skills.rest
openakita
openakitaCOMMUNITY
Explore all skills by this author →
GitHub

idea-research

Pulls trending videos, breaks down viral content, and remixes scripts across six social platforms.

2.0k|274|Updated Jan 30, 2026
Search Tags:#content analysis#Bilibili#trend research#Douyin#Social Media#viral breakdown#script remix
One-click install
npx skills add https://github.com/openakita/openakita --skill idea-research
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: idea-research
Source: https://github.com/openakita/openakita/tree/main/plugins/idea-research
Command: npx skills add https://github.com/openakita/openakita --skill idea-research

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires httpx, pydantic, yt-dlp, ffmpeg, playwright, cryptography, keyring, faster-whisper.

What problem does it solve?

Content creators and marketers struggle to systematically discover trending topics, understand why viral videos succeed, and adapt winning formats to their own brand voice across platforms like Bilibili, YouTube, Douyin, Xiaohongshu, Kuaishou, and Weibo.

Core Features & Use Cases

  • Trend Radar (radar_pull): Concurrently pulls hot content from up to 5 platforms with keyword filtering, time windows (24h/7d/30d), and engagement-based scoring, using safe public APIs by default or an optional cookie-based browser crawler.
  • Viral Breakdown (breakdown_url): A 7-step pipeline that downloads a video via yt-dlp, extracts keyframes, describes them with Qwen-VL, transcribes audio, clusters comments, and produces a structured hook analysis report.
  • Account Comparison & Script Remix: Compares hook patterns and posting cadence across competitor accounts, then rewrites any trend item into 1-5 branded script variants using 12 built-in personas.
  • MDRM Memory Graph: Learns successful hook patterns over time via dual-track writes to vector store and memory manager, so recommendations improve with use.
  • Use Case: A Xiaohongshu operator pastes a competitor's viral video URL, gets a full structural breakdown of its hooks and comments, then generates three 60-second script variants for their own skincare brand.

Quick Start

Ask the assistant to pull today's trending AI videos from Bilibili and YouTube, then paste any video URL to receive a full viral breakdown with hook analysis and script suggestions.

Frequently Asked Questions about idea-research

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I pull trending videos from Bilibili and Douyin?▼

Use the radar_pull mode with a list of platforms, keywords, and a time window such as 24h, 7d, or 30d. Bilibili and YouTube use safe public APIs by default, while Douyin, Xiaohongshu, Kuaishou, and Weibo can use RSS feeds or an optional cookie-based browser crawler.

How to analyze why a viral video succeeded?▼

Use the breakdown_url mode with the video URL. It runs a pipeline that downloads the media with yt-dlp, extracts keyframes, describes frames with Qwen-VL, transcribes audio, clusters comments, and outputs a structured hook analysis with persona takeaways.

What is the difference between Engine A and Engine B?▼

Engine A uses official APIs and RSS feeds with zero risk but limited interaction data. Engine B is an optional Playwright browser crawler requiring user-supplied cookies; it accesses login-gated data but may violate platform terms and trigger anti-bot controls.

Does the trend radar work without cookies or API keys?▼

Yes, Engine A works out of the box for Bilibili and via RSS Hub for Douyin, Xiaohongshu, and Weibo. A YouTube Data API key improves YouTube results, and DashScope keys are only needed for ASR and VLM media analysis.

Why did my crawler cookies stop working?▼

Cookies expire or get invalidated when platforms flag automated access, producing cookies_expired or crawler_blocked errors. Re-export cookies from your browser into Settings, test connectivity, or switch back to Engine A if the account is rate-limited.

What are the limits of video breakdown and crawling?▼

Engine B caps single fetches at 30-100 items per platform, local ASR handles videos up to 10 minutes before falling back to cloud Paraformer, and VLM frame description runs at 5 concurrent requests subject to DashScope QPS limits.

Related Skills

Explore More Content & Communication Skills→
reomgitreomgit

setup-browser-cookies

Import cookies from a Chromium browser into a headless browser session.

Community
Intermediate
espressifespressif
2.1k

scheduled_task

Creates cron and interval scheduled tasks for agent wake-ups, IM messages, and Lua script runs.

Official
Intermediate
different-aidifferent-ai
249

video-subtitle-cutter

Transcribe video speech, detect filler words and pauses, and generate FFmpeg cut commands with SRT subtitles.

Official
Advanced
jguadagnojguadagno
18

jjgnet-broadcast

Monitor RSS feeds and video streams to post updates to Twitter and Facebook.

Community
Intermediate