Supported platforms & reference
August 6, 2026 · View on GitHub
Full platform coverage, the reference files behind each route, dependencies, and example prompts. Most sites need no explicit entry — the Phase 0→3 adaptive scheduler handles them automatically. This page lists the special endpoints and the reference map.
What's in the index
Only special endpoints that the generic chain can't discover on its own. Everything else — blogs, commerce sites, professional networks, newsletters, news sites, and most forums — is handled by the adaptive scheduler without explicit entries.
Platform-specific APIs
| Platform | Method | Reference |
|---|---|---|
| X/Twitter | single tweet → cdn.syndication tweet-result + oEmbed · timeline → syndication · keyword → WebSearch → tweet-result | twitter.md |
.rss feed via curl_cffi (the unauth .json now 403s) | json-api.md | |
| Threads | video post → inline video_versions JSON, block nearest to the URL shortcode (engine Phase 0 — yt-dlp has no extractor; signed CDN URLs expire, download immediately) | media.md |
| Bluesky | AT Protocol (public.api.bsky.app/xrpc/...) | public-api.md |
| Mastodon | Per-instance public API | public-api.md |
| Hacker News | Firebase API + Algolia Search (hn.algolia.com/api/v1/search) | json-api.md |
| Stack Overflow | SE API v2.3 | public-api.md |
| Lobste.rs / V2EX / dev.to | Public JSON APIs | json-api.md |
Media (CLI tool required)
| Platform | Method | Reference |
|---|---|---|
| YouTube / Vimeo / Twitch / TikTok / SoundCloud + 1,853 others | yt-dlp --dump-json | media.md |
Academic & registry
| Platform | Method | Reference |
|---|---|---|
| arXiv | Atom API | public-api.md |
| CrossRef | REST API | public-api.md |
| Wikipedia | REST API | json-api.md |
| OpenLibrary | JSON API | public-api.md |
| GitHub | gh CLI / REST API | public-api.md |
| npm / PyPI | Registry API | json-api.md |
| Wayback Machine | CDX API | public-api.md |
Korea-specific
| Platform | Method | Reference |
|---|---|---|
| Naver Search | curl_cffi + search.naver.com (통합/블로그/뉴스) | naver.md |
| Naver Finance (stock prices) | api.finance.naver.com/siseJson.naver (unofficial, no auth) | naver.md |
Everything else flows through Phase 1~3 automatically — commerce sites, professional networks, blogs, newsletters, and most forums via the generic curl_cffi + JSON-LD/metadata chain or Jina, plus any site with /rss or /feed endpoints.
Reference files
The skill is organized as a set of reference files (skills/insane-search/references/), each covering one class of techniques.
| File | Covers |
|---|---|
fallback.md | Phase 0→3 adaptive scheduler, escalation signals, response validation |
jina.md | Jina Reader (no-key reader at r.jina.ai) |
json-api.md | Public JSON APIs (Reddit, HN, dev.to, Wikipedia, npm, PyPI, etc.) |
public-api.md | Bluesky, Mastodon, Stack Exchange, arXiv, CrossRef, OpenLibrary, GitHub, Wayback |
media.md | yt-dlp usage for 1,858 media sites |
twitter.md | X/Twitter tweet-result + oEmbed + syndication + WebSearch keyword search |
naver.md | Naver Search (curl_cffi), blog mobile URLs, Finance JSON API |
rss.md | Korean news RSS (9 outlets), Google News RSS, feedparser, SearXNG |
tls-impersonate.md | curl_cffi multi-target + identity spoofing (cookie warming, referrer chain) + behavioral challenge detection |
playwright.md | Playwright MCP full toolkit (snapshot, evaluate, network_requests) |
cache-archive.md | Google AMP cache, archive.today, Wayback Machine |
metadata.md | OGP, JSON-LD, Schema.org, Next.js RSC payload extraction |
Dependencies
Required: Claude Code only.
Auto-installed when needed (the skill installs these transparently on first use):
pip install curl_cffi # TLS impersonation for WAF-blocked sites (>= 0.15.0)
pip install feedparser # RSS/Atom parsing
pip install yt-dlp # 1,858 media sites
Optional, improves coverage:
brew install gh # GitHub (faster than REST API)
claude mcp add playwright -- npx @playwright/mcp@latest # JS-rendered sites
If a dependency is missing, the skill doesn't skip the method — it installs the dependency and tries.
What insane-search is not
- Not a scraper — It's a method-selection layer. It uses public APIs and standard techniques.
- Not API-key based — Everything uses no-auth public endpoints or URL transformations.
- Not a hand-maintained answer key — The index is minimal (~15 groups). Everything else is discovered by the adaptive scheduler.
- Not bias-forming — There's no "access denied" list. If a site can be reached, the chain will find the way.
Example prompts
There are no commands. Just talk normally — the skill triggers when a URL is blocked or when accessing platforms that need special handling.
"What's on the front page of Hacker News right now?"
→ Firebase API → top stories with scores and comments
"Find AI papers published this week on arXiv"
→ arXiv Atom API with date filter
"Read a commerce listing page for product data"
→ Phase 2: curl_cffi → JSON-LD ItemList
"Summarize this Medium article"
→ Phase 1: Jina Reader → clean markdown
"Check what people are saying about Claude Code on Reddit"
→ Reddit .rss feed (curl_cffi) → posts
"Search X for insane-search"
→ Intent routing: keyword → WebSearch(site:x.com) → tweet-result → full tweets
"네이버에서 클로드코드 뉴스 찾아줘"
→ Naver Search (identity spoofing) → news tab → article URLs → Jina Reader
"Find articles on a professional-network site"
→ WebSearch → curl_cffi → JSON-LD articleBody