Netra PII And Input Guardrails

Use this skill to protect prompts and logs from sensitive data and malicious inputs.

When To Use

You must detect/mask/block PII before sending text to LLMs.
You need prompt-injection scanning before agent/tool execution.
You want explicit security actions (MASK, FLAG, BLOCK).

Procedure

Choose a PII detector (get_default_detector or RegexPIIDetector).
Set action type based on policy (MASK for safe continuation is common).
Scan user input with InputScanner before processing.
Stop requests on blocked violations.
Send masked content downstream when allowed.

Python Pattern

from netra.pii import get_default_detector
from netra.input_scanner import InputScanner, ScannerType

pii_detector = get_default_detector(
    action_type="MASK",
    entities=["EMAIL_ADDRESS", "PHONE_NUMBER", "CREDIT_CARD"],
)
scanner = InputScanner(scanner_types=[ScannerType.PROMPT_INJECTION])


def secure_input_pipeline(user_text: str) -> str:
    scan_result = scanner.scan(user_text, is_blocked=True)
    if scan_result.has_violation:
        raise ValueError("Malicious input detected")

    pii_result = pii_detector.detect(user_text)
    return pii_result.masked_text if pii_result.has_pii else user_text

TypeScript Pattern

import { Netra } from "netra-sdk";

const EMAIL_RE = /[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}/g;
const INJECTION_RE = /ignore previous instructions|reveal system prompt/i;

export function secureInputPipeline(userText: string): string {
    const span = Netra.startSpan("security-input-scan");
    try {
        if (INJECTION_RE.test(userText)) {
            span.setError("prompt_injection_detected");
            throw new Error("Malicious input detected");
        }

        const masked = userText.replace(EMAIL_RE, "[EMAIL_ADDRESS]");
        span.setAttribute("security.masked", masked !== userText);
        span.setSuccess();
        return masked;
    } finally {
        span.end();
    }
}

Checklist

Scanner runs before model/tool calls.
PII action type matches policy requirements.
Blocked inputs fail fast with safe error responses.
Security metadata is logged without storing secret values.

netra-pii-and-input-guardrails

Netra PII And Input Guardrails

When To Use

Procedure

Python Pattern

TypeScript Pattern

Checklist

References