docs(tiktok): document Creator Search Insights (CSI) scraping

[eyalgaon]

Summary

Adds a new domain skill at domain-skills/tiktok/csi-scraping.md for TikTok Creator Search Insights (tiktok.com/csi), an auth-walled dashboard exposing trending search queries by category. Captures everything the next agent needs to skip the discovery work I had to do:

• URL + login wall detection (/csi redirects to /login for anonymous users)
• The private API endpoint /api/search/inspired_query/recommended_queries, request body shape, and category filter values (Food, Fashion, Gaming, Tourism, Science, Sports)
• Response fields worth keeping (query_text, popularity_v2, trending_seq_v2 time series, query_labels incl. content_gap, textnet hierarchy)
• Pagination strategy (has_more + cursor, 50/page accepted)
• A gotcha that cost me a few iterations: js() wraps in a non-async IIFE, so top-level await inside the captured fetch loop won't compile — use .then() chains or paginate from the Python side, one page per js() call

No coordinates, no cookies, no user-specific state — only the durable shape of the site, per the SKILL.md guidance on what a domain skill should capture.

Test plan

• Used the documented flow end-to-end against a real logged-in session — pulled 174 food trends, 28 flagged content_gap
• Verified the captured endpoint shape is reproducible across runs
• Confirmed the js()-non-async IIFE behavior matches the documented workaround

🤖 Generated with Claude Code

---

Summary by cubic

Adds a new domain skill documenting how to scrape TikTok Creator Search Insights (tiktok.com/csi) behind login. Covers private API capture, pagination, and regional targeting via tt-target-idc to get country-appropriate trends.

• New Features
  • Added domain-skills/tiktok/csi-scraping.md with login-wall detection and session reuse.
  • Captures /api/search/inspired_query/recommended_queries via a window.fetch interceptor; documents body (category/language/tab), limit 50, offset/cursor, and session_refresh_index.
  • Lists key response fields to keep: query_text, popularity_v2, trending_seq_v2, query_labels, textnet, video_num.
  • Regional targeting: swap tt-target-idc/store-idc and drop tt-target-idc-sign to reroute; recommend language_filters: ["en"] for best coverage.
  • Gotchas: js() wraps a non-async IIFE (use .then()), and Filters dropdown toggle behavior.

^{Written for commit 5b8c980. Summary will update on new commits. Review in cubic}

[browser-harness-review]

✅ Skill review passed

Reviewed 1 file(s) — no findings.

[cubic-dev-ai]

No issues found across 1 file

_{Re-trigger cubic}