websh_skill

Overview

This skill is a shell for the web that treats URLs like directories and the DOM like a filesystem. You navigate with familiar shell commands (cd, ls, cat, grep, follow) and interact with page content locally and instantly. Background tasks fetch and parse pages so the prompt never blocks. It’s designed for fast, programmatic browsing and content extraction.

How this skill works

When you cd to a URL the session updates immediately and a background agent fetches the page, caches the HTML, and extracts parsed content. Commands like ls, cat, and grep operate on the cached extraction; if content is still loading the shell returns graceful partial results or status messages. The system also prefetches linked pages in the background to make follow and navigation feel instant.

When to use it

• You want to navigate websites using shell-style commands and workflows.
• You need to query or extract text from web pages programmatically (grep, cat, selectors).
• You prefer non-blocking interactions where the prompt remains responsive while fetching occurs.
• You want to prefetch and explore link graphs for faster multi-page workflows.
• You need a reproducible, local cache of page content for scripting or automation.

Best practices

• Start sessions at a stable landing page (e.g., news.ycombinator.com) to leverage eager prefetching.
• Use selectors with cat or ls to target specific elements and reduce noisy output.
• Turn off eager prefetch on slow connections using prefetch off to save bandwidth.
• Rely on grep and pipe chains (ls | grep | head) for quick filtering without re-fetching.
• Treat cd as asynchronous: expect a fetching status and use commands that tolerate partial cache.

Example use cases

• Quickly list top links on a news homepage: cd https://news.ycombinator.com then ls | head 10.
• Extract article titles or headings with selectors: cat .title or cat h1.
• Search multiple pages for a keyword: cd a site root, then grep "AI" across cached pages.
• Follow links instantly during research because background prefetch already cached targets.
• Automate periodic snapshots by caching pages and comparing parsed Markdown outputs.