- Home
- Skills
- Testing & Debugging
- Narrated Evidence
Narrated Evidence
Record narrated screen videos (voice + subtitles) of bugs and features from a JSON script, using the free open-source narrate CLI and Playwright.
Free
Works with the AI Agents you already use
Narrated Evidence
Example session with this skill installed
Record a narrated video of the passengers form bug in our demo app: it says "Passenger saved" even when the server returns 500. Use the bug template and a Spanish voice.
- Read your context and instructions
- Compiled the narrated evidence
Done. I wrote the script from the bug template, validated it and recorded it with the headless Playwright recorder.
Result
| Item | Value |
|---|---|
| Video | out/narration/narration.mp4 (49.8 s, H.264 + AAC + soft subtitles) |
| Subtitles | out/narration/narration.srt |
| Voice | edge-tts es-CO-SalomeNeural, no fallback needed |
| Recorder | playwright (headless Chrome) |
Scenes
| Scene | What is shown |
|---|---|
| contexto | Chapter card: "Bug: false save in Passengers" |
| pasos | Name and phone filled, Save clicked |
| espera | Backend wait, sped up x4 |
| actual | Red callout: shows "Passenger saved" |
| esperado | Green callout: should show an error |
| evidencia | POST /api/passengers -> 500 from evidence.json |
| impacto | Chapter card: users believe data was saved |
Evidence captured
- 1 failed response:
POST /api/passengersreturned 500. - 2 console errors: the 500 resource error and
POST /api/passengers failed with 500.
I checked frames at each scene start: the screen matches what the voice says.
Next steps
- Watch the video once before attaching it.
- Attach the MP4 and SRT to the bug report or PR.
Connects securely to your tools. The creator never sees your data.
About this skill
Screenshots cannot show a sequence, a timing problem or a state that changes. This skill makes your agent produce a narrated video instead: it writes a short script, records the browser and delivers an MP4 with voice and subtitles that you can attach to a bug report or a pull request.
What it does
- Writes the script for you: one scene per idea, each with the sentence to narrate and the browser actions (go to, click, fill, wait).
- Keeps voice and screen in sync: the voice is generated first and every scene waits until its narration ends, so the voice never runs ahead of the screen.
- Bug and feature templates: context, steps, actual result (red callout), expected result (green callout), evidence and impact.
- Speeds up slow waits (backend calls, loading) without touching the narration.
- Collects evidence: failed network responses and console errors from the run, ready to quote in the video.
- Delivers an H.264 MP4 with AAC audio and soft subtitles, plus an .srt file.
How it works
- The agent checks the requirements and asks before installing anything.
- It installs the open-source CLI:
npm install -g github:jeider05/ai-video-evidence. - It writes
narration.jsonfrom the matching template and validates it. - It runs
narrate run narration.json: free neural voices (edge-tts), headless Chrome recording, ffmpeg assembly. - It checks frames and audio, then hands you the video and a scene timeline.
Requirements
An agent with a local terminal (Claude Code, Codex CLI, Cursor, Gemini CLI). Node.js 22+, ffmpeg and ffprobe, Google Chrome and Python with edge-tts. No paid API is needed; Deepgram is an optional paid fallback voice.
Safety
- The agent never types passwords: for apps behind a login you sign in yourself with
narrate login <url>. - Test data only, with a
QA_prefix; existing records are never edited or deleted. - Outputs and session files stay out of git.
Known limitations
- Headless recording cannot show native
<select>popups or OS dialogs. - The experimental
gdigrabrecorder (Windows + ego-browser) is for those cases and needs a visible window. - Web-only chats cannot run it: the CLI runs on your machine.
Source code (MIT): https://github.com/jeider05/ai-video-evidence
Needs an agent with a local terminal, Node.js 22+, ffmpeg/ffprobe, Google Chrome and Python with edge-tts. Headless recording cannot show native <select> popups or OS dialogs; the experimental gdigrab recorder (Windows + ego-browser) covers them. The agent never types passwords: you sign in with narrate login.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
Requires
Permissions
Terminal to run the narrate CLI and ffmpeg; files to write the script and the video; browser to record the app; network for edge-tts voices and the GitHub install; environment variables only for the optional Deepgram key.