Skip to content

Quick Start

Three steps to analyze your first video.

1. Install

/plugin marketplace add mnvsk97/eyeroll
/plugin install eyeroll@mnvsk97-eyeroll
pip install eyeroll[gemini]

2. Initialize

/eyeroll:init

This configures your API key and generates codebase context. Takes about 30 seconds.

3. Watch a video

/eyeroll:watch https://loom.com/share/abc123

Or with a local file:

/eyeroll:watch ./bug-recording.mp4

Example flow

You: /eyeroll:watch https://loom.com/share/abc123 --context "checkout page broken after PR #432"

eyeroll: Acquiring: https://loom.com/share/abc123
         Media type: video
         Duration: 00:45
         Strategy: frame-by-frame
         Extracted 12 key frames
         Analyzing frame 0 @ 00:00...
         Analyzing frame 1 @ 00:04...
         ...
         Synthesizing report...

Agent:   Here's what the video shows:

         **What's happening**: The user navigates to /checkout, fills in payment
         details, clicks "Pay", and gets a 500 error. The error toast reads
         "Internal Server Error".

         **The bug**: POST /api/checkout returns 500. The console shows
         "TypeError: Cannot read properties of undefined (reading 'stripe_id')".

         **Key evidence**:
         - Error: "TypeError: Cannot read properties of undefined (reading 'stripe_id')"
         - Route: /api/checkout (POST)
         - The error appears after clicking the "Pay" button

         **Suggested fix**: Search for `stripe_id` in the checkout handler.
         The user object may not have a Stripe customer ID set.

         Want me to fix this?

Adding context

Context text helps eyeroll understand what you want. Most screen recordings are silent, so the context is often more informative than the video itself.

/eyeroll:watch ./demo.mp4 --context "create a skill from this tutorial"
/eyeroll:watch ./bug.mp4 --context "broken after deploying the auth changes"

What's next