You wrote a script for a Reel, a TikTok or a Short. It reads well. Then you film it and the first sentence takes six seconds to say, and the whole thing runs a minute over. Re-shooting is the expensive way to find that out.
I made a small Hermes skill, short-video-script, that handles the part before the camera. You give your agent a topic, the platform and the one thing you want viewers to do at the end. It writes the script in a fixed format: a HOOK, a few BEATs, a TURN (where the promise is kept) and a single CTA. Every line also carries a visual note in brackets, so you know what changes on screen.
Then a tiny Python file estimates how long each line takes to say, flags a hook that is too long or a missing call to action, and splits the words into short caption chunks you can paste into your editor.
Here is a real run on a five-line script:
00:00.00-00:03.60 HOOK 9 words 3.6s 00:03.60-00:07.60 BEAT 10 words 4.0s ... TOTAL: 20.4s (target 45s) hook takes about 3.6s to say, limit is 3.0s (9 words) The hook is only a little over my 3-second limit, so I shorten it before filming. That is the whole point.
Some honest notes. The timing is an estimate from word count at a fixed pace, so record one take and adjust the pace setting. The skill never makes up numbers or testimonials: if a figure would help, it leaves a placeholder for you. And if the video is a paid partnership, it reminds you to label it.
Skill and instructions are in the hermes-skills repository, MIT licensed.
Skill (MIT): https://github.com/RobinBeraud/hermes-skills/tree/main/content-creation/short-video-script
All skills: https://github.com/RobinBeraud/hermes-skills
This post was written with AI assistance and reviewed through automated checks.
