界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

bilibili-viral-topic

bilibili-viral-topic

【效率与协作】Find low-follower viral Bilibili videos for account-growth topic selection. Use when the user asks for B站低粉爆款, Bilibili low-fan hits, 起号选题 from B站, niche Bilibili references, or videos ranked by UP主粉丝数, 播放, 点赞, 投币, 收藏, 分享, 评论, 弹幕, and 播放粉丝比.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: bilibili-viral-topic description: Find low-follower viral Bilibili videos for account-growth topic selection. Use when the user asks for B站低粉爆款, Bilibili low-fan hits, 起号选题 from B站, niche Bilibili references, or videos ranked by UP主粉丝数, 播放, 点赞, 投币, 收藏, 分享, 评论, 弹幕, and 播放粉丝比.

Bilibili Viral Topic

Use this skill to find B站 videos that significantly outperform the UP主's follower base.

Quick Start

Run the bundled script from this skill directory:

python3 scripts/search_bilibili_viral_topic.py --topic "AI工具,大模型,AI编程" --days 30 --max-followers 100000 --min-play 10000 --limit 50 --format markdown

No API key is required. The script uses public Bilibili JSON endpoints with browser-like headers, conservative rate limits, retries, and a local follower cache.

Optional environment:

export BILIBILI_COOKIE="..."

Only set BILIBILI_COOKIE if public requests are repeatedly rate-limited. Never write cookies into skill files or output artifacts.

Workflow

  1. Convert the user's niche into focused Chinese and product-name keywords.
  2. Search B站 videos through x/web-interface/search/type with order=pubdate and order=click.
  3. Optionally supplement with x/web-interface/ranking/v2 when the user asks for broader platform signals.
  4. Deduplicate by bvid.
  5. Filter to the requested recent window using pubdate.
  6. Enrich each unique mid through x/relation/stat?vmid=<mid> and cache follower counts.
  7. Filter:
    • follower_count <= --max-followers
    • view_count >= --min-play
    • view_count / follower_count >= --min-play-follower-ratio
  8. Rank by breakout_score, then play_follower_ratio, then view_count.
  9. Return JSON for automation or Markdown for topic review.

Output Contract

Each result must preserve source and UP主 evidence:

{
  "platform": "bilibili",
  "content_id": "",
  "url": "",
  "title": "",
  "author_id": "",
  "author_name": "",
  "follower_count": 0,
  "published_at": "",
  "view_count": 0,
  "like_count": 0,
  "comment_count": 0,
  "share_count": 0,
  "save_count": 0,
  "coin_count": 0,
  "danmaku_count": 0,
  "hot_score": 0,
  "breakout_score": 0,
  "evidence": []
}

Read references/scoring.md before changing thresholds or combining B站 results with other platforms. Read references/rate_limits.md before increasing pages, ranking rids, or request speed.

Guardrails

  • Do not scrape B站 HTML pages when the JSON endpoints are enough.
  • Use browser-like headers. Naked script user agents can trigger -352 or HTTP 412.
  • Keep request speed conservative. Default --request-sleep 0.6 is intentional.
  • Cache follower counts for at least 24 hours. Re-querying the same mid repeatedly increases risk and adds little value.
  • Do not call a video confirmed low-fan if follower count is missing unless the user explicitly allows unknown followers.
  • Use videos as topic signals, hook references, title structures, and format references. Do not copy full scripts or captions wholesale.
我的收藏