Skip to content
Back
  • Skill
  • Claude Code
  • ElevenLabs
  • Refining

Product demo videos: a Claude Code skill for narrated walkthroughs

Makes short, voiced walkthrough videos of a web app. The pointer moves in time with the narration, and the finished video comes with zooms, music, captions and a branded intro and outro, with no editing afterwards.

View on GitHub
A 40-second walkthrough of a sample app, made start to finish by the skill: script, voice, recording, zooms, music, captions, intro and outro. Nothing was edited by hand.
Up close
The pointer moves in a slight arc and hovers the other choices before picking the one the voice names, right as it’s said.

What it makes

Short product walkthrough videos, thirty seconds to two minutes, with a voice that talks the viewer through every click. You describe the flow. The agent writes the narration, records the app while a pointer moves through it in time with the words, then adds zooms, a branded intro and outro, music and captions. The video above came straight out of it, with no editing by hand.

How it works

Every video is a list of beats: one spoken line plus the on-screen actions that go with it.

Script

The agent scouts the app first, so the narration uses the exact labels on screen.

Voice

ElevenLabs reads each line and returns word-by-word timings. Only changed lines get regenerated.

Take

A hidden browser records the app at twice the output resolution while the pointer glides, hovers and clicks. Each beat lasts as long as its line, and a click can wait for the exact word that names it.

Assemble

The voice lands where each beat started, the frame eases in where the take asked for a zoom, and the intro, outro, music and captions go on in one pass. Music ducks under the voice on its own.

Check

The agent pulls a frame from every beat and looks at each one before calling the video done.

Change a line, regenerate that clip, record again. Nothing gets edited by hand.

What you need

Node and ffmpeg

The setup script installs Playwright and its browser.

An ElevenLabs API key

Ideally restricted to Text to Speech, Voices and Music Generation. A free system voice works for rehearsing. For production quality, we recommend their v4 model; the skill’s defaults are a good starting point.

A demo account

Never record real customer data.

Any desktop OS

Recording happens in a hidden browser, so it never takes over your mouse. We use it on macOS, and it's built to run on Linux and Windows too. A macOS-only screen mode covers things outside the page, like native menus or a terminal.

The first time, the agent runs a setup check that names anything missing along with the command that fixes it. Then it makes a short test video from a bundled sample app, so you know the machine is ready before you record your own.

Example prompts

"Make a getting-started video that shows signing up and creating a first project."

"Record a one-minute walkthrough of inviting a teammate, in the same voice as the last one, with our logo on the intro."

"The pricing changed. Update the third line of video two and record it again."

Install

Use Hand to your agent above and paste the result into Claude Code or another coding agent. To install it by hand, copy the product-demo-videos folder from the strideUX skills repo into ~/.claude/skills/.

Ready to try it? Hand it to your agent

Work with us

Tell us what you’re building. We read every message and reply shortly.

We only use this to reply. Privacy policy