A sophisticated AI-powered skill for analyzing, summarizing, and querying local and remote video content using Gemini or Moonshot AI.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install video-understand
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install video-understand using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Video Understand?
Video Understand is a comprehensive tool within the library of Openclaw Skills designed to give AI agents the ability to perceive and interpret video data. It enables users to process local files, YouTube URLs, and direct HTTP video links to extract deep insights, generate summaries, or perform complex Q&A sessions. By leveraging state-of-the-art models like Google Gemini and Moonshot AI (Kimi), it bridges the gap between raw video files and actionable text-based intelligence.
The skill is engineered for efficiency, featuring local content hash caching to prevent redundant uploads and native URL handling for supported providers. Whether you are managing a library of screen recordings or need to parse hours of YouTube content, this skill provides a structured, developer-friendly interface to automate video comprehension tasks.
Video Understand Use Cases
- Generating automated, timestamped summaries of long-form webinars or tutorials.
- Extracting specific information or data points from recorded meetings and presentations.
- Performing multi-turn Q&A on video content to clarify specific visual details.
- Building automated metadata for video libraries by generating descriptions and identifying key moments.
- Analyzing remote video content from YouTube or HTTP servers without manual downloads.
How Video Understand Works
- The user initiates an analysis by providing a local file path or a remote URL (YouTube/HTTP).
- For local files, the skill generates a content hash to check against the local upload cache, ensuring files are only uploaded once.
- If using Google Gemini, remote URLs are passed natively to the API; if using Kimi, the skill handles the download via yt-dlp or fetch before uploading.
- The chosen AI model processes the video stream and the user's prompt to generate a detailed response.
- The output is returned as formatted Markdown or structured JSON, including optional timestamps for chronological reference.
- Subsequent questions can be asked about the same video using the file reference, avoiding the need for repeated processing.
Video Understand Setup
Check if the skill is installed and verify your version:
video-understand --version
Configure your preferred AI provider and authenticate with your API key:
video-understand config set-provider gemini
# or
video-understand config set-provider kimi
If using Kimi with YouTube links, ensure yt-dlp is installed on your system:
# macOS
brew install yt-dlp
# Windows
winget install yt-dlp
Video Understand Data Schema & Taxonomy
Video Understand manages configuration and caching through a specific directory structure and metadata format:
| Path | Description |
|---|---|
~/.video-understand/config.json |
Global settings, including default provider and API keys. |
~/.video-understand/uploads.json |
Cache of uploaded file references and content hashes for deduplication. |
.video-understand/ |
Local directory used for storing analysis output files when requested. |
Supported File Formats: MP4, MPEG, MOV, AVI, FLV, MPG, WebM, WMV, 3GPP, and MKV.
Video Understand Advanced Features
- Content hash caching to skip re-uploading identical local files across different sessions.
- Timestamped breakdowns that correlate AI insights with specific video markers.
- Multi-provider support allowing users to toggle between Gemini and Moonshot AI models.
- Programmatic JSON output mode for seamless integration into larger automation pipelines.
- Stateless follow-up capabilities using the ask command for previously indexed video content.
- Environment variable support for API keys, enabling secure usage in CI/CD environments.
SKILL.md
Loading
Related Openclaw Skills
wacli
A command-line utility for sending WhatsApp messages, searching chat history, and syncing local archives.
Self-Improving Agent (With Self-Reflection)
An autonomous learning framework that enables AI agents to evaluate their work, incorporate user corrections, and build a permanent knowledge base through self-reflection.
Save Money
An intelligent behavioral skill that automatically routes tasks between high-performance and low-cost AI models to slash API expenses by over 50%.
QMD Memory Skill
A local hybrid search engine that eliminates API costs by running embeddings and semantic queries directly on your machine.
Signalgrid Live Activities
Integrate real-time mobile push notifications and live activity tracking into your AI workflows via Signalgrid.
DECK-0
Enable AI agents to interact with the DECK-0 digital collectibles platform to manage, buy, and open trading card packs on-chain.
Featured*
