Score a URL for phishing risk before an agent opens or trusts it. Keyless, deterministic heuristics - typosquat / homoglyph of a known brand, punycode / mixed-script hosts, credentials-in-URL, raw-IP hosts, suspicious TLDs, over-deep subdomains - plus a best-effort domain-age check (young domains are a top phishing signal). Returns a 0-100 risk score, a SAFE / SUSPICIOUS / DANGEROUS verdict, and labelled reasons. No API keys, no LLM.