Viral Topic

Use this skill as the control layer for finding topic references across platforms. It does not fetch data itself; it routes the request to the platform skill whose collection logic matches the platform.

Package Layout

This repository publishes viral-topic/ as one skill package:

viral-topic/
├── SKILL.md
├── references/platforms.md
├── wechat-viral-topic/
├── x-viral-topic/
├── bilibili-viral-topic/
└── youtube-viral-topic/

The root SKILL.md is the controller. The platform directories are child skills/adapters with their own SKILL.md, references/, agents/, and scripts.

Platform Routing

Read references/platforms.md before deciding which child skills to run.

Implemented child skills:

• $wechat-viral-topic: WeChat public-account articles that exceed the account's average reads.
• $x-viral-topic: X/Twitter posts from low-follower authors that outperform their audience size.
• $bilibili-viral-topic: B站 videos from low-follower UP主 that outperform their follower base.
• $youtube-viral-topic: recent high-performing YouTube videos without low-subscriber filtering.

If the user gives no platform, default to the implemented set: WeChat, X, Bilibili, and YouTube. If the user asks for 小红书 or 抖音, explain that those adapters are not implemented yet and propose the available API/browser-automation path before building it.

Workflow

1. Normalize the user's niche into 3-8 focused search terms.
2. Choose platforms from the request. For "全平台" or "各个平台", run all implemented child skills.
3. Run each platform skill independently with its own defaults:
   • WeChat: --days 7, --min-read 10000, --min-read-month-avg-ratio 2
   • X: --days 7, --max-followers 50000, --min-engagement 100
   • Bilibili: --days 30, --max-followers 100000, --min-play 10000
   • YouTube: --days 30, --min-views 10000
4. Preserve source URLs, author/account metrics, timing, and evidence labels.
5. Merge outputs into a cross-platform topic table only after preserving each platform's original evidence.
6. Label confidence carefully:
   • WeChat confirms average-read breakout, not low-follower breakout.
   • X and Bilibili can confirm low-follower viral only when follower counts are present.
   • YouTube confirms viral strength, not low-subscriber breakout.

Output Shape

When combining platforms, use this compact table first:

platform | title | author | published_at | main_metric | scale_metric | breakout_reason | url

Then add a short synthesis:

• reusable topic angle
• title/hook pattern
• format pattern
• why it likely worked
• whether the evidence is confirmed or proxy-only

Guardrails

• Do not collapse all platforms into one scoring formula without keeping platform-specific evidence.
• Do not call YouTube results low-fan by default.
• Do not use WeChat follower_count as the default low-fan signal; compare article reads with month_read_avg.
• Never write API keys, cookies, app secrets, or access tokens into skill files or outputs.
• Use results as topic references and structure inspiration. Do not copy posts, articles, video scripts, captions, or thumbnails wholesale.