skillZs
★ LIVE SKILL TAGS ★
>>> LIVE SKILLS INDEX <<<
* OPEN SOURCE *
NO LOGIN, NO TRACKING
※ REAL INSTALL DATA ※
← back to all skills
us/crw63 installs

crw-search

Search the web with fastCRW and get titles, URLs, and descriptions. Use when you have a question or topic but not a URL — "search for", "find pages about", "look up", "what is", "who is", "latest news on", "find docs for". Own search backend: self-hosted, no API key, no per-query cost, high recall via meta-search aggregation. Step 1 of the crw workflow ladder.

How do I install this agent skill?

npx skills add https://github.com/us/crw --skill crw-search
view source ↗

Is this agent skill safe to install?

  • Gen Agent Trust Hubpass

    The skill is a web search utility that uses a self-hosted backend. While generally safe, it introduces a surface for indirect prompt injection because it ingests untrusted data from the internet (search snippets and titles) into the agent's context.

  • Socketpass

    No alerts

  • Snykwarn

    Risk: MEDIUM · 1 issue

What does this agent skill do?

crw-search — web search via crw's own search backend

When to use

  • You have a question or topic, not a URL. Get candidate URLs first, then scrape the one that looks right.
  • Step 1 in the crw ladder: most searches should end at step 2 (crw-scrape) — pick the best result URL and scrape it for full content.
  • Token-heavy context? Pipe through a subprocess filter instead of dumping raw JSON. See crw-dynamic-search.
  • The search backend is self-hosted and free — no API key, no per-query billing, no usage cap. Queries never leave your infrastructure in embedded/local mode.

Quick start

CLI (binary on PATH):

crw search "rust async http client"                              # text output
crw search "site:docs.rs tokio" --json --fields title,url,snippet --limit 5
crw search "CVE-2024-1234" --category news --time-range week
crw search "climate policy 2025" --json -o .crw/results.json
crw search "rust crates" --language en --limit 20

MCP (inside an agent harness):

crw_search(query="rust async http client", limit=5, lang="en")
crw_search(query="latest CVE nginx", tbs="qdr:w", categories="news")
crw_search(query="openai pricing", scrapeOptions={"formats": ["markdown"]})

REST (drop-in for Firecrawl SDKs — just swap the base URL):

curl -X POST "$CRW_API_URL/v1/search" -H "Authorization: Bearer $CRW_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"query":"rust async http","limit":5,"lang":"en"}'

Options

NeedCLI flagMCP / REST field
Result count-l/--limit N (default 10)limit (default 5)
JSON output--json or --format json— (always JSON)
Field projection--fields title,url,snippet—
Output to file-o FILE—
Filter by category--category news|images|videos|general|…categories
Language--language enlang
Time filter--time-range day|week|month|yeartbs: qdr:h|qdr:d|qdr:w|qdr:m|qdr:y
Safe search--safesearch 0|1|2—
Custom search backend--searxng-url URL / $CRW_SEARXNG_URL—
Group by source—sources: ["web","news","images"]
Scrape results inline— (use crw scrape separately)scrapeOptions: {formats:["markdown"]}

--fields available values: title, url, description, snippet, position, score, category. snippet is an alias for description.

A note on result scores

The search backend is a meta-search aggregator — it merges results from multiple engines (Google, Bing, DuckDuckGo, etc.) and the score field reflects internal engine weighting, not a universal relevance measure. Do not rely on score for ranking or filtering. Use position (1-based rank) or result order instead — position 1 is the most relevant result surfaced.

Tips

  • No results / 403? The search backend needs JSON output enabled in its config. Run crw setup --local to spin up a pre-configured sidecar automatically. Public instances usually block JSON with 403/429.
  • --fields saves context. --json --fields title,url,snippet --limit 5 is one call; piping to jq is two. Prefer the flag.
  • Inline scraping via MCP. Pass scrapeOptions: {formats: ["markdown"]} to get page content alongside search results in one round-trip. There is no --scrape CLI flag — use the MCP/REST path for this.
  • Time-sensitive queries. Use --time-range week (CLI) or tbs: "qdr:w" (MCP/REST) for news, CVEs, releases, or any freshness-sensitive topic.
  • After search, scrape the winner. crw search "…" returns candidates; crw scrape "<url>" gets the full content. Don't try to read content from search snippets alone.
  • Write large result sets to .crw/. Never stream a 20-result JSON blob to stdout into context. Use -o .crw/results.json then jq/grep.

See also

  • crw-scrape — scrape the URL you found
  • crw-dynamic-search — filter output in a subprocess to save context (use this on token-heavy tasks)
  • crw — hub skill with the full workflow ladder

Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.

<a href="https://skillzs.dev/skills/us/crw/crw-search">View crw-search on skillZs</a>