web-research

Neural web search and content extraction using x402-protected APIs. Better than WebSearch for deep research and WebFetch for blocked sites. USE FOR: - Deep web research and investigation - Finding similar pages to a reference URL - Extracting clean text from web pages - Scraping sites that block standard fetchers - Getting direct answers to factual questions - Research requiring multiple sources - Crawling multiple pages from a website TRIGGERS: - "research", "investigate", "deep dive", "find

15stars8forksUpdated 7/17/2026

#agentcash #x402 #neural-search #web-scraping #content-extraction #web-research

Security Assessment

Low Risk(70/100)

Detected risks:

Secret Exposure([SKILL.md] token=)

Security Score70/100

About web-research

web-research provides neural web search and content extraction through x402-protected APIs, giving agents a stronger alternative to standard WebSearch for deep investigation and to WebFetch for sites that block ordinary fetchers. It solves the problem of shallow keyword searches and inaccessible or JavaScript-heavy pages by routing requests to Exa (semantic/neural search) and Firecrawl (web scraping), plus a Cloudflare crawl endpoint for multi-page site crawls. It is intended for deep research, finding pages similar to a reference URL, extracting clean text, scraping blocked sites, direct factual Q&A, and crawling whole sites or sections.

Endpoints are reached with the command `npx agentcash@latest fetch` against stableenrich.dev URLs. Available endpoints include Exa search ($0.01), find-similar ($0.01), contents ($0.002), and answer ($0.01); Firecrawl scrape ($0.0126) and search ($0.0252); and a Cloudflare crawl ($0.10) with free polling of results. Exa search accepts options such as query, numResults (max 25), type ("neural" or "keyword"), includeDomains/excludeDomains, date filters, and a category filter (company, research paper, news, pdf, github, tweet, personal site, linkedin profile, financial report). Firecrawl scrape supports formats, onlyMainContent, and waitFor for JS rendering. The Cloudflare crawl uses an async two-step pattern: start the crawl (returns a JWT token) then poll the jobs endpoint every 3-5 seconds until complete. Balance can be checked with `npx agentcash@latest balance`.

The skill targets research agents and developers who need reliable extraction from difficult sources. It emphasizes cost optimization: try free WebSearch/WebFetch first, use Exa contents ($0.002) when URLs are already known, batch URL extraction, and limit result counts. Setup requires installation and wallet configuration described in rules/getting-started.md, and detailed tool-selection guidance lives in rules/when-to-use.md.

FAQ

How do I call the research endpoints?

Use the command `npx agentcash@latest fetch` against the relevant stableenrich.dev URL, passing `-m POST` and a JSON body with `-b` where required. For example, Exa search posts a body with a `query` field.

When should I use this skill instead of WebSearch or WebFetch?

Use it for deep research, semantic/conceptual search, finding pages similar to a URL, extracting clean text from blocked or JavaScript-heavy sites, quick factual answers, and full-site crawls. The docs recommend trying free WebSearch/WebFetch first and falling back to these x402 endpoints when needed.

How does the website crawl work?

The Cloudflare crawl uses an async two-step pattern: start the crawl (paid, $0.10) which returns a 202 with a JWT token, then poll the jobs endpoint with that token (free) every 3-5 seconds until the crawl is complete. Parameters include url, limit (max 25), depth (max 3), formats, render, and include/exclude patterns.

How do I keep costs down?

Use Exa contents ($0.002) when you already have URLs, batch multiple URLs into a single contents call, request only as many results as needed, and fall back to the paid endpoints only after free WebSearch/WebFetch. You can check your balance with `npx agentcash@latest balance`.

What setup is required?

Installation and wallet setup are covered in rules/getting-started.md. The endpoints are x402-protected, so a funded wallet is needed to pay per-request charges.