Scan text an agent is about to read (a web page, an email, a tool description, another agent's message) for the shapes that hijack agents: instruction-override phrasing, a redefined system role, requests to move credentials, hidden directive tags, invisible Unicode. Verdict plus every pattern found. Pattern-based; nothing is stored. Pay-per-call via x402. An empty body is answered for the published example text (one harmless sentence); send {"text":"…"} to scan your own.

Payment Options