SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsbggg-tiktok-search

bggg-tiktok-search

General & other

View repository

Updated 10-15 20:06

About this skill

BGGG TikTok Search

Goal

Use the user's local real Chrome TikTok login state to complete content research, creator search, video sampling, profile collection, and screenshot evidence. By default, only read public or user-visible pages; do not like, follow, comment, send direct messages, publish, or modify account settings.

Prerequisites

Choose one control path:

  1. WebBridge: local WebBridge daemon and browser extension are available.
  2. Chrome/CDP: the user starts Chrome with remote debugging themselves, and the script connects via Playwright.
  3. Apple Events: after enabling View > Developer > Allow JavaScript from Apple Events on macOS Chrome, use scripts/tk_real_chrome.py。
  4. Manual assistance: when login, CAPTCHA, region pop-up, or risk control appears, have the user handle it in real Chrome and then continue.

WebBridge health check:

WEBBRIDGE_URL="${WEBBRIDGE_URL:-http://127.0.0.1:10086}"
curl -s "$WEBBRIDGE_URL/status"

CDP health check:

python3 <skill-dir>/scripts/tk_research.py check-cdp

Workflow

  1. Determine task type：
    • Keyword video search → search
    • Keyword creator search → authors
    • Existing creator profile → creator
    • The user has already navigated Chrome to the target page → current
  2. Operate Chrome：
    • navigate Open the TikTok search page/creator page.
    • fill Enter the search term,click trigger the search.
    • snapshot Read the page content and locate video cards.
    • evaluate Extract structured data (URL, author, title, engagement counts).
    • Take screenshots for evidence, and avoid pasting base64 screenshots directly into the context.
    • Scroll to load more and repeat extraction.
  3. Save results：
    • Write structured data to collected_items.json。
    • Save screenshots to screenshots/。
    • Generate research_notes.md for business distillation.
  4. Follow-up handoff：
    • To download videos → hand off to bggg-tiktok-downloader
    • To analyze videos → hand off to bggg-tiktok-readvideo

WebBridge operation examples

1. Open TikTok and search for keywords

# 打开 TikTok 首页（新标签页，session 隔离）
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -H 'Content-Type: application/json' \
  -d '{"action":"navigate","args":{"url":"https://www.tiktok.com","newTab":true,"group_title":"TikTok Research"},"session":"tiktok-search"}'

# 获取页面 snapshot，找到搜索框的 @e ref
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"snapshot","session":"tiktok-search"}'

# 在搜索框填入关键词（用 @e ref 或 CSS selector）
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"fill","args":{"selector":"input[type=search]","value":"skincare routine"},"session":"tiktok-search"}'

# 点击搜索按钮或按回车
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"click","args":{"selector":"button[type=submit]"},"session":"tiktok-search"}'

2. Extract video card data

# 用 evaluate 提取视频列表的结构化数据
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"evaluate","args":{"code":"(() => { const cards = [...document.querySelectorAll(\"[data-e2e=search-card-video]\")].slice(0, 10); return cards.map((c, i) => { const link = c.querySelector(\"a\"); const author = c.querySelector(\"[data-e2e=search-card-user-avatar]\"); const metrics = [...c.querySelectorAll(\"[data-e2e=video-views]\")]; return { index: i, url: link ? link.href : null, title: c.innerText.slice(0, 200), author: author ? author.getAttribute(\"href\") : null, raw_text: c.innerText.slice(0, 500) }; }); })()"},"session":"tiktok-search"}'

TikTok DOM changes frequently; the selectors above are only examples. In actual use, first use snapshot to read the current page structure, then write the corresponding evaluate extraction logic.

3. Screenshot evidence

python3 <skill-dir>/scripts/tk_research.py screenshot "search_001"

4. Scroll to load more

curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"evaluate","args":{"code":"window.scrollTo(0, document.body.scrollHeight); return document.body.scrollHeight;"},"session":"tiktok-search"}'

After scrolling, wait 2-3 seconds for content to load, then snapshot or evaluate to extract newly appearing cards.

5. Open a creator profile to collect works

# 导航到博主主页
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"navigate","args":{"url":"https://www.tiktok.com/@creator","newTab":true},"session":"tiktok-creator"}'

# 提取主页视频列表
curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"evaluate","args":{"code":"(() => { const items = [...document.querySelectorAll(\"[data-e2e=user-post-item]\")].slice(0, 10); return items.map((item, i) => { const link = item.querySelector(\"a\"); const views = item.querySelector(\"[data-e2e=video-views]\"); return { index: i, url: link ? link.href : null, title: item.innerText.slice(0, 100), views: views ? views.innerText : null, raw_text: item.innerText.slice(0, 300) }; }); })()"},"session":"tiktok-creator"}'

Output directory structure

bggg-tiktok-search/projects/tiktok-research/YYYYMMDD_HHMMSS_<slug>/
├── collected_items.json
├── research_notes.md
└── screenshots/
    ├── search_001.png
    ├── search_002.png
    └── ...

collected_items.json structure

{
  "method": "kimi-webbridge",
  "query": "skincare routine",
  "source_url": "https://www.tiktok.com/search?q=skincare%20routine",
  "session": "tiktok-search",
  "items": [
    {
      "url": "https://www.tiktok.com/@creator/video/1234567890123456789",
      "author": "@creator",
      "author_url": "https://www.tiktok.com/@creator",
      "title": "visible caption or card text",
      "metric_1": "visible metric",
      "metric_2": "visible metric",
      "date": "visible date",
      "raw_text": "short visible evidence",
      "screenshot": "screenshots/search_001.png",
      "notes": "why this item matters"
    }
  ]
}

Research output suggestions

When finally replying to the user, prioritize giving:

  1. Research package paths: JSON, CSV, Markdown, screenshots.
  2. List of highest-value candidate videos or creators.
  3. Content structure insights: topic selection, opening hooks, account positioning, sales intent, shot/script patterns.
  4. Next-step suggestions: whether to proceed to download, transcription, deep analysis, or Feishu Bitable archival.

Notes

  • TikTok DOM changes frequently，evaluate The selector in it needs to be adjusted according to the actual page structure. First use snapshot to confirm the current structure before writing the extraction logic.
  • When encountering login, risk control, regional restrictions, or CAPTCHA, have the user handle it manually in a real Chrome and then continue. Do not solve CAPTCHAs on their behalf.
  • Do not perform external side-effect actions such as liking, following, commenting, sending direct messages, posting, or changing account settings; if the user explicitly requests it, confirmation must be obtained before the action.
  • Do not submit cookies, browser profiles, screenshots, CSV, JSON scraping results, or any runtime artifacts to a public repository.
  • Write screenshot files to projects/tiktok-research/.../screenshots/; do not paste base64 screenshots directly into the context.
  • Close the session after the task ends： 
    curl -s -X POST "$WEBBRIDGE_URL/command" \
  -d '{"action":"close_session","session":"tiktok-search"}'
  • references/implementation-notes.md records design patterns learned from other browser-use skills; third-party reference projects are not vendored with this skill.

Read more implementation details as needed from references/implementation-notes.md。