Deterministic prompt-injection pre-filter for untrusted text (web pages, emails, tool output) before it reaches an LLM: instruction overrides, role/template spoofing, secret and system-prompt exfiltration, markdown image beacons, tool-call markup, crypto transfer instructions, hidden Unicode (zero-width, bidi, tag smuggling), hidden HTML, base64 payloads. Returns score 0-100, risk level, matched rules and spans. Heuristic, no LLM; max 32 KB. PAID: $0.002 USDC per call on Base via x402 (MCP transport: retry with params._meta["x402/payment"]). Same tool over plain HTTP: POST /v1/text/injection-scan.