😰 Problem

Agents Open Links Blind

  • LLM agents fetch URLs without validation
  • SSRF exposure through untrusted redirects
  • Phishing pages injected into context windows
  • Prompt injection hidden in HTML, titles, meta tags
  • One malicious link → full agent compromise
😌 Solution

Gate Every Link Before It Opens

CLI Tool

  • sancheck scan for one-off checks
  • sancheck gate for CI/CD pipelines
  • --stdin for inline message filtering

Codex Plugin

  • Intercepts links before agent opens them
  • Auto-blocks flagged URLs
  • Zero config, drops into existing workflow

Exit Codes

  • 0 — allow, safe to proceed
  • 2 — block, threat detected
  • 1 — error, investigate
😮 Demo

Three Ways to Run It

Scan a URL directly

python -m sancheck scan https://example.com --format text

Pipe through the gate script

printf 'check https://example.com' | ./scripts/sancheck-gate

Gate a message file via stdin

python -m sancheck gate --stdin --format json < message.md

Output includes findings + allow/block decision. Checks: URL tricks, DNS/SSRF, TLS, redirects, prompt injection.

🔥 Ask

Stdlib Python. Zero Dependencies.

Try the Gate

  • Drop into any pipeline
  • Gate agent messages before execution

Install the Codex Plugin

  • Auto-intercept links in agent sessions
  • Block threats before they reach context

github.com/fozagtx/sancheck

Made with