You finish a 40-minute video and the chores begin: scroll the timeline to write chapters, hunt for a line worth posting, remember what you actually said in minute 23. It is slow, and it is the part most people skip.
I wrote a small Hermes skill, transcript-to-chapters, to take over the mechanical half. You give it the subtitle file your editor or platform already exports (SRT or VTT). A tiny script with no dependencies and no API key reads the timestamps, compares neighbouring stretches of speech and flags where the vocabulary changes, which usually means a new topic. It also lists sentences that could work as quotes.
The agent does the part that needs judgement: it writes short, honest chapter titles, moves a boundary to the sentence that really starts the topic, and drafts a short summary using only what was said. Quotes are copied word for word with their timestamp, never reworded.
Before you paste anything, it checks the rules YouTube applies to chapters: the first one starts at 0:00, there are at least three, and each lasts at least ten seconds.
Example result for a baking video:
0:00 Starter and flour basics
2:10 Folding the dough
3:00 Preheating and baking
Honest limits: the boundaries are suggestions based on word counts, so short or uniform talks may need manual fixes, and auto-captions contain mistakes you should check. I tested it on a small sample and a self-test, not yet on hours of real recordings.
The skill is MIT licensed in the hermes-skills library.
Skill (MIT): https://github.com/RobinBeraud/hermes-skills/tree/main/media/transcript-to-chapters
All skills: https://github.com/RobinBeraud/hermes-skills
This post was written with AI assistance and reviewed through automated checks.
Top comments (0)