YouTube chapters from a transcript

What it does

Builds the chapter list (table of contents) for a YouTube description from the video's transcript. Your agent finds topic changes and writes titles; the bundled script handles the format checks.

• Reads SRT/VTT captions, speech-to-text JSON (start/end/text), 0:00 text lines and transcripts copied from YouTube
• Condenses long transcripts into windows (for example 30 s) so the agent can read the whole video
• Validates the proposed chapters against YouTube's rules: first at 0:00, at least 3, each at least 10 seconds, none after the end
• Snaps each chapter start to the nearest sentence start in the transcript
• Shows style hints such as duplicate or very long titles
• Prints 00:00 style under one hour and 0:00:00 for longer videos, consistently

How to use

python scripts/chapters.py parse transcript.srt --window 30 --out condensed.txt
python scripts/chapters.py validate chapters.txt --transcript transcript.srt --snap 5

Ask your agent to make chapters from a transcript. It reads the condensed transcript, proposes chapters, fixes them until validation passes and returns a paste-ready list.

Requirements

• Python 3 (standard library only)

Caveats

• It does not connect to YouTube or read video files; you paste the result into the description yourself.
• Plain text transcripts have no video length; pass --duration so the last chapter can be checked too.
• If an automatic transcript is inaccurate, the titles can only be as good as the transcript.