Convert a Douyin, Bilibili, or Xiaohongshu video link into a locally transcribed, AI-analyzed, interactive HTML report.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install tri-platform-video-analysis
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 tri-platform-video-analysis 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 Tri-Platform Video Analysis?
Tri-Platform Video Analysis is an Openclaw Skills workflow for collecting and understanding videos from Douyin, Bilibili, and Xiaohongshu. Provide one video URL, and the skill downloads the media, transcribes speech offline with faster-whisper, returns structured transcript data, and enables the calling AI to produce a standardized content analysis.
The final deliverable is a clean, shareable HTML report rather than a raw transcript. It includes platform metadata, a one-sentence summary, source-grounded quotes, and four expandable insight capsules covering key points, content structure, editorial judgment, and highlights. Transcription and rendering run locally without an external LLM API key or API charges.
Tri-Platform Video Analysis Use Cases
- Analyze short-form videos from Douyin, Bilibili, or Xiaohongshu for research and competitive intelligence.
- Turn spoken content into structured notes, summaries, and reusable editorial insights.
- Evaluate hooks, pacing, narrative structure, credibility, positioning, and content opportunities.
- Create presentation-ready HTML reports for clients, teammates, or personal knowledge management.
- Extract verbatim quotes and memorable highlights from videos you are authorized to process.
- Process Chinese-language video content locally when transcript privacy and zero external LLM API fees matter.
- Review product reviews, workplace commentary, educational content, interviews, and creator videos from a single workflow.
How Tri-Platform Video Analysis Works
- Accept exactly one video URL explicitly provided by the user; the skill does not scan for or guess URLs.
- Inspect the target metadata and confirm the platform, title, and duration with the user before processing.
- Download the video from Douyin, Bilibili, or Xiaohongshu. Douyin prefers a direct iesdouyin API route, while Bilibili and Xiaohongshu use yt-dlp; cookies may be supplied when authentication is required.
- Transcribe the audio locally with faster-whisper. The default Whisper
smallmodel is downloaded on first use and supports CPU execution. - Parse the single-line JSON emitted on stdout by
scripts/parse.py; ignore stderr because it contains progress logs. The result includes metadata, transcript text, and paths to timed and plain transcript files. - Have the calling AI read the
transcriptfield and generate an analysis JSON object. The script performs collection, transcription, and rendering, but does not perform the semantic analysis itself. - Save the analysis object as
output/_analysis.json, preserving the required metadata, summary, quotes, and capsule structure. - Run the HTML generation command to render the analysis as an interactive card-based report.
- Present the generated HTML file at the returned
html_pathto the user, with no additional processing required.
Tri-Platform Video Analysis Setup
Install the required Python packages:
pip install yt-dlp faster-whisper
The first execution automatically downloads the Whisper small model from Hugging Face, approximately 500 MB. A GPU is not required; CPU execution is supported.
Collect and transcribe one video:
python scripts/parse.py "VIDEO_URL"
Read the JSON object from stdout, use its transcript value to create the required analysis JSON, and save that JSON to output/_analysis.json. Then generate the interactive report:
python scripts/parse.py --generate-html output/_analysis.json
Useful options include:
python scripts/parse.py "VIDEO_URL" --model small --language zh --out-dir output/
python scripts/parse.py "VIDEO_URL" --cookies cookies.txt
python scripts/parse.py "VIDEO_URL" --keep-video
Before processing, verify that you have permission to download and transcribe the content. Keep private output directories under your control, treat cookies as sensitive authentication material, and inspect HTML files from untrusted sources before opening them.
Tri-Platform Video Analysis Data Schema & Taxonomy
Collection output
The collection command emits one JSON object on stdout. Its primary fields are:
| Field | Description |
|---|---|
status |
Processing state, such as transcribed |
title |
Detected video title |
platform |
Douyin, Bilibili, or Xiaohongshu |
duration |
Video duration in seconds |
timed_path |
Path to the timestamped .txt transcript |
plain_path |
Path to the continuous transcript .txt file |
transcript |
Full recognized subtitle text used for analysis |
Generated files
All artifacts are written to output/ by default:
- Timed transcript:
.txtfile containing timestamp-aware transcription. - Plain transcript:
.txtfile containing continuous readable text. - Analysis JSON:
output/_analysis.json. - HTML report: platform- and title-based filename ending in
-报告.htmlor an equivalent generated report name. - Temporary video: retained only when
--keep-videois provided; otherwise deleted after transcription.
Analysis JSON taxonomy
| Section | Required content |
|---|---|
meta |
platform, title, duration, optional author, category, and date |
summary |
One-sentence conclusion with 2–5 key concepts wrapped in <em> tags |
quotes |
Verbatim subtitle excerpts; quotes must not be invented |
capsules |
Exactly four primary cards: points, structure, judgment, and highlights |
Capsule content follows these conventions:
points: 3–5 numbered insight items with titles and explanatory bodies.structure: 4–6 labeled elements such as opening hook, development, pacing, and takeaway.judgment: exactly three items covering creator perspective, credibility, and practical reusability.highlights: notable moments, rankings, or memorable content points.- Supported capsule types also include
contrast, although the standard report uses four fixed capsules. - Each capsule can include
id,icon,title,subtitle,detail_title,type, andcontentfields.
The HTML presentation contains a hero area, platform badge, title, duration, category and date, a highlighted summary card, quote cards, four expandable insight cards, and a footer. It intentionally excludes the full timed or continuous transcript.
Tri-Platform Video Analysis Advanced Features
- Fully local transcription with faster-whisper, reducing dependence on third-party transcription services and avoiding external LLM API keys for semantic analysis.
- Automatic model management through Hugging Face, with configurable Whisper model size such as
smallorlarge-v3. - Platform-aware acquisition: direct iesdouyin API access is preferred for Douyin, with yt-dlp fallback; Bilibili and Xiaohongshu use yt-dlp.
- Netscape-format cookie support for authenticated or restricted Bilibili and Xiaohongshu downloads; Douyin commonly works without cookies through the preferred API route.
- Configurable language, output directory, model, and video-retention behavior through CLI flags.
- Source-grounded quote extraction that requires every quote to come directly from the transcript.
- Structured AI handoff through a stable analysis JSON schema, making the workflow compatible with different calling agents or models.
- Interactive HTML delivery with expandable capsule panels, screenshot-friendly quote cards, and a consistent report layout.
- Privacy-conscious temporary-file handling: downloaded videos are deleted after transcription by default, while user-provided cookies are used only for platform authentication and are not uploaded.
- Explicit safety boundaries requiring user-provided URLs, pre-processing confirmation, permission to process copyrighted material, and caution around personally identifiable information.
SKILL.md
Loading
Related Openclaw Skills
A-Share Financial Literacy Q&A
An AI-powered A-share financial literacy assistant that explains market concepts, personalizes learning, analyzes risk, and supports simulation-based investment research.
doc-consistency
An offline document QA skill that finds internal consistency errors in long DOCX, Markdown, and TXT files and produces a repeatable re-check command.
Gitcrawl
Gitcrawl provides a freshness-aware GitHub archive workflow for finding issues, triaging pull requests, and identifying duplicate discussion clusters safely.
Frontend UI Engineer
Frontend UI Engineer automates production-quality, accessible, responsive UI development and related engineering workflows with structured outputs and built-in error recovery.
Financial Literacy Q&A
An AI-powered financial literacy and A-share analysis skill for answering questions, personalizing education, evaluating risk, and supporting investment workflows.
DEX Token Analysis
An AI-assisted DEX token analysis skill for holder clustering, advanced trader tracking, liquidity-pool research, and structured crypto market insights.
Featured*
