PII Guards

RegexPiiDetector

The detection engine used by all PII guards. Patterns are compiled at construction time with RegexOptions.Compiled | RegexOptions.IgnoreCase for maximum performance.

The 10 Detection Patterns

Two detection methods

public interface IPiiDetector
{
    IReadOnlyList<PatternDefinition> DetectionRules { get; }

    // Returns {piiType → count} — fast check for Pre-phase blocking
    Dictionary<string, int> DetectPii(string text);

    // Returns full match details — used by Post-phase redaction
    List<PiiMatch> DetectPiiDetailed(string text);
}

public class PiiMatch
{
    public string PiiType { get; set; }       // e.g. "SSN", "Email"
    public string MatchedText { get; set; }   // the actual matched string
    public int StartPosition { get; set; }
    public int EndPosition { get; set; }
    public string Pattern { get; set; }       // the regex pattern that matched
    public double ConfidenceLevel { get; set; } // always 0.95
}

PiiDetectionGuard v2

How it works

Converts context.Input to string via .ToString(), calls IPiiDetector.DetectPii(inputStr). If any PII is found, returns Blocked with detected types and counts in metadata.

Configuration

{
  "name": "PiiDetectionGuard",
  "enabled": true,
  "config": {
    "sensitivityLevel": "medium",  // required; "low" | "medium" | "high"
    "detection_rules": []          // optional; custom rules (currently all 10 patterns always run)
  }
}

Block result

{
  "IsAllowed": false,
  "ErrorMessage": "PII detected in input: SSN, Email",
  "Metadata": {
    "detected_pii_types": ["SSN", "Email"],
    "pii_counts": { "SSN": 1, "Email": 2 },
    "sensitivity_level": "medium"
  }
}

PiiRedactionGuard

How it works

Converts context.Output to string, calls IPiiDetector.DetectPiiDetailed() to get exact positions of all PII matches. Applies redaction in reverse positional order (to preserve string offsets). Directly sets context.Output to the redacted string and returns Success(outputModified: true).

Three redaction methods

Configuration

{
  "name": "PiiRedactionGuard",
  "enabled": true,
  "config": {
    "redaction_method": "mask",  // required; "mask" | "hash" | "partial"
    "fields": []                 // optional; target specific field names (empty = all output)
  }
}

Success result

{
  "IsAllowed": true,
  "OutputModified": true,
  "Metadata": {
    "redacted_pii_count": 3,
    "redaction_method": "mask"
  }
}

PiiAnonymizationGuard

Less destructive than redaction — preserves structure but replaces PII values with anonymized equivalents. Use when downstream systems need data in a recognizable format but without real personal data.

Configuration

{
  "name": "PiiAnonymizationGuard",
  "enabled": true,
  "config": {
    "anonymization_method": "pseudonymize",  // required
    "preserve_format": true                  // optional; keep same data type/format
  }
}

PiiNodeGuard (Composite)

A composite INodeGuardExecutor that wires PiiDetectionGuard (Pre) and PiiRedactionGuard (Post) into a single integration point. Use this when you want full two-layer PII protection without configuring each guard separately.

// PiiNodeGuard implements INodeGuardExecutor
ExecutePreAsync()  → runs PiiDetectionGuard (blocks if PII found)
ExecutePostAsync() → runs PiiRedactionGuard (redacts PII from output)
ExecuteOnErrorAsync() → no-op (returns Success)

Registering PiiNodeGuard

// In DI setup, for nodes that need full PII protection:
services.AddSingleton<INodeGuardExecutor, PiiNodeGuard>();

// Or use the standard full-stack registration which includes PiiNodeGuard:
services.AddGuardRailsExecution();

Full PII Workflow