A powerful cross-platform tool to transform video URLs or local files into searchable, summarized local knowledge using Whisper and yt-dlp.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install video-knowledge-ingest
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-knowledge-ingest 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 Knowledge Ingest?
The Video Knowledge Ingest skill is a comprehensive workflow designed for AI agents to process video content from diverse platforms like YouTube, Bilibili, and Xiaohongshu. By integrating robust media downloaders and local AI transcription models, this Openclaw Skills entry automates the transition from raw video URLs to structured, summarized markdown notes.
This skill prioritizes efficiency by attempting to fetch native subtitles first, falling back to high-fidelity Whisper transcription via GPU or CPU when necessary. It is an essential component for building a local second brain, ensuring that video-based information is persisted, indexed, and easily retrievable within a private workspace.
Video Knowledge Ingest Use Cases
- Creating a searchable local archive of educational YouTube or Bilibili video transcripts.
- Summarizing content from Xiaohongshu where native subtitles are frequently unavailable.
- Converting local MP4 or MKV media files into text-based summaries for research.
- Debugging video ingestion failures like 403 Forbidden errors or anti-bot blocks on major video platforms.
- Automating the population of a local knowledge base for sub-agents to reference.
How Video Knowledge Ingest Works
- Source Normalization: The skill cleans the input URL, stripping tracking parameters while preserving essential timing metadata.
- Subtitle Retrieval: It attempts to fetch existing subtitles (preferring English and Chinese) using yt-dlp as a best-effort first step.
- Media Fallback: If subtitles are missing, the skill downloads the audio stream and utilizes ffmpeg for normalization.
- AI Transcription: The bundled Whisper-GPU script transcribes the audio, falling back to CPU mode if hardware acceleration is unavailable.
- Summary Generation: The resulting transcript is processed through a codex-based summarizer to create a concise markdown digest.
- Data Persistence: All artifacts, including metadata, transcripts, and summaries, are saved to the local knowledge base and indexed.
Video Knowledge Ingest Setup
To use this within the Openclaw Skills ecosystem, ensure you have yt-dlp, ffmpeg, and a configured Whisper environment. Use the following command to start ingesting:
# Ingest a YouTube video
skills/video-knowledge-ingest/scripts/video-ingest.sh "https://www.youtube.com/watch?v=EXAMPLE"
# Ingest a local media file
skills/video-knowledge-ingest/scripts/video-ingest.sh /path/to/video.mp4
Video Knowledge Ingest Data Schema & Taxonomy
The skill organizes data within the /knowledge/video-notes/ directory using the following structure:
| File | Description |
|---|---|
source.info.json |
Original video metadata (title, uploader, etc.) |
transcript.txt |
The full raw text of the video content |
summary.md |
AI-generated summary of the transcript |
record.json |
Processing logs and execution metadata |
index.jsonl |
A global append-only index for cross-video search |
Video Knowledge Ingest Advanced Features
- Automatic GPU-to-CPU fallback for Whisper transcription to ensure workflow continuity.
- Smart URL normalization specifically tuned for Bilibili and YouTube tracking bypass.
- Support for local
.srtor.vttfile injection to skip the download phase entirely. - Multi-platform error handling for common issues like YouTube 429 rate limits and Bilibili 403 errors.
- Integration with
summarize --cli codexfor high-quality, developer-focused content synthesis.
SKILL.md
Loading
Related Openclaw Skills
Novel
A lightning-fast terminal utility for managing novel metadata, including chapters, characters, and plot timelines without leaving the command line.
Frontend Performance Audit
An AI-driven performance analyzer that generates structured optimization reports for web applications based on Lighthouse and bundle data.
Newsticker (claw://News)
A specialized newsdesk skill connecting OpenClaw agents to German-language AI, security, and ecosystem updates from ClawNews.de.
ClawJobs
ClawJobs is a collaboration plugin for OpenClaw that enables multiple agent peers to share tasks and synchronize status through a centralized hub.
Porteden Docs
A secure CLI-driven tool for comprehensive Google Docs lifecycle management and content manipulation optimized for AI agents.
Project Context Manager
A project-based context management system that enforces structured documentation and persistent memory for AI coding agents.
Featured*
