When should you use it?
- Your app changes often and recorded tutorials go out of date
- You support customers who prefer video to reading
- You already have a web app a coding agent can work in
Not for: Apps you can't run with safe fake data; never record real customer records
What do you need, and what do you get?
You need
- A web app the agent can run locally
- A list of tasks customers ask how to do
- A text-to-speech option and a video tool such as ffmpeg
You get
- A demo setting filled with invented records, so nothing real is shown or saved
- One scene script per tutorial: page, narration, cursor actions
- Rendered videos with captions, transcripts and a manifest
- A searchable help page, including search inside transcripts
How does it work, step by step?
- Pick the first five tutorialsStart with the questions your support team answers most.
- Make it safe to filmAsk the agent for a demo setting filled with invented records that saves nothing.
- Script the scenesUse our prompt so each video is defined in code and rebuilt automatically.
- Review and publishWatch every video once, fix narration, then let the build regenerate them after app changes.
The prompt
Our own prompt, inspired by Nadia I (@NIftek) and written to work in Claude Code, Custom agent and Grok Bot. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
I want how-to videos for our app that we can rebuild automatically after every release.
Start by adding a demo setting, filled with invented sample records, that never saves anything.
Write each tutorial as a script file listing the screens to visit, what the narrator says, and where the pointer moves. First tutorials: {tutorial_list}.
Build a pipeline that films each script in a headless browser, adds a synthetic voice and subtitles, and exports the clip with a text transcript.
Publish them on one searchable support page (searching the transcripts too), usable by keyboard. If a render fails, resume from the failed clip, and check each clip for blank or broken frames.Inspired by Nadia I (@NIftek): A Replit agent prompt for an auto-generated help video library. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Generated narration can describe a button that moved. Check each frame against the narration before publishing.
- Long builds fail halfway. Make rendering resumable so one broken scene doesn't restart everything.
Our take
Help videos are valuable and almost always out of date, because re-recording them after every interface change is nobody's favorite job. Nadia's approach treats them like code: each tutorial is a short script of pages, narration and cursor moves, recorded automatically against a demo mode with fake data, narrated, captioned and stitched together. That makes them cheap to regenerate and safe, since no real customer data appears on screen. Her prompt also asks for details that separate a demo from something maintainable, such as resumable builds, frame checks on every video, search across transcripts and keyboard accessibility. We'd add one human step, watching each new video once, because narration that confidently describes the wrong button confuses customers more than no video at all. In our dry run against a small sample app, our prompt produced a demo mode, 3 scene scripts, rendered 3 captioned videos with transcripts and a manifest, and the frame check flagged one scene where the cursor clicked before the page finished loading.
Where the time goes back: Updating tutorials after a release becomes a rebuild instead of re-recording, editing and captioning each video by hand.
FAQ
Why not just record my screen?
Is the narration a robot voice?
Source and credit
Inspired by Nadia I (@NIftek). We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- A Replit agent prompt for an auto-generated help video library
- Creator
- Nadia I (@NIftek)
- Published on
- X
- Original date
- Jul 9, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Custom agent
- AI dry run
- Oct 20, 2026: passed. Dry-run tested by our AI test agent on sample input. Videos regenerated from code; one timing issue caught.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.