search

name

search

description

Search the live web, scrape pages to clean markdown, search open-source code, and pull library docs — all through the local `ketch` CLI. Use this whenever the user wants to find something online, look something up, research a topic, gather sources, read/summarize a web page, scrape a site, crawl a site, find code examples in OSS repos, or get version-aware library documentation. Free and local: web search goes through the user's private SearXNG; scraping uses ketch's own readability engine (Brave only for JS-rendered pages). Replaces the older searxng/firecrawl/exa/web-search skills. For INTERACTIVE control of the user's real browser (clicking, logins, tabs), use the `browser` skill instead.

research: web search + scrape + code + docs via ketch

ketch is a fast, stateless Go CLI (no daemon, every call independent and cached). One binary, four surfaces: web search, page scraping, OSS code search, library docs. Always add --json when you need to parse the output.

Already configured (in ~/.config/ketch/config.json)

• backend = searxng → web search hits the user's private SearXNG at http://192.168.0.25:8888 (by IP — Go can't resolve search.local/mDNS).
• browser = /opt/brave-origin-bin/brave → JS pages render via the user's Brave binary (see "How ketch uses Brave" below). No Google Chromium dependency.
• Other search backends available: exa, brave, ddg (override per call with -b).
• brave_api_key is set, so brave works as a paid premium backend.

Resilient search — use ./search.sh (auto-fallback)

ketch has no built-in search fallback: if SearXNG is down, ketch search -b searxng errors (exit 4). For search that must not fail, call the wrapper instead of ketch search:

./search.sh "<query>" [--limit N --json]

It tries the chain searxng → brave (private primary → Brave paid fallback), falling through only on backend ERROR (a valid empty result is respected, not retried). The chosen backend is in the JSON (backend:) and a meta: fallback backend used = … line goes to stderr. Edit CHAIN=(…) in search.sh to change. ketch scrape/code/docs are unaffected — they have no SearXNG dependency.

Commands

# WEB SEARCH (via SearXNG). --scrape pulls full content from each hit.
ketch search "claude opus pricing" --limit 5
ketch search "claude opus pricing" --scrape --json
ketch search "neural retrieval" -b exa            # one-off: use Exa instead

# SCRAPE a URL -> clean markdown (HTTP fast-path; Brave only if JS-rendered)
ketch scrape https://example.com/article
ketch scrape https://a.com https://b.com --json   # concurrent batch
ketch scrape https://spa.example --force-browser   # force Brave render
ketch scrape https://example.com --select "main"   # CSS selector, skip readability
ketch scrape https://example.com --raw             # raw HTML instead of markdown

# CRAWL a site (BFS or sitemap; background for big ones)
ketch crawl https://docs.example.com --depth 2
ketch crawl https://example.com/sitemap.xml --sitemap --background
ketch crawl status                                 # list crawls

# OSS CODE SEARCH (grep.app default, token-free; or sourcegraph/github)
ketch code "http.NewRequestWithContext" --lang go
ketch code "rate limit middleware" --lang go -b github --json

# LIBRARY DOCS (Context7 — needs context7_api_key; optional)
ketch docs "word wrap rendering" --library /charmbracelet/glamour

Useful flags: --limit/-l N, --json, --minimal (one result/line), --trim (strip markdown to text), --max-chars N (truncate), --no-cache.

How ketch uses the user's Brave (important)

ketch does NOT attach to or touch the user's running/daily Brave. It only uses the Brave binary as a rendering engine for JavaScript-heavy pages:

1. Static pages → plain HTTP fetch, no browser at all (fast path).
2. JS-rendered page (auto-detected) or --force-browser → ketch (via go-rod) launches a separate, headless Brave process with a fresh temporary --user-data-dir: no logins, no cookies, no history — ephemeral / incognito-like. It's killed and the temp profile deleted when the scrape finishes.
3. This throwaway instance has a different --user-data-dir than the daily Brave-Browser-Pi, so the two never conflict and ketch never sees the user's sessions.

Net: ketch borrows the Brave program to render pages in a clean sandbox — no Google Chromium, no privacy leak, no interference with interactive browsing.

Choosing this skill vs browser

• research (ketch) → read-only: find, fetch, scrape, crawl, code/docs lookup.
• browser → interactive control of the user's REAL Brave (their logins, clicking, typing, tabs, authenticated dashboards). ketch cannot do that.

Config management

ketch config                                 # show effective config + backends (JSON)
ketch config set backend searxng
ketch config set searxng_url http://192.168.0.25:8888
ketch config set browser /opt/brave-origin-bin/brave