insane-search for Claude Code: a skill that reads blocked public pages another way

insane-search is a Claude Code skill that retrieves public web pages through alternative routes when the usual request gets blocked. It ships as a single skill inside a plugin and returns body text as Markdown along with metadata. This post covers the problem it solves, who it suits, how to install it and run it the first time, and the limit you should know before relying on it.

What problem it solves

Plenty of pages that anyone can open in a browser refuse an agent’s plain request. A community thread, a paper listing, or a video page comes back as a challenge screen, an empty shell, or an error, and the agent either gives up or summarizes the wrong thing. For research work that pulls from the public web, that turns a simple read into manual copying.

insane-search changes the order of attack. It looks for an official public route first, such as RSS for Reddit, the Firebase API for Hacker News, or the Atom API for arXiv. If that route is blocked, it escalates through public APIs, feeds, TLS impersonation, and finally a real browser using the Chrome already on your system. The README presents it as needing no API keys or proxy configuration.

Who it fits

The chapter rates four situations. Three cover most decisions.

Situation	Fit	Reason
Collecting community reactions	Good fit	Supports public access routes
Obtaining video subtitles as text	Good fit	Extracts subtitles
Collecting dashboards behind a login	Poor fit	Stops at authentication barriers

Checking prices on product pages sits in the middle, because it relies on whatever structured data the page exposes. The README does not provide Codex installation instructions, so treat it as a Claude Code skill.

Install and first run

Add the marketplace in Claude Code, install the plugin, and reload.

# Claude Code: add the marketplace, then install the plugin
/plugin marketplace add https://github.com/fivetaku/gptaku_plugins.git
/plugin install insane-search@gptaku-plugins
/reload-plugins

Required packages install automatically on first use. On that first run the skill also runs an installation script and asks once whether you want to star the GitHub repository. Starring is optional, and declining changes nothing about its features.

There is no separate command to remember. You ask for a public page the way you normally would, and the README says the skill activates on its own when retrieval is blocked.

Read [public URL] and summarize the key points

What makes the result trustworthy is the success check. An HTTP 200 alone does not count. The skill inspects the response for challenge text and checks its size before it accepts anything as body text, then converts HTML to Markdown or extracts text from PDFs.

What trips people up

The skill does not cross login barriers or paywalls, and that is by design. When a page needs authentication, it reports authentication required and stops rather than trying to get around it. If your target is an internal dashboard or a subscriber-only article, this is the wrong tool. Even on public pages, the chapter is clear that you remain responsible for each site’s terms of service, robots.txt, and rate limits. One more cost note: X search can use xAI credentials for additional results, which may incur charges, and there is a setting that keeps it on free routes only.

Where this comes from

This post adapts part of chapter 03-04 of The Claude Code & Codex Skills Guidebook. The full chapter covers the platform-specific routes it documents, including yt-dlp for YouTube subtitles and how it treats retrieved text as untrusted data, and the table of included commands and settings files such as python3 -m engine and waf_profiles.yaml, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Gumroad.