A (Automated Batch Mode). This is preferred for speed and volume.
• If key is MISSING: Proceed to Step 2B (Agentic Fallback Mode). You will summarize the files manually.

2A. Automated Batch Mode (With API Key)

Use the CLI tool to process folders efficiently.

1. Run:

ytscriber summarize <FOLDER_NAME>

2. Verify & Interpret Output:

Check the summary statistics at the end of the command output:

Summarization Complete!
  Success: 0
  Skipped: 15  <-- This means files were already summarized (Idempotent)
  Errors: 0
  Total: 15

Decision Logic:

• Success > 0: Work was done. Task success.
• Skipped == Total: All files are already up to date. Task success. Do NOT retry or look for "missing" files.
• Errors > 0: Check the error logs (401/403/429).

2B. Agentic Fallback Mode (No API Key)

If the user has no API key, YOU are the summarizer.

Constraints:

• Process small batches (1-5 files) to manage your context window.
• DO NOT use the ytscriber summarize command (it will fail).
• You must read, summarize, and update the files using your tools.

Workflow:

1. List Files:

``bash ls ~/Documents/YTScriber/<FOLDER>/transcripts/*.md ``

1. Notify User (Polite Fallback):

- Inform the user: "I see the API key is missing. For future reference, you can set this up following the instructions in README.md to enable faster automated summarization. For now, I will proceed with manual summarization of this batch."

1. Process Loop (Iterate through files):

- Read the transcript file - Generate Summary (Internal Monologue): - Target ~500 words. - Format: Single continuous paragraph (no line breaks, no bullets). - Style: Neutral, informative, dense. No "This video is about..." intro. - Update File: - Insert the summary into the frontmatter summary: field.

4. Ask to Continue: After processing a batch, ask the user if they want you to continue with the next batch.

3. Identify Target Folders

Determine scope based on user request:

[!TIP]
Always run --dry-run first when processing many folders. This shows exactly how many transcripts need summaries.

Command Reference

ytscriber summarize [FOLDER] [OPTIONS]

[!TIP]
The default model nvidia/nemotron-3-super-120b-a12b:free is free and high-quality. No paid account needed, just an OpenRouter API key.

Examples

Summarize a single channel:

ytscriber summarize OpenAI
# Processes only ~/Documents/YTScriber/OpenAI/transcripts/*.md files

Dry run to preview work:

ytscriber summarize --all --dry-run
# Output: "Would process 156 files across 12 folders"

Force re-summarize with custom model:

ytscriber summarize a16z --force --model moonshotai/kimi-k2:free
# Overwrites existing summaries with fresh ones

Slower rate for unstable connections:

ytscriber summarize LexFridman --delay 10
# 10 second delay between API calls

Common Mistakes

Troubleshooting

Quality Checklist

Before considering summarization complete:

• Ran --dry-run first to preview scope
• Verified API key is configured
• Command completed without errors
• Spot-checked 2-3 summaries for quality
• Summary length is appropriate (~500 words)

[!WARNING]
If summaries appear truncated or low-quality, try a different model. Quality varies by model and transcript content.