Analyze video containers, durations, resolutions, audio channels, and extracted keyframes using bradautomates/claude-video patterns (commit 03ceb42) and local ffprobe/ffmpeg, with explicit WhisperX boundary reporting. Use when inspecting rendered or supplied video files with grounded timestamps.
8 stars
0 votes
0 copies
0 views
Added November 18, 2026
ai-agentsgonodegitapisecurity
Works with
cli
api
Security analysis
A92/100
mediumInstalls packages at runtime which could introduce malicious dependencies
Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of Rei Video Understanding?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/rootcastleco-rei-video-understanding)
---
name: rei-video-understanding
description: "Analyze video containers, durations, resolutions, audio channels, and extracted keyframes using bradautomates/claude-video patterns (commit 03ceb42) and local ffprobe/ffmpeg, with explicit WhisperX boundary reporting. Use when inspecting rendered or supplied video files with grounded timestamps."
category: data-ai
risk: safe
source: rootcastle-rei
license: MIT
metadata:
tags:
- video
- claude-video
- ffprobe
- ffmpeg
- keyframes
- analysis
---
# REI Video Understanding (`claude-video` Adapter)
Extract verifiable stream metadata, keyframes, and timestamped scene observations from video files using [`lib/multimodal-providers.js`](../../lib/multimodal-providers.js) (`runVideoUnderstandingStage`).
## When to Use
- Auditing rendered videos (`final.mp4`) or supplied engineering clips (`.mp4`, `.mov`, `.mkv`, `.webm`, `.m4v`, `.avi`).
- Checking real duration in seconds, pixel dimensions (`width x height`), video/audio codecs, and extracting keyframes at grounded timestamps.
## Inputs and Outputs
- **Inputs:** Validated workspace/temp video path (`validateFileAccessBoundary` enforced; supported extensions only).
- **Outputs:** JSON inspection report containing `ffprobe` container/stream metrics, `extractedFrameCount`, timestamped `segments[]`, SHA-256 digest, and explicit `transcriptStatus` / uncertainty disclosures.
## Prerequisites & License Constraints
- Upstream: `bradautomates/claude-video` (`MIT`, commit `03ceb42f7fa2c4439aca01752118044baabffb8f`).
- Requires local `ffprobe` and `ffmpeg`. Optional `whisperx` (3.8.6, ~3 GB free disk) is never installed automatically during `npm install`.
## Offline vs Online Capabilities
- **Offline:** Full container inspection and FFmpeg keyframe extraction operate locally with zero network calls or API keys.
- **Online / Extended:** Cloud transcription APIs or remote `yt-dlp` downloads remain opt-in and disabled by default for security and cost control.
## Procedure
1. Validate path confinement and reject unsupported extensions or traversal attempts (`../`, credential paths).
2. Run `ffprobe -show_format -show_streams` and extract representative 1 fps keyframes with `ffmpeg`.
3. Attach exact timestamps and report `transcriptStatus: NOT_VERIFIED` when WhisperX is absent.
```sh
ffprobe -v quiet -print_format json -show_format -show_streams video/final.mp4
node bin/install.js pipeline validate --output ./examples/multimodal-rotor-demo --json
```
## Do not use
Constraints and safety: Operational boundary — do not claim semantic ASR speech transcription when WhisperX is missing, do not fabricate video timestamps, and never download copyrighted external streams without rights.
## Verification Criteria & Failure Recovery
- Verify `metadata.ok === true`, `durationSec > 0`, valid `width x height` integers, and non-empty `segments[]`.
- Clean temporary extracted `.jpg` frames immediately after QA audit.
## Authoritative product references
- https://github.com/bradautomates/claude-video